chore: release v0.25.8

HTTPS to a relay board could not work. Nearly every board that offers it signs
its own certificate — there is no authority anywhere that could have signed it —
so the request failed verification before it left.

A checkbox, per board, off by default. Not a blanket switch, because the other
HTTPS case is real and opposite: a board reached from outside through a proxy
with a genuine certificate, where verification is the only thing standing
between an antenna switch and the internet. Same setting, two boards, different
answers.

Off by default is only safe if the failure explains itself, so a certificate
error now names the box to tick. Go's own "x509: certificate signed by unknown
authority" is accurate and tells an operator nothing about what to do next.

Shown only once an https:// URL is actually in the board's configuration. A
board on plain HTTP has no certificate to argue about, and an option that cannot
matter yet is one more thing to wonder about.

The flag joins the driver cache key: ticking it has to rebuild the driver, or
the cached one would go on refusing the certificate with the verifying client it
already holds.

The boards that take a bare host — WebSwitch, KMTronic — keep verification. An
https:// typed there is the proxy case by construction, since they default to
plain HTTP on the LAN.

Three tests against a real self-signed TLS server: accepted with the box,
refused with a message naming it without the box, and one board's setting not
leaking into another's.

amps_linked_test.go

@@ -1,6 +1,11 @@
 package main
 
-import "testing"
+import (
+	"errors"
+	"sync"
+	"testing"
+	"time"
+)
 
 // The coupling is a SET, not a global switch.
 //
@@ -61,3 +66,73 @@ func TestLinkedAmpsNeedsTwo(t *testing.T) {
 		}
 	}
 }
+
+// Two combined amplifiers must be commanded AT THE SAME TIME, not one after the
+// other.
+//
+// Sequentially, the second was commanded only once the first had answered — and
+// an SPE answers over its own link, in its own time. The combiner heard power
+// appear on one input before the other and beeped about it, on every OFF and
+// every ON. This is what an operator hears, so it is worth a test that would
+// hear it too.
+func TestLinkedAmpCommandsLeaveTogether(t *testing.T) {
+	a := &App{}
+	const slow = 150 * time.Millisecond
+
+	var mu sync.Mutex
+	starts := map[string]time.Time{}
+	err := a.ampFanOut([]string{"one", "two"}, func(id string) error {
+		mu.Lock()
+		starts[id] = time.Now()
+		mu.Unlock()
+		time.Sleep(slow) // an amplifier taking its time to answer
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("fan-out: %v", err)
+	}
+	if len(starts) != 2 {
+		t.Fatalf("%d amplifiers were commanded, want both", len(starts))
+	}
+	// Both goroutines wait on one channel and are released by closing it, so the
+	// gap is scheduling noise. Sequential execution would put a full command
+	// between them.
+	gap := starts["one"].Sub(starts["two"])
+	if gap < 0 {
+		gap = -gap
+	}
+	if gap > slow/3 {
+		t.Errorf("the two amplifiers were commanded %v apart — the combiner hears that as one input arriving late", gap)
+	}
+}
+
+// The amplifier the operator clicked comes first, and its failure is the one
+// reported: "the amp I pressed did not respond" beats the same message about
+// its silent partner.
+func TestLinkedAmpErrorNamesTheOneClicked(t *testing.T) {
+	a := &App{}
+	clicked := errors.New("the one clicked")
+	other := errors.New("the other one")
+	err := a.ampFanOut([]string{"clicked", "other"}, func(id string) error {
+		if id == "clicked" {
+			return clicked
+		}
+		return other
+	})
+	if !errors.Is(err, clicked) {
+		t.Errorf("fan-out reported %v, want the amplifier the operator pressed", err)
+	}
+}
+
+// One amplifier is the ordinary case and must not change: run inline, no
+// goroutine, no barrier, and the error straight back.
+func TestSingleAmpRunsInline(t *testing.T) {
+	a := &App{}
+	boom := errors.New("not running")
+	if err := a.ampFanOut([]string{"solo"}, func(string) error { return boom }); !errors.Is(err, boom) {
+		t.Errorf("a single amplifier reported %v, want the error itself", err)
+	}
+	if err := a.ampFanOut(nil, func(string) error { return boom }); err != nil {
+		t.Errorf("an empty group reported %v, want nothing to do", err)
+	}
+}

app.go

@@ -67,6 +67,8 @@ import (
 	"hamlog/internal/solar"
 	"hamlog/internal/spe"
 	"hamlog/internal/steppir"
+	"hamlog/internal/syncfolder"
+	"hamlog/internal/tciserver"
 	"hamlog/internal/tunergenius"
 	"hamlog/internal/uls"
 	"hamlog/internal/ultrabeam"
@@ -119,6 +121,8 @@ const (
 	keyCATDigitalDefault  = "cat.digital_default" // mode to use when CAT reports DATA
 	keyCATShareEnabled    = "cat.share.enabled"   // expose CAT to other programs (Hamlib NET rigctl)
 	keyCATSharePort       = "cat.share.port"      // TCP port for that server (rigctld default 4532)
+	keyCATShareProto      = "cat.share.proto"     // which sharing server runs: "rigctl" (Hamlib NET) or "tci"
+	keyCATShareTCIPort    = "cat.share.tci_port"  // WebSocket port for the TCI server (TCI default 40001)
 	keyCATXieguPort       = "cat.xiegu.port"      // Xiegu CI-V serial port (G90/X6100…)
 	keyCATXieguBaud       = "cat.xiegu.baud"      // Xiegu CI-V baud (G90 default 19200)
 	keyCATXieguAddr       = "cat.xiegu.addr"      // Xiegu CI-V address (factory 0x70)
@@ -456,8 +460,13 @@ type CATSettings struct {
 	PollMs          int    `json:"poll_ms"`         // poll interval in ms (default 250)
 	DelayMs         int    `json:"delay_ms"`        // pause between commands (default 0)
 	DigitalDefault  string `json:"digital_default"` // when CAT says DATA, surface this mode (FT8/FT4/RTTY/…)
-	ShareEnabled    bool   `json:"share_enabled"`   // serve CAT to other programs (Hamlib NET rigctl)
-	SharePort       int    `json:"share_port"`      // TCP port for it (default 4532)
+	ShareEnabled    bool   `json:"share_enabled"`   // serve CAT to other programs
+	SharePort       int    `json:"share_port"`      // TCP port for the rigctl server (default 4532)
+	// ShareProto picks WHICH server runs — "rigctl" or "tci". One or the other,
+	// not both: they are two ways of asking the same radio the same questions,
+	// and a second listener is only a second thing to go wrong.
+	ShareProto   string `json:"share_proto"`
+	ShareTCIPort int    `json:"share_tci_port"` // WebSocket port for the TCI server (default 40001)
 	// PTT hotkey — a keyboard key that keys the transmitter while OpsLog is
 	// focused (hold-to-talk, or toggle). Uses the configured Audio → PTT method,
 	// falling back to CAT keying when that is VOX/none.
@@ -595,8 +604,12 @@ type App struct {
 	// port: without it, choosing native CAT locks WSJT-X and friends out of the
 	// radio entirely. nil when the operator has not enabled sharing.
 	catShare *rigctld.Server
-	dxcc     *dxcc.Manager
-	cluster  *cluster.Manager
+	// catShareTCI serves the same link to programs built around Expert
+	// Electronics' TCI instead. One or the other runs, never both — they answer
+	// the same questions about the same radio, and nothing speaks both.
+	catShareTCI *tciserver.Server
+	dxcc        *dxcc.Manager
+	cluster     *cluster.Manager
 	// Cluster spots/lines are processed OFF the socket-read goroutine. Enriching a
 	// spot (DXCC/POTA), emitting it to the UI, running alert rules — which can hit
 	// a remote MySQL via isWorkedBandMode — and mirroring it to the Flex all used
@@ -735,6 +748,11 @@ type App struct {
 	confDLCancel   context.CancelFunc
 	udpLogMu       sync.Mutex             // serialises UDP auto-log so concurrent packets can't both pass the dedup check
 	adifMonMu      sync.Mutex             // guards the ADIF-monitor config (file list + per-file read offsets)
+	syncMu         sync.Mutex             // serialises folder synchronisation: config, the seq counter, and the append to our own file
+	syncSent       int64                  // changes written to the folder this session
+	syncReceived   int64                  // changes taken from the other machines this session
+	syncLast       time.Time              // last completed pass, for the status panel
+	syncErr        string                 // last folder error, shown in settings — a share that dropped is otherwise invisible
 	relayAutoMu    sync.Mutex             // serialises relay auto-control evaluation
 	relayAutoLast  map[string]bool        // deviceID|relay → last applied on/off, so we only switch on a real change
 	relayAutoOn    atomic.Bool            // cached "auto-control enabled" so the CAT hot path skips work when off
@@ -766,6 +784,8 @@ type App struct {
 	liveTableMu        sync.Mutex    // guards liveTableFor
 	liveTableFor       *sql.DB       // logbook whose live_status DDL has been ensured (once per connection, not per call)
 	awardSnapMu        sync.Mutex    // guards the award QSO snapshot
+	awardSnapBuild     sync.Mutex    // serialises BUILDING it — see awardSnapshot
+	awardSnapCap       int           // rows the last build produced, the capacity hint for the next
 	awardSnap          []qso.QSO     // light-scanned + enriched logbook snapshot reused across award computations
 	awardSnapRev       string        // logbook revision the snapshot was built at ("" = none)
 	awardSnapUsed      time.Time     // last read — the snapshot is dropped once it goes cold (see awardSnapshotJanitor)
@@ -1162,6 +1182,7 @@ func (a *App) startup(ctx context.Context) {
 	a.backfillAwardRefsOnce()                     // one-time: materialise award_refs for pre-existing QSOs
 	go a.rebuildWorkedIndex()                     // in-memory worked-index for per-spot alert checks
 	go a.adifMonitorLoop()                        // watch external ADIF files (fldigi, N1MM…) for new QSOs
+	go a.folderSyncLoop()                         // one operator, several PCs: read the other machines' change logs
 	a.relayAutoOn.Store(a.GetRelayAuto().Enabled) // prime the relay auto-control hot-path flag
 
 	// cty.dat for offline DXCC / country resolution. Cached on disk; first
@@ -1682,6 +1703,10 @@ func (a *App) shutdown(ctx context.Context) {
 		a.catShare.Stop()
 		a.catShare = nil
 	}
+	if a.catShareTCI != nil {
+		a.catShareTCI.Stop()
+		a.catShareTCI = nil
+	}
 	if a.cat != nil {
 		a.cat.Stop()
 	}
@@ -2645,8 +2670,14 @@ func (a *App) reloadLookupProviders() {
 		fmt.Println("OpsLog: settings load error:", err)
 		return
 	}
-	if days, _ := strconv.Atoi(m[keyCacheTTL]); days > 0 {
-		a.cache.SetTTL(time.Duration(days) * 24 * time.Hour)
+	// An EXPLICIT zero switches the cache off; an ABSENT key leaves the thirty-day
+	// default the cache was built with. The difference matters: every operator who
+	// has never opened this setting has no value stored, and reading that blank as
+	// a zero would silently turn the cache off for all of them.
+	if raw := strings.TrimSpace(m[keyCacheTTL]); raw != "" {
+		if days, err := strconv.Atoi(raw); err == nil && days >= 0 {
+			a.cache.SetTTL(time.Duration(days) * 24 * time.Hour)
+		}
 	}
 
 	build := func(name string) lookup.Provider {
@@ -2785,6 +2816,14 @@ func (a *App) AddQSO(q qso.QSO) (id int64, err error) {
 			a.maybeAutoSendEQSL(qc)
 			a.maybeSelfSpot(qc)
 			a.publishSoon() // refresh the published web page, debounced
+			// Tell the operator's other PCs. Down here with the rest of the
+			// after-the-fact work because a folder on a network share can block
+			// for seconds, and a contact belongs in the database and on screen
+			// long before another machine needs to hear about it.
+			//
+			// Read back rather than sent from `qc`: award_refs was materialised
+			// a few lines above and is not on the copy taken at insert time.
+			a.syncPublishAsync(syncfolder.OpAdd, id, nil)
 			if a.udp != nil {
 				rec := adif.SingleRecordADIF(qc)
 				a.udp.EmitLoggedADIF(rec)
@@ -4348,8 +4387,41 @@ func (a *App) awardSnapshot() ([]qso.QSO, error) {
 		a.awardSnapMu.Unlock()
 	}
 
+	// ONE BUILDER AT A TIME.
+	//
+	// The cache lock above is released before the pull, so every caller that
+	// arrives while the logbook is being read used to miss and start its own.
+	// Opening the Awards panel does that: a field log showed three pulls of the
+	// same 123 615 QSOs within ten seconds, and three copies alive at once took
+	// the heap from 725 MB to 2.5 GB. The work was identical each time.
+	//
+	// Waiting here costs the second and third caller the seconds the first was
+	// going to take anyway — they were already paying that, plus a second and
+	// third trip to the database.
+	a.awardSnapBuild.Lock()
+	defer a.awardSnapBuild.Unlock()
+	// Re-check: whoever held the build lock has just finished, and their result
+	// is what we came for.
+	if revErr == nil {
+		a.awardSnapMu.Lock()
+		if a.awardSnap != nil && a.awardSnapRev == rev {
+			qs := a.awardSnap
+			a.awardSnapUsed = time.Now()
+			a.awardSnapMu.Unlock()
+			return qs, nil
+		}
+		a.awardSnapMu.Unlock()
+	}
+
 	t0 := time.Now()
-	var all []qso.QSO
+	// Sized from the last build. Growing a slice to 123 000 structs by doubling
+	// copies the whole thing a dozen times and holds the old and the new array
+	// together at every step — on the biggest object OpsLog keeps, that transient
+	// is worth avoiding.
+	a.awardSnapMu.Lock()
+	hint := a.awardSnapCap
+	a.awardSnapMu.Unlock()
+	all := make([]qso.QSO, 0, hint)
 	if err := a.qso.IterateForAwards(a.ctx, func(q qso.QSO) error {
 		a.enrichQSOForAwards(&q)
 		all = append(all, q)
@@ -4365,13 +4437,14 @@ func (a *App) awardSnapshot() ([]qso.QSO, error) {
 	applog.Printf("awardSnapshot: pulled %d qsos from logbook in %v (rev=%s) — go heap now %d MB",
 		len(all), time.Since(t0).Round(time.Millisecond), rev, ms.HeapAlloc/(1024*1024))
 
+	a.awardSnapMu.Lock()
+	a.awardSnapCap = len(all) // next build starts the right size
 	if revErr == nil {
-		a.awardSnapMu.Lock()
 		a.awardSnap = all
 		a.awardSnapRev = rev
 		a.awardSnapUsed = time.Now()
-		a.awardSnapMu.Unlock()
 	}
+	a.awardSnapMu.Unlock()
 	return all, nil
 }
 
@@ -5167,6 +5240,37 @@ func (a *App) SaveAwardReference(code string, ref awardref.Ref) error {
 	}
 	a.markAwardEdited(code)
 	a.mirrorAwards()
+	// A reference's name is what the award column SHOWS for awards displaying by
+	// name, and its pattern is part of what matches at all — so editing one
+	// changes rows in the log, exactly as deleting or replacing the list does.
+	// Those already recomputed; this did not, and left the grid showing the old
+	// label until something else happened to trigger a pass.
+	a.recomputeAwardRefsAsync()
+	return nil
+}
+
+// RenameAwardReference changes a reference's code on an award the operator
+// already runs.
+//
+// The one field the editor could not touch, and the one that was wrong: WAJA
+// shipped numbered by the Japanese state rather than by the JARL. Correcting it
+// meant deleting the whole list and importing another — throwing away anything
+// the operator had adjusted in it.
+func (a *App) RenameAwardReference(code, oldRef, newRef string) error {
+	if a.awardRefs == nil {
+		return fmt.Errorf("db not initialized")
+	}
+	if err := a.awardRefs.Rename(a.ctx, code, oldRef, newRef); err != nil {
+		return err
+	}
+	a.markAwardEdited(code)
+	a.mirrorAwards()
+	// The materialised award columns hold a LABEL computed from the definition —
+	// the reference code for most awards, the name for those displaying by name.
+	// A renumbered reference changes the first kind, so the log is recomputed
+	// exactly as it is for every other reference-list change.
+	a.recomputeAwardRefsAsync()
+	applog.Printf("awards: %s reference %s renumbered to %s", code, oldRef, newRef)
 	return nil
 }
 
@@ -5978,6 +6082,7 @@ func (a *App) UpdateQSO(q qso.QSO) error {
 	if err == nil {
 		a.invalidateAwardStats()
 		a.materializeAwardRefs(q) // fields may have changed → refresh award_refs
+		a.syncPublishAsync(syncfolder.OpUpdate, q.ID, nil)
 	}
 	return err
 }
@@ -6071,6 +6176,7 @@ func (a *App) DeleteQSO(id int64) error {
 		return fmt.Errorf("db not initialized")
 	}
 	a.deleteRemoteCopies([]int64{id})
+	a.syncPublishDeletes([]int64{id})
 	return a.qso.Delete(a.ctx, id)
 }
 
@@ -6141,6 +6247,7 @@ func (a *App) DeleteQSOs(ids []int64) (int64, error) {
 		return 0, fmt.Errorf("db not initialized")
 	}
 	a.deleteRemoteCopies(ids)
+	a.syncPublishDeletes(ids)
 	return a.qso.DeleteMany(a.ctx, ids)
 }
 
@@ -7091,8 +7198,14 @@ func (a *App) SaveCabrilloFile() (string, error) {
 // LookupCallsign returns the cached or freshly-fetched info for a callsign.
 // Errors are returned as-is to the frontend; ErrNotFound surfaces as
 // "callsign not found".
-func (a *App) LookupCallsign(callsign string) (lookup.Result, error) {
-	return a.lookupCallsign(callsign, false)
+// qsoDate is the entry form's date as "YYYY-MM-DD" (or empty for now). It is
+// what the ClubLog exception is resolved AGAINST: an exception has a validity
+// window, and a DXpedition's window closes. Entering 3Y0K by hand months later,
+// with the activation's own date in the form, resolved the exception at TODAY's
+// date, found none, and fell back to cty.dat — Antarctica instead of Bouvet
+// Island. A backdated QSO must be enriched as of when it happened.
+func (a *App) LookupCallsign(callsign string, qsoDate string) (lookup.Result, error) {
+	return a.lookupCallsign(callsign, false, qsoDate)
 }
 
 // LookupCallsignFresh is the same, but SKIPS the cache and refreshes it.
@@ -7103,11 +7216,11 @@ func (a *App) LookupCallsign(callsign string) (lookup.Result, error) {
 // subscription went on getting the thin free-account record, and deleting the
 // cached row by hand was the only way out. A deliberate click must reach the
 // provider and overwrite what was stored.
-func (a *App) LookupCallsignFresh(callsign string) (lookup.Result, error) {
-	return a.lookupCallsign(callsign, true)
+func (a *App) LookupCallsignFresh(callsign string, qsoDate string) (lookup.Result, error) {
+	return a.lookupCallsign(callsign, true, qsoDate)
 }
 
-func (a *App) lookupCallsign(callsign string, force bool) (lookup.Result, error) {
+func (a *App) lookupCallsign(callsign string, force bool, qsoDate string) (lookup.Result, error) {
 	if a.lookup == nil {
 		return lookup.Result{}, fmt.Errorf("lookup not initialized")
 	}
@@ -7152,10 +7265,15 @@ func (a *App) lookupCallsign(callsign string, force bool) (lookup.Result, error)
 			r.ImageURL = ""
 		}
 	}
-	// ClubLog exception override (live entry → today's date): for an active
-	// DXpedition the entered call gets the right entity/zones immediately.
+	// ClubLog exception override, resolved AT THE QSO'S DATE.
+	//
+	// An exception carries a validity window and a DXpedition's window closes.
+	// Resolving at today's date is right for a contact happening now and wrong
+	// for one being entered afterwards: 3Y0K typed months later, with the
+	// activation's date in the form, found no live exception and fell back to
+	// cty.dat — Antarctica, where the log says Bouvet Island.
 	if a.clublogCtyEnabled() && a.clublog != nil {
-		if e, ok := a.clublog.Resolve(callsign, time.Now().UTC()); ok {
+		if e, ok := a.clublog.Resolve(callsign, lookupWhen(qsoDate)); ok {
 			r.Country = titleEntity(e.Entity)
 			if e.Cont != "" {
 				r.Continent = e.Cont
@@ -7172,6 +7290,19 @@ func (a *App) lookupCallsign(callsign string, force bool) (lookup.Result, error)
 		}
 	}
 	a.enrichFromULS(&r, callsign)
+	// An entity that IS a single island group answers for itself. QRZ carries
+	// <iota> only for the operators who filled it in, and most have not; but a
+	// station in Ascension is on AF-003 whatever its callbook record says, and
+	// the entity is known for every callsign from cty.dat alone.
+	//
+	// So this reaches the operator with no QRZ subscription, and the station
+	// that has never touched a callbook — and it is right when the two
+	// disagree only in the sense that the callbook wins: filled ONLY when the
+	// callbook gave nothing, because an operator who typed their own reference
+	// knows something a table cannot (an IOTA-heavy entity, a rare island).
+	if r.IOTA == "" && r.DXCC != 0 {
+		r.IOTA = awardref.IOTAForDXCC(r.DXCC)
+	}
 	// Custom outbound rows bound to a lookup. After the enrichment, so a
 	// template sees the same grid and county the entry panel is about to show —
 	// two answers for one callsign is how a rotator ends up pointed elsewhere
@@ -7273,9 +7404,13 @@ func (a *App) GetLookupSettings() (LookupSettings, error) {
 	if err != nil {
 		return LookupSettings{}, err
 	}
-	ttl, _ := strconv.Atoi(m[keyCacheTTL])
-	if ttl <= 0 {
-		ttl = 30
+	// Same rule as reloadLookupProviders: blank means "never set" and gets the
+	// default, while a stored zero is the operator asking for no cache at all.
+	ttl := 30
+	if raw := strings.TrimSpace(m[keyCacheTTL]); raw != "" {
+		if n, err := strconv.Atoi(raw); err == nil && n >= 0 {
+			ttl = n
+		}
 	}
 	return LookupSettings{
 		QRZUser:           m[keyQRZUser],
@@ -7295,8 +7430,9 @@ func (a *App) SaveLookupSettings(s LookupSettings) error {
 	if a.settings == nil {
 		return fmt.Errorf("db not initialized")
 	}
-	if s.CacheTTLDays <= 0 {
-		s.CacheTTLDays = 30
+	// Zero is kept — it means no cache. Only a negative number is nonsense.
+	if s.CacheTTLDays < 0 {
+		s.CacheTTLDays = 0
 	}
 	// Reject a primary == failsafe routing combo — would just hit the same
 	// provider twice. Frontend should prevent this but defend in depth.
@@ -7371,7 +7507,7 @@ func (a *App) GetCATSettings() (CATSettings, error) {
 	if a.settings == nil {
 		return CATSettings{Backend: "omnirig", OmniRigNum: 1, PollMs: 250}, fmt.Errorf("db not initialized")
 	}
-	m, err := a.settings.GetMany(a.ctx, keyCATEnabled, keyCATBackend, keyCATOmniRigNum, keyCATOmniRigVFO, keyCATFlexHost, keyCATFlexPort, keyCATFlexSpots, keyCATFlexDecodeSpots, keyCATFlexDecodeSecs, keyCATXieguPort, keyCATXieguBaud, keyCATXieguAddr, keyCATXieguPTTLine, keyCATYaesuPort, keyCATYaesuBaud, keyCATKenwoodPort, keyCATKenwoodBaud, keyCATKenwoodHost, keyCATYaesuLowLines, keyCATKenwoodLowLines, keyCATKenwoodDataMode, keyCATIcomPort, keyCATIcomBaud, keyCATIcomAddr, keyCATIcomNetHost, keyCATIcomNetUser, keyCATIcomNetPass, keyCATIcomNetAudio, keyCATTCIHost, keyCATTCIPort, keyCATTCISpots, keyCATPttHotkeyEnabled, keyCATPttHotkey, keyCATPttHotkeyToggle, keyCATPollMs, keyCATDelayMs, keyCATDigitalDefault, keyCATShareEnabled, keyCATSharePort)
+	m, err := a.settings.GetMany(a.ctx, keyCATEnabled, keyCATBackend, keyCATOmniRigNum, keyCATOmniRigVFO, keyCATFlexHost, keyCATFlexPort, keyCATFlexSpots, keyCATFlexDecodeSpots, keyCATFlexDecodeSecs, keyCATXieguPort, keyCATXieguBaud, keyCATXieguAddr, keyCATXieguPTTLine, keyCATYaesuPort, keyCATYaesuBaud, keyCATKenwoodPort, keyCATKenwoodBaud, keyCATKenwoodHost, keyCATYaesuLowLines, keyCATKenwoodLowLines, keyCATKenwoodDataMode, keyCATIcomPort, keyCATIcomBaud, keyCATIcomAddr, keyCATIcomNetHost, keyCATIcomNetUser, keyCATIcomNetPass, keyCATIcomNetAudio, keyCATTCIHost, keyCATTCIPort, keyCATTCISpots, keyCATPttHotkeyEnabled, keyCATPttHotkey, keyCATPttHotkeyToggle, keyCATPollMs, keyCATDelayMs, keyCATDigitalDefault, keyCATShareEnabled, keyCATSharePort, keyCATShareProto, keyCATShareTCIPort)
 	if err != nil {
 		return CATSettings{}, err
 	}
@@ -7413,6 +7549,8 @@ func (a *App) GetCATSettings() (CATSettings, error) {
 		DigitalDefault:   m[keyCATDigitalDefault],
 		ShareEnabled:     m[keyCATShareEnabled] == "1",
 		SharePort:        4532,
+		ShareProto:       "rigctl",
+		ShareTCIPort:     tciserver.DefaultPort,
 	}
 	if n, _ := strconv.Atoi(m[keyCATFlexPort]); n > 0 && n <= 65535 {
 		out.FlexPort = n
@@ -7426,6 +7564,12 @@ func (a *App) GetCATSettings() (CATSettings, error) {
 	if n, _ := strconv.Atoi(m[keyCATSharePort]); n > 0 && n <= 65535 {
 		out.SharePort = n
 	}
+	if p := strings.ToLower(strings.TrimSpace(m[keyCATShareProto])); p == "tci" {
+		out.ShareProto = p
+	}
+	if n, _ := strconv.Atoi(m[keyCATShareTCIPort]); n > 0 && n <= 65535 {
+		out.ShareTCIPort = n
+	}
 	if n, _ := strconv.Atoi(m[keyCATXieguBaud]); n > 0 {
 		out.XieguBaud = n
 	}
@@ -7483,6 +7627,12 @@ func (a *App) SaveCATSettings(s CATSettings) error {
 	if s.SharePort <= 0 || s.SharePort > 65535 {
 		s.SharePort = 4532
 	}
+	if s.ShareProto != "tci" {
+		s.ShareProto = "rigctl"
+	}
+	if s.ShareTCIPort <= 0 || s.ShareTCIPort > 65535 {
+		s.ShareTCIPort = tciserver.DefaultPort
+	}
 	if s.XieguBaud <= 0 {
 		s.XieguBaud = 19200
 	}
@@ -7583,6 +7733,8 @@ func (a *App) SaveCATSettings(s CATSettings) error {
 		keyCATDigitalDefault:   strings.ToUpper(strings.TrimSpace(s.DigitalDefault)),
 		keyCATShareEnabled:     shareEnabled,
 		keyCATSharePort:        strconv.Itoa(s.SharePort),
+		keyCATShareProto:       s.ShareProto,
+		keyCATShareTCIPort:     strconv.Itoa(s.ShareTCIPort),
 	} {
 		if err := a.settings.Set(a.ctx, k, v); err != nil {
 			return err
@@ -10224,6 +10376,15 @@ func (a *App) TestCloudlogUpload() (string, error) {
 // ── QSL Manager (manual upload) ────────────────────────────────────────
 
 // uploadColumnFor maps a service id to its QSO sent-status column.
+// manualUploadPace is the shortest gap between two consecutive single-QSO
+// uploads in a bulk "Send to …" run, for the services with no batch endpoint
+// (QRZ.com, HRDLog). Selecting twenty-five thousand QSOs in the QSL Manager and
+// firing them off as fast as the link allows is exactly the traffic a logbook
+// service reads as a robot rather than an operator — Club Log threatens to block
+// an IP for it. The gap is free in practice: a round trip to either already
+// takes longer than it.
+const manualUploadPace = 200 * time.Millisecond
+
 func uploadColumnFor(service string) string {
 	switch extsvc.Service(service) {
 	case extsvc.ServiceQRZ:
@@ -10318,10 +10479,13 @@ func (a *App) runManualUpload(svc extsvc.Service, ids []int64, cfg extsvc.Extern
 				}
 			}
 		}
-	} else if svc == extsvc.ServiceClublog || svc == extsvc.ServiceHRDLog {
+	} else if svc == extsvc.ServiceClublog || svc == extsvc.ServiceHRDLog || svc == extsvc.ServiceEQSL {
 		statusCol, dateCol := "clublog_qso_upload_status", "clublog_qso_upload_date"
-		if svc == extsvc.ServiceHRDLog {
+		switch svc {
+		case extsvc.ServiceHRDLog:
 			statusCol, dateCol = "hrdlog_qso_upload_status", "hrdlog_qso_upload_date"
+		case extsvc.ServiceEQSL:
+			statusCol, dateCol = "eqsl_sent", "eqsl_sent_date"
 		}
 		type item struct {
 			id   int64
@@ -10385,10 +10549,62 @@ func (a *App) runManualUpload(svc extsvc.Service, ids []int64, cfg extsvc.Extern
 					applog.Printf("extsvc: Club Log batch FAILED (%s) — QSOs: %s", msg, strings.Join(who, ", "))
 				}
 			}
+		} else if svc == extsvc.ServiceEQSL {
+			// eQSL's ImportADIF.cfm is a file importer — it answers "X out of Y
+			// records added" — so send a whole chunk per request instead of one
+			// request per contact. eQSL asks that an upload stay under about a
+			// thousand records; 100 keeps a single refused record from taking the
+			// rest of the chunk with it.
+			const chunk = 100
+			emit(fmt.Sprintf("eQSL: uploading %d QSO(s) in batches of %d…", len(items), chunk))
+			for start := 0; start < len(items); start += chunk {
+				end := start + chunk
+				if end > len(items) {
+					end = len(items)
+				}
+				batch := items[start:end]
+				recs := make([]string, len(batch))
+				batchIDs := make([]int64, len(batch))
+				for i, it := range batch {
+					recs[i] = it.rec
+					batchIDs[i] = it.id
+				}
+				res, err := extsvc.UploadEQSLBatch(ctx, nil, cfg.EQSL.Username, cfg.EQSL.Password, cfg.EQSL.QTHNickname, recs)
+				if err == nil && res.OK {
+					if merr := a.qso.MarkUploadedBatch(ctx, statusCol, dateCol, date, batchIDs); merr != nil {
+						applog.Printf("extsvc: eQSL batch mark: %v", merr)
+					}
+					uploaded += len(batch)
+					// eQSL took the file but left records out. It never says WHICH
+					// — normally they are QSOs it already holds — so quote its own
+					// count rather than claim a clean run.
+					if res.Ignored {
+						emit(fmt.Sprintf("eQSL: %d/%d uploaded — %s", end, len(items), res.Message))
+					} else {
+						emit(fmt.Sprintf("eQSL: %d/%d uploaded", end, len(items)))
+					}
+				} else {
+					msg := res.Message
+					if err != nil {
+						msg = err.Error()
+					}
+					// Name the QSOs in the failing batch, same as Club Log: a
+					// per-record rejection is otherwise impossible to locate.
+					who := make([]string, 0, len(batch))
+					for _, it := range batch {
+						who = append(who, fmt.Sprintf("%s#%d", it.call, it.id))
+					}
+					emit(fmt.Sprintf("eQSL: batch of %d FAILED: %s", len(batch), msg))
+					applog.Printf("extsvc: eQSL batch FAILED (%s) — QSOs: %s", msg, strings.Join(who, ", "))
+				}
+			}
 		} else {
 			// HRDLog's NewEntry.aspx inserts only the FIRST record of a multi-
 			// record ADIF, so upload ONE record per request. The DB stays cheap:
 			// bulk fetch above + the marks flushed in batches (not one per QSO).
+			// Paced: HRDLog is the one service here with no way to batch, and a
+			// few thousand requests as fast as the link allows is what a logbook
+			// reads as a robot.
 			emit(fmt.Sprintf("HRDLog: uploading %d QSO(s) (one request each)…", len(items)))
 			var doneIDs []int64
 			flush := func() {
@@ -10401,6 +10617,9 @@ func (a *App) runManualUpload(svc extsvc.Service, ids []int64, cfg extsvc.Extern
 				doneIDs = doneIDs[:0]
 			}
 			for i, it := range items {
+				if i > 0 {
+					time.Sleep(manualUploadPace)
+				}
 				res, err := extsvc.UploadHRDLog(ctx, nil, cfg.HRDLog.Callsign, cfg.HRDLog.Code, it.rec)
 				if err == nil && res.OK {
 					doneIDs = append(doneIDs, it.id)
@@ -10422,34 +10641,26 @@ func (a *App) runManualUpload(svc extsvc.Service, ids []int64, cfg extsvc.Extern
 			flush()
 		}
 	} else {
-		// QRZ.com: one record per request (its logbook API has no batch upload).
-		for _, id := range ids {
+		// QRZ.com: one record per request (its logbook API has no batch upload),
+		// paced for the same reason as HRDLog above.
+		for i, id := range ids {
+			if i > 0 {
+				time.Sleep(manualUploadPace)
+			}
 			q, gerr := a.qso.GetByID(ctx, id)
 			call := ""
 			if gerr == nil {
 				call = q.Callsign
 			}
-			force := ""
-			if svc == extsvc.ServiceQRZ {
-				force = cfg.QRZ.ForceStationCallsign
-			}
-			rec, ok := a.buildUploadADIF(id, force)
+			// QRZ rewrites STATION_CALLSIGN to the registered call.
+			rec, ok := a.buildUploadADIF(id, cfg.QRZ.ForceStationCallsign)
 			if !ok {
 				emit(call + " — skipped (no record)")
 				continue
 			}
-			var res extsvc.UploadResult
-			var err error
-			switch svc {
-			case extsvc.ServiceQRZ:
-				res, err = extsvc.UploadQRZ(ctx, nil, cfg.QRZ.APIKey, rec)
-			case extsvc.ServiceHRDLog:
-				res, err = extsvc.UploadHRDLog(ctx, nil, cfg.HRDLog.Callsign, cfg.HRDLog.Code, rec)
-			case extsvc.ServiceEQSL:
-				res, err = extsvc.UploadEQSL(ctx, nil, cfg.EQSL.Username, cfg.EQSL.Password, cfg.EQSL.QTHNickname, rec)
-			default:
-				res, err = extsvc.UploadClublog(ctx, nil, cfg.Clublog, rec)
-			}
+			// Only QRZ reaches this branch: LoTW, Club Log, HRDLog and eQSL are
+			// all handled above, and UploadQSOsManual rejects anything else.
+			res, err := extsvc.UploadQRZ(ctx, nil, cfg.QRZ.APIKey, rec)
 			if err == nil && res.OK {
 				a.markExtUploaded(svc, id, "")
 				uploaded++
@@ -14788,6 +14999,13 @@ type StationDevice struct {
 	OffURLs []string `json:"off_urls,omitempty"`
 	OnPat   string   `json:"on_pattern,omitempty"` // fallback, {relay} substituted
 	OffPat  string   `json:"off_pattern,omitempty"`
+	// InsecureTLS accepts an HTTPS certificate that cannot be verified — which
+	// is the only kind a relay board on the LAN can present, having signed it
+	// itself. Off by default, because the other HTTPS case is the opposite one:
+	// a board reached from outside through a proxy with a real certificate,
+	// where verification is what stands between an antenna switch and the
+	// internet.
+	InsecureTLS bool `json:"insecure_tls,omitempty"`
 }
 
 // deviceRelayCount is the relay count for a configured device — fixed by type,
@@ -14848,7 +15066,7 @@ func buildDeviceDriver(d StationDevice) relaydev.Device {
 		// generic board fall through to the WebSwitch driver below: it answered
 		// the WebSwitch's own address, never sent one configured URL, and
 		// reported itself offline so every relay button stayed greyed out.
-		return relaydev.NewHTTPGeneric(d.OnURLs, d.OffURLs, d.OnPat, d.OffPat, d.User, d.Pass, deviceRelayCount(d), d.Labels)
+		return relaydev.NewHTTPGeneric(d.OnURLs, d.OffURLs, d.OnPat, d.OffPat, d.User, d.Pass, deviceRelayCount(d), d.Labels, d.InsecureTLS)
 	default:
 		return relaydev.NewWebswitch(d.Host)
 	}
@@ -14868,7 +15086,10 @@ func deviceKey(d StationDevice) string {
 			// The labels are part of the wire format here: {value} sends them.
 			// Renaming a relay re-addresses it, and the cached driver would keep
 			// commanding the old name.
-			"|" + strings.Join(d.Labels, "\x1f")
+			"|" + strings.Join(d.Labels, "\x1f") +
+			// Ticking the box has to rebuild the driver: the cached one holds the
+			// verifying client and would go on refusing the certificate.
+			fmt.Sprintf("|%t", d.InsecureTLS)
 	}
 	return k
 }
@@ -16588,13 +16809,55 @@ func (a *App) AmpOperate(id string, on bool) error {
 	// Fan out here rather than in the UI: the card and the docked widget both
 	// call this, and a coupling implemented in one of them would be missing from
 	// the other — which on a combiner means one amplifier keyed and one not.
-	var firstErr error
-	for _, tid := range a.ampTargets(id, a.GetLinkedAmps()) {
-		if err := a.ampOperateOne(tid, on); err != nil && firstErr == nil {
-			firstErr = err
+	return a.ampFanOut(a.ampTargets(id, a.GetLinkedAmps()), func(tid string) error {
+		return a.ampOperateOne(tid, on)
+	})
+}
+
+// ampFanOut runs one command against every amplifier AT THE SAME TIME.
+//
+// Sequentially, the second amplifier of a combined pair was commanded only once
+// the first had answered — and an SPE answers over its own link, in its own
+// time. The combiner heard power appear on one input before the other and
+// complained about it, on every OFF and every ON.
+//
+// So each target gets a goroutine, and they are all parked on the same channel
+// until every one of them is ready. Closing it releases them together: the
+// difference between "start one, then start the other" and "both leave at
+// once". They have separate clients and separate connections, so nothing
+// downstream re-serialises them.
+//
+// A single target — the ordinary case of one amplifier — runs inline. No
+// goroutine, no barrier, nothing to go wrong for the operators who have one amp.
+func (a *App) ampFanOut(targets []string, do func(id string) error) error {
+	if len(targets) == 0 {
+		return nil
+	}
+	if len(targets) == 1 {
+		return do(targets[0])
+	}
+	errs := make([]error, len(targets))
+	start := make(chan struct{})
+	var wg sync.WaitGroup
+	for i, id := range targets {
+		wg.Add(1)
+		go func(i int, id string) {
+			defer wg.Done()
+			<-start
+			errs[i] = do(id)
+		}(i, id)
+	}
+	close(start)
+	wg.Wait()
+	// The amplifier the operator actually clicked is first, and its failure is
+	// the one worth reporting: "the amp I pressed did not respond" beats the
+	// same message about its silent partner.
+	for _, err := range errs {
+		if err != nil {
+			return err
 		}
 	}
-	return firstErr
+	return nil
 }
 
 func (a *App) ampOperateOne(id string, on bool) error {
@@ -16622,15 +16885,11 @@ func (a *App) AmpPower(id string, on bool) (err error) {
 			applog.Printf("amp %s: power %v failed: %v", id, on, err)
 		}
 	}()
-	linked := a.GetLinkedAmps()
-	targets := a.ampTargets(id, linked)
-	var firstErr error
-	for _, tid := range targets {
-		if e := a.ampPowerOne(tid, on, len(targets) > 1); e != nil && firstErr == nil {
-			firstErr = e
-		}
-	}
-	return firstErr
+	targets := a.ampTargets(id, a.GetLinkedAmps())
+	multi := len(targets) > 1
+	return a.ampFanOut(targets, func(tid string) error {
+		return a.ampPowerOne(tid, on, multi)
+	})
 }
 
 func (a *App) ampPowerOne(id string, on, linked bool) error {
@@ -16660,12 +16919,31 @@ func (a *App) ampPowerOne(id string, on, linked bool) error {
 }
 
 // AmpPowerLevel selects the output power level — SPE only (L/M/H).
+//
+// Coupled like ON, OFF and OPERATE, which it was not: two combined amplifiers
+// left at different power levels feed the combiner unevenly, which is the thing
+// the coupling exists to prevent. It was simply the command nobody had linked.
+//
+// Running the pair concurrently matters more here than anywhere else. Setting a
+// level is not one command: an SPE has no "set level" at all, so the driver taps
+// the POWER key and waits for the amp to report the new level before tapping
+// again — up to three taps, up to two seconds each. One after the other, the two
+// amplifiers would sit at different levels for as long as six seconds.
 func (a *App) AmpPowerLevel(id, level string) error {
-	inst := a.ampInstByID(id)
-	if inst == nil || inst.spe == nil {
-		return fmt.Errorf("power level is an SPE feature")
-	}
-	return inst.spe.SetPowerLevel(level)
+	targets := a.ampTargets(id, a.GetLinkedAmps())
+	multi := len(targets) > 1
+	return a.ampFanOut(targets, func(tid string) error {
+		inst := a.ampInstByID(tid)
+		if inst == nil || inst.spe == nil {
+			// A PowerGenius sitting in the group has no L/M/H, and saying so
+			// would make a successful SPE pair look like a failure.
+			if multi {
+				return nil
+			}
+			return fmt.Errorf("power level is an SPE feature")
+		}
+		return inst.spe.SetPowerLevel(level)
+	})
 }
 
 // AmpFanMode sets the fan mode — PGXL only (STANDARD/CONTEST/BROADCAST).
@@ -17494,16 +17772,26 @@ func (a *App) SendClusterCommand(cmd string) error {
 	if cmd == "" {
 		return fmt.Errorf("empty command")
 	}
-	servers, err := a.listClusterServers()
+	srv, err := a.masterClusterServer()
 	if err != nil {
 		return err
 	}
+	return a.cluster.SendCommand(srv.ID, cmd)
+}
+
+// masterClusterServer returns the master node — the first ENABLED server in
+// sort order, which is where commands and spots go.
+func (a *App) masterClusterServer() (cluster.ServerConfig, error) {
+	servers, err := a.listClusterServers()
+	if err != nil {
+		return cluster.ServerConfig{}, err
+	}
 	for _, s := range servers {
 		if s.Enabled {
-			return a.cluster.SendCommand(s.ID, cmd)
+			return s, nil
 		}
 	}
-	return fmt.Errorf("no enabled cluster server to send to")
+	return cluster.ServerConfig{}, fmt.Errorf("no enabled cluster server to send to")
 }
 
 // SendClusterSpot announces a DX spot on the **master** cluster (first
@@ -17525,8 +17813,26 @@ func (a *App) SendClusterSpot(call string, freqKHz float64, comment string) erro
 	if c := strings.TrimSpace(comment); c != "" {
 		cmd += " " + c
 	}
+	if a.cluster == nil {
+		return fmt.Errorf("cluster not initialized")
+	}
+	srv, err := a.masterClusterServer()
+	if err != nil {
+		return err
+	}
 	applog.Printf("cluster: send spot — freqKHz=%v → command %q", freqKHz, cmd)
-	return a.SendClusterCommand(cmd)
+	if err := a.cluster.SendCommand(srv.ID, cmd); err != nil {
+		return err
+	}
+	// Show it in OUR OWN spot list straight away. Most nodes never broadcast a
+	// spot back to the station that sent it, so the operator saw nothing appear
+	// and reported the spot as not sent — several times. It goes through the same
+	// queue as a spot off the wire, so it gets the same DXCC/POTA enrichment,
+	// alert evaluation and panadapter mirroring, and the UI de-dupes it against
+	// the node's echo when there is one.
+	sp := cluster.NewLocalSpot(srv, a.resolveClusterLogin(srv.LoginOverride), call, freqKHz, comment)
+	a.enqueueClusterEvent(clusterEvent{spot: &sp})
+	return nil
 }
 
 // GetClusterStatus returns a snapshot of every active session. Used by
@@ -18099,6 +18405,18 @@ func (r catShareRig) Split() (bool, int64) {
 	return true, st.FreqHz
 }
 
+// RxFreq is where we LISTEN. Only the TCI server asks for it: TCI's channel A
+// is the receive frequency and channel B the transmit one, the opposite way
+// round from RigState, and a client handed these two the wrong way about would
+// transmit on the DX's own frequency.
+func (r catShareRig) RxFreq() int64 {
+	st := r.a.cat.State()
+	if st.Split && st.RxFreqHz > 0 {
+		return st.RxFreqHz
+	}
+	return st.FreqHz
+}
+
 func (r catShareRig) SetFreq(hz int64) error { return r.a.cat.SetFrequency(hz) }
 func (r catShareRig) SetMode(m string) error { return r.a.cat.SetMode(m) }
 func (r catShareRig) SetPTT(on bool) error   { return r.a.cat.SetPTT(on) }
@@ -18113,17 +18431,35 @@ func (r catShareRig) SetSplit(on bool, txHz int64) error {
 
 // reloadCATShare starts, stops or restarts the sharing server to match the
 // settings. Called from reloadCAT so one "Save & Close" settles both.
+//
+// One server or the other, never both. rigctl and TCI are two ways of asking
+// the same radio the same questions; running both would only double the ways a
+// port clash or a confused client can go wrong, and no program speaks both.
 func (a *App) reloadCATShare(s CATSettings) {
 	want := s.Enabled && s.ShareEnabled
-	// Always tear down first: the port may have changed, and a listener bound to
-	// the old one would keep answering while the client is told to use the new.
+	// Always tear down first: the port — or the protocol — may have changed, and
+	// a listener bound to the old one would keep answering while the client is
+	// told to use the new.
 	if a.catShare != nil {
 		a.catShare.Stop()
 		a.catShare = nil
 	}
+	if a.catShareTCI != nil {
+		a.catShareTCI.Stop()
+		a.catShareTCI = nil
+	}
 	if !want {
 		return
 	}
+	if s.ShareProto == "tci" {
+		srv := tciserver.New(s.ShareTCIPort, catShareRig{a: a}, applog.Printf)
+		if err := srv.Start(); err != nil {
+			applog.Printf("cat share: %v", err)
+			return
+		}
+		a.catShareTCI = srv
+		return
+	}
 	srv := rigctld.New(s.SharePort, catShareRig{a: a}, applog.Printf)
 	if err := srv.Start(); err != nil {
 		// The usual cause is another rigctld — or a previous OpsLog — already on
@@ -18476,3 +18812,31 @@ func wsjtLoggedQSO(q qso.QSO) udp.LoggedQSO {
 	}
 	return out
 }
+
+// lookupWhen turns the entry form's date into the instant a ClubLog exception
+// should be resolved at.
+//
+// Empty, unparseable, or in the future → now. A date is only trusted to move
+// the resolution BACKWARDS: a half-typed "2026-0" must not send the lookup to
+// the year 20, and a mistyped future date must not resolve against an exception
+// window that has not opened.
+//
+// Midday UTC rather than midnight: an exception window given in whole days is
+// inclusive of its end date, and resolving at 00:00 of that day sits on the
+// boundary where an off-by-one in either direction changes the answer.
+func lookupWhen(qsoDate string) time.Time {
+	now := time.Now().UTC()
+	s := strings.TrimSpace(qsoDate)
+	if s == "" {
+		return now
+	}
+	t, err := time.Parse("2006-01-02", s)
+	if err != nil {
+		return now
+	}
+	t = t.Add(12 * time.Hour)
+	if t.After(now) {
+		return now
+	}
+	return t
+}

awardsnap_test.go

@@ -0,0 +1,84 @@
+package main
+
+import (
+	"sync"
+	"testing"
+	"time"
+)
+
+// The award snapshot must be built ONCE however many callers ask at once.
+//
+// The cache lock is released before the logbook is read, so every caller that
+// arrives during a build used to miss the cache and start its own. Opening the
+// Awards panel does exactly that: a field log showed three pulls of the same
+// 123 615 QSOs inside ten seconds, and three copies alive together took the Go
+// heap from 725 MB to 2.5 GB.
+//
+// This models the same shape — a cheap cache check, a slow build, a shared
+// result — against the pattern awardSnapshot now uses, so the invariant is
+// pinned without needing a logbook.
+func TestSnapshotBuildsOncePerRevision(t *testing.T) {
+	var (
+		cacheMu sync.Mutex
+		buildMu sync.Mutex
+		cached  []int
+		rev     = "r1"
+		gotRev  string
+		builds  int
+	)
+
+	get := func() []int {
+		cacheMu.Lock()
+		if cached != nil && gotRev == rev {
+			defer cacheMu.Unlock()
+			return cached
+		}
+		cacheMu.Unlock()
+
+		buildMu.Lock()
+		defer buildMu.Unlock()
+		// Re-check: whoever held the build lock has just finished.
+		cacheMu.Lock()
+		if cached != nil && gotRev == rev {
+			defer cacheMu.Unlock()
+			return cached
+		}
+		cacheMu.Unlock()
+
+		time.Sleep(50 * time.Millisecond) // the logbook read
+		out := []int{1, 2, 3}
+
+		cacheMu.Lock()
+		builds++
+		cached, gotRev = out, rev
+		cacheMu.Unlock()
+		return out
+	}
+
+	var wg sync.WaitGroup
+	results := make([][]int, 8)
+	for i := range results {
+		wg.Add(1)
+		go func(i int) { defer wg.Done(); results[i] = get() }(i)
+	}
+	wg.Wait()
+
+	if builds != 1 {
+		t.Errorf("%d builds for one revision — each concurrent caller pulled the whole logbook again", builds)
+	}
+	for i, r := range results {
+		if len(r) != 3 {
+			t.Errorf("caller %d got %v", i, r)
+		}
+	}
+
+	// A new revision must rebuild: the guard is against duplicate work, not
+	// against a logbook that changed.
+	cacheMu.Lock()
+	rev = "r2"
+	cacheMu.Unlock()
+	get()
+	if builds != 2 {
+		t.Errorf("builds = %d after the revision moved, want 2 — a changed logbook must be re-read", builds)
+	}
+}

changelog.json

@@ -1,4 +1,66 @@
 [
+  {
+    "version": "0.25.8",
+    "date": "",
+    "en": [
+      "An entity that is a single island group now fills the IOTA reference on its own — no callbook subscription needed.",
+      "CAT sharing can now speak TCI instead of Hamlib, split included, so a TCI-only program reaches whatever radio you are on.",
+      "Lookup cache: a TTL of 0 switches it off, so a callbook record you are correcting is re-read every time.",
+      "TCI radios: when the rig forbids transmitting, PTT says so instead of doing nothing silently.",
+      "WAJA carried Japan’s civil prefecture numbers instead of the JARL’s: 35 of the 47 references are renumbered.",
+      "Award references can be renumbered in the editor — the number was the one field it would not let you correct.",
+      "The compass fills the moment Station Control opens, instead of waiting out the rest of a polling interval.",
+      "Combined amplifiers: the power level (L/M/H) is coupled too, and both amps are commanded at once so the combiner stops beeping.",
+      "Generic HTTP relay: an https:// board can be accepted with its own self-signed certificate, per board.",
+      "Club Log: the on-close upload now goes out as one batch — sending hundreds of contacts one at a time got operators blocked.",
+      "eQSL: uploads go out in batches of 100 too, and QRZ.com and HRDLog — which have no batch upload — are spaced out instead.",
+      "A spot you send now shows in your own spot list — most nodes never echo it back, so it looked like nothing had gone out.",
+      "Icom with JTDX in Fake It split: a lost PTT acknowledgement is sent again instead of failing, which made JTDX drop the rig.",
+      "A cluster whose only greeting is “login:” and which then asks for a password now connects — both prompts were being missed."
+    ],
+    "fr": [
+      "Une entité qui est un seul groupe d’îles remplit désormais la référence IOTA toute seule, sans abonnement callbook.",
+      "Le partage CAT peut désormais parler TCI au lieu de Hamlib, split compris : un logiciel TCI atteint la radio, quelle qu’elle soit.",
+      "Cache des recherches : un TTL à 0 le désactive, pour relire à chaque fois une fiche callbook en cours de correction.",
+      "Radios TCI : quand la radio interdit l’émission, le PTT le dit au lieu de ne rien faire en silence.",
+      "WAJA portait les numéros civils des préfectures japonaises et non ceux de la JARL : 35 des 47 références sont renumérotées.",
+      "Les références d’un diplôme se renumérotent dans l’éditeur : le numéro était le seul champ qu’il refusait de corriger.",
+      "La boussole se remplit dès l’ouverture de Station Control, au lieu d’attendre la fin d’un intervalle d’interrogation.",
+      "Amplis combinés : le niveau de puissance (L/M/H) est couplé lui aussi, et les deux amplis sont commandés en même temps — fini le bip du combineur.",
+      "Relais HTTP générique : une carte en https:// peut être acceptée avec son certificat auto-signé, carte par carte.",
+      "Club Log : l’envoi à la fermeture part désormais en un lot — envoyer des centaines de contacts un par un faisait bloquer l’opérateur.",
+      "eQSL : les envois partent aussi par lots de 100, et QRZ.com et HRDLog — qui n’ont pas d’envoi groupé — sont espacés à la place.",
+      "Un spot que tu envoies apparaît maintenant dans ta liste : la plupart des nœuds ne le renvoient pas, il semblait n’être jamais parti.",
+      "Icom avec JTDX en split Fake It : un accusé de réception PTT perdu est renvoyé au lieu d’échouer — JTDX lâchait le poste.",
+      "Un cluster dont tout l’accueil est « login: » puis qui réclame un mot de passe se connecte : les deux invites étaient ignorées."
+    ]
+  },
+  {
+    "version": "0.25.7",
+    "date": "",
+    "en": [
+      "Opening the Awards panel no longer pulls the whole logbook several times at once — a large log briefly took gigabytes of memory.",
+      "Awards: the Missing refs button now only appears where it means something — a worldwide award like POTA could only ever answer “nothing found”.",
+      "Worked before: a prefixed call like ZA/OE8NDR matched every other visitor to that country instead of that one operator.",
+      "French: seventeen strings were still in English, the whole update panel among them, plus Spot lifetime and Chase new grids.",
+      "Selecting a QSO shows the entity the QSO records, not one re-derived from its callsign — a 3Y0K contact logged as Bouvet showed the Antarctica matrix.",
+      "Back-entering a QSO resolves the ClubLog exception at the CONTACT’S date, so a DXpedition entered months later gets the entity it had then.",
+      "QRZ.com sends an island reference for an operator on one, and OpsLog read past it — it now fills the IOTA award reference before the QSO is logged.",
+      "Sync across PCs: point every OpsLog at one folder you already synchronise and your contacts follow you between machines.",
+      "TCI: a station spotted by several operators is drawn once on the panorama instead of two or three times a few hertz apart."
+    ],
+    "fr": [
+      "Ouvrir le panneau Awards ne tire plus plusieurs fois le journal entier en même temps — un gros log occupait brièvement des gigaoctets de mémoire.",
+      "Awards : le bouton Réf. manquantes n’apparaît plus que là où il a un sens — un diplôme mondial comme POTA ne pouvait répondre que « aucun manque ».",
+      "Déjà contacté : un indicatif préfixé comme ZA/OE8NDR rapprochait tous les autres visiteurs du pays au lieu de ce seul opérateur.",
+      "Français : dix-sept textes étaient restés en anglais, dont tout le panneau de mise à jour, la durée de vie des spots et Chasser les nouveaux locators.",
+      "Sélectionner un QSO affiche l’entité que le QSO enregistre, pas une recalculée depuis l’indicatif — un 3Y0K logué Bouvet montrait la matrice Antarctique.",
+      "Saisir un QSO a posteriori résout l’exception ClubLog à la date DU CONTACT : une DXpedition entrée des mois après retrouve l’entité qu’elle avait alors.",
+      "QRZ.com envoie la référence d’île d’un opérateur sur une île, et OpsLog l’ignorait — elle remplit désormais la référence IOTA avant l’enregistrement du QSO.",
+      "Synchro entre PC : fais pointer chaque OpsLog vers un dossier déjà synchronisé et tes contacts te suivent d’une machine à l’autre.",
+      "TCI : une station spottée par plusieurs opérateurs n’est tracée qu’une fois sur le panorama, au lieu de deux ou trois fois."
+    ]
+  },
   {
     "version": "0.25.6",
     "date": "",

frontend/src/App.tsx

@@ -56,7 +56,7 @@ import {
   QSLViaRepairStatus, RepairQSLVia, DismissQSLViaRepair,
 } from '../wailsjs/go/main/App';
 import { Combobox } from '@/components/ui/combobox';
-import { applyAwardRefs, parseAwardRefs as parseManualRefs, spotRefList } from '@/lib/awardRefs';
+import { applyAwardRefs, parseAwardRefs as parseManualRefs, spotRefList , withIOTARef } from '@/lib/awardRefs';
 import { EventsOn, BrowserOpenURL, WindowMinimise, WindowToggleMaximise, WindowIsMaximised, Quit } from '../wailsjs/runtime/runtime';
 import type { adif as adifModels, lookup as lookupModels, cat as catModels } from '../wailsjs/go/models';
 import type { QSOForm, WorkedBeforeView, StationSettingsForm, ListsSettingsForm, ModePresetForm } from '@/types';
@@ -1780,7 +1780,10 @@ export default function App() {
   // Stats (F1) matrix whenever the entry form is empty — clicking a past contact
   // is the natural way to ask "what else have I got with this one?", and until
   // now the panel just sat blank.
-  const [selQso, setSelQso] = useState<{ call: string; band: string; mode: string } | null>(null);
+  // The selected row's own entity travels with it. Re-deriving it from the
+  // callsign is what showed Antarctica for a 3Y0K contact the log records as
+  // Bouvet Island — see the WorkedBefore call below.
+  const [selQso, setSelQso] = useState<{ call: string; band: string; mode: string; dxcc: number } | null>(null);
   const [bulkEditIds, setBulkEditIds] = useState<number[]>([]);
   const [bulkEditOpen, setBulkEditOpen] = useState(false);
   const [showSettings, setShowSettings] = useState(false);
@@ -3752,7 +3755,19 @@ export default function App() {
     if (!call || callsign.trim()) { setSelWb(null); return; }
     let dead = false;
     setSelWbBusy(true);
-    WorkedBefore(call, 0)
+    // The SELECTED QSO's own entity, not one re-derived from its callsign.
+    //
+    // Passing 0 lets the backend resolve the entity from cty.dat and the
+    // ClubLog exceptions AS THEY ARE TODAY, which is right for a live contact
+    // and wrong for one being looked at in the log. A 3Y0K contact recorded as
+    // Bouvet Island showed the Antarctica matrix, and counted five contacts
+    // "with this entity" against eleven with the call — two different entities,
+    // one of them nobody asked about.
+    //
+    // Browsing the log shows what the log says, even where the log is wrong.
+    // Correcting an entity is a deliberate act (right-click → Update from
+    // ClubLog), not something a panel does behind the operator's back.
+    WorkedBefore(call, selQso?.dxcc || 0)
       .then((w: any) => { if (!dead) setSelWb(w); })
       .catch(() => { if (!dead) setSelWb(null); })
       .finally(() => { if (!dead) setSelWbBusy(false); });
@@ -3861,7 +3876,11 @@ export default function App() {
     const gen = lookupGenRef.current; // invalidated by ESC / resetEntry
     setLookupBusy(true);
     try {
-      const r = await LookupCallsign(call);
+      // The ENTRY'S date, so a ClubLog exception resolves at the moment the
+      // contact happened. Live, that is now and nothing changes; back-entering
+      // a past QSO, it is what makes 3Y0K come back Bouvet Island instead of
+      // Antarctica — the exception's window had closed by today.
+      const r = await LookupCallsign(call, qsoStartedAt ? qsoStartedAt.toISOString().slice(0, 10) : '');
       // Discard a STALE result: the operator already moved to another call
       // (clicked a new spot / typed) OR cleared the entry (ESC) while this lookup
       // was in flight. Applying it would clobber the current fields and zoom the
@@ -3915,6 +3934,24 @@ export default function App() {
         email:   d.email   || (r.email ?? ''),
         web:     d.web     || (r.web ?? ''),
         qsl_via: d.qsl_via || (r.qsl_via ?? ''),
+        // An island reference from the callbook becomes an IOTA award reference
+        // on the entry, exactly like one picked by hand.
+        //
+        // QRZ carries <iota> for an operator on an island and OpsLog read past
+        // it. That matters more for IOTA than it would for another award:
+        // unlike POTA there is no live "who is on an island right now" feed
+        // anywhere, so the callbook record is the practical source — and it is
+        // known BEFORE the contact is logged, which is when it is useful.
+        //
+        // Only when the operator has not set one already: a reference typed or
+        // picked by hand outranks a callbook that may be years out of date.
+        //
+        // This belongs HERE, on the provider result, and not in the backfill
+        // from the last QSO where it first went: that runs only when there IS a
+        // previous contact with the call, so the island appeared for stations
+        // already in the log and never for the new one on the island — which is
+        // the entire point of the feature.
+        award_refs: withIOTARef(d.award_refs ?? '', String((r as any)?.iota ?? '')),
       }));
       // Backfill anything the provider didn't supply from the last time we worked
       // this call (call not found on QRZ/HamQTH, or lookup off → cty.dat only).
@@ -5359,7 +5396,7 @@ export default function App() {
               onExportFiltered={exportFilteredADIF}
               onDelete={(ids) => setDeletingIds(ids)}
               onRowSelected={(ids) => { setSelectedIds(ids); setSelectedId(ids[0] ?? null); }}
-              onRowSelectedQso={(r) => setSelQso(r ? { call: String(r.callsign ?? ""), band: String(r.band ?? ""), mode: String(r.mode ?? "") } : null)}
+              onRowSelectedQso={(r) => setSelQso(r ? { call: String(r.callsign ?? ""), band: String(r.band ?? ""), mode: String(r.mode ?? ""), dxcc: Number(r.dxcc ?? 0) } : null)}
             />
           </div>
         );
@@ -6677,7 +6714,7 @@ export default function App() {
                 onExportCabrilloFiltered={exportFilteredCabrillo}
                 onDelete={(ids) => setDeletingIds(ids)}
                 onRowSelected={(ids) => { setSelectedIds(ids); setSelectedId(ids[0] ?? null); }}
-              onRowSelectedQso={(r) => setSelQso(r ? { call: String(r.callsign ?? ""), band: String(r.band ?? ""), mode: String(r.mode ?? "") } : null)}
+              onRowSelectedQso={(r) => setSelQso(r ? { call: String(r.callsign ?? ""), band: String(r.band ?? ""), mode: String(r.mode ?? ""), dxcc: Number(r.dxcc ?? 0) } : null)}
               />
               <div className="px-3 py-1.5 border-t border-border/60 text-[11px] text-muted-foreground flex items-center justify-between gap-3 bg-muted/30">
                 <div className="flex items-center gap-3">

frontend/src/components/AwardEditor.tsx

@@ -14,7 +14,7 @@ import { useI18n } from '@/lib/i18n';
 import {
   GetAwardDefs, SaveAwardDefs, ResetAwardDefs, AwardFields,
   GetAwardReferenceMeta, UpdateAwardReferenceList,
-  ListAwardReferences, SearchAwardReferences, SaveAwardReference, DeleteAwardReference,
+  ListAwardReferences, SearchAwardReferences, SaveAwardReference, DeleteAwardReference, RenameAwardReference,
   ImportAwardReferencesText, GetAwardPresets, ApplyAwardPreset,
   ListCountries, DXCCForCountry, DXCCName,
   PopulateBuiltinReferences, HasBuiltinReferences,
@@ -915,6 +915,10 @@ function ReferencesPanel({ code, presets, meta, awardValidFrom, awardValidTo, on
   const [refs, setRefs] = useState<AwardRef[]>([]);
   const [q, setQ] = useState('');
   const [selCode, setSelCode] = useState<string | null>(null);
+  // The code as TYPED. The list and every patch key off selCode, so editing the
+  // code in place would make the editor lose the reference mid-edit; the draft
+  // is applied as a rename when the operator saves.
+  const [codeDraft, setCodeDraft] = useState('');
   const [busy, setBusy] = useState(false);
   const [bulk, setBulk] = useState('');
   const [showBulk, setShowBulk] = useState(false);
@@ -952,6 +956,7 @@ function ReferencesPanel({ code, presets, meta, awardValidFrom, awardValidTo, on
   }
 
   const sel = refs.find((r) => r.code === selCode) || null;
+  useEffect(() => { setCodeDraft(selCode ?? ''); }, [selCode]);
   // Large lists are already filtered by the server; small lists filter locally.
   const filtered = useMemo(() => {
     if (large) return refs;
@@ -965,6 +970,27 @@ function ReferencesPanel({ code, presets, meta, awardValidFrom, awardValidTo, on
     try { await SaveAwardReference(code, r as any); load(); onChanged(); }
     catch (e: any) { setErr(String(e?.message ?? e)); }
   }
+  // Save the selected reference, renumbering it first when the code was edited.
+  //
+  // The rename has to come first and has to be a rename: saving under the new
+  // code would simply create a second reference and leave the old one behind,
+  // which is how a list quietly grows duplicates.
+  async function saveSelected(r: AwardRef) {
+    const next = codeDraft.trim().toUpperCase();
+    if (!next) { setErr(t('awed.refCodeEmpty')); return; }
+    if (next !== r.code) {
+      try {
+        await RenameAwardReference(code, r.code, next);
+      } catch (e: any) {
+        // Most often the number is already taken by another reference. Said
+        // here rather than swallowed: the save has NOT happened.
+        setErr(String(e?.message ?? e));
+        return;
+      }
+      setSelCode(next);
+    }
+    await saveRef({ ...r, code: next });
+  }
   async function addRef() {
     const c = prompt(t('awed.newRefCodePrompt'))?.trim().toUpperCase();
     if (!c) return;
@@ -1046,7 +1072,13 @@ function ReferencesPanel({ code, presets, meta, awardValidFrom, awardValidTo, on
           ) : (
             <div className="space-y-2">
               <div className="flex items-center gap-2">
-                <Input className="h-8 w-28 font-mono font-semibold" value={sel.code} readOnly />
+                {/* Editable, because a shipped list can be wrong about it: WAJA
+                    went out carrying Japan's civil prefecture numbers instead of
+                    the JARL's, and correcting that meant deleting all 47
+                    references and importing a new list. */}
+                <Input className="h-8 w-28 font-mono font-semibold" value={codeDraft}
+                  title={t('awed.refCodeTip')}
+                  onChange={(e) => setCodeDraft(e.target.value)} />
                 <label className="flex items-center gap-1.5 text-xs cursor-pointer"><Checkbox checked={sel.valid} onCheckedChange={(c) => patchSel({ valid: !!c })} /> {t('awed.valid')}</label>
                 <div className="flex-1" />
                 <button className="text-muted-foreground hover:text-destructive" onClick={() => delRef(sel.code)}><Trash2 className="size-4" /></button>
@@ -1084,7 +1116,7 @@ function ReferencesPanel({ code, presets, meta, awardValidFrom, awardValidTo, on
                   ? t('awed.refValidHintAward', { from: openEnded(awardValidFrom), to: openEnded(awardValidTo) })
                   : t('awed.refValidHint')}
               </p>
-              <div className="flex justify-end pt-1"><Button size="sm" className="h-7" onClick={() => sel && saveRef(sel)}><Save className="size-3.5 mr-1" /> {t('awed.saveReference')}</Button></div>
+              <div className="flex justify-end pt-1"><Button size="sm" className="h-7" onClick={() => sel && saveSelected(sel)}><Save className="size-3.5 mr-1" /> {t('awed.saveReference')}</Button></div>
             </div>
           )}
         </div>

frontend/src/components/AwardsPanel.tsx

@@ -64,7 +64,9 @@ function ProgressBar({ worked, confirmed, total }: { worked: number; confirmed:
   );
 }
 
-type AwardListItem = { code: string; name: string; valid?: boolean; bands?: string[]; emission?: string[] };
+// scoped: the award is limited to one or more DXCC entities. Missing-reference
+// detection only means anything for those — see the Missing refs button.
+type AwardListItem = { code: string; name: string; valid?: boolean; bands?: string[]; emission?: string[]; scoped?: boolean };
 
 export function AwardsPanel({ onEditQSO, onAwardsChanged }: { onEditQSO?: (id: number) => void; onAwardsChanged?: () => void } = {}) {
   const { t } = useI18n();
@@ -150,7 +152,7 @@ export function AwardsPanel({ onEditQSO, onAwardsChanged }: { onEditQSO?: (id: n
       ]);
       const follow = new Set(tracked);
       let list: AwardListItem[] = defs
-        .map((d) => ({ code: d.code, name: d.name, valid: d.valid, bands: d.valid_bands ?? [], emission: d.emission ?? [] }))
+        .map((d) => ({ code: d.code, name: d.name, valid: d.valid, bands: d.valid_bands ?? [], emission: d.emission ?? [], scoped: (d.dxcc_filter ?? []).length > 0 }))
         .sort((a, b) => a.code.localeCompare(b.code));
       if (follow.size > 0) list = list.filter((a) => follow.has(a.code));
       setAwardList(list);
@@ -427,13 +429,21 @@ export function AwardsPanel({ onEditQSO, onAwardsChanged }: { onEditQSO?: (id: n
                 ))}
               </div>
               <span className="text-xs text-muted-foreground">{filteredRefs.length} {t('awp.refs')}</span>
-              <button
-                onClick={() => setShowMissing(true)}
-                className="flex items-center gap-1 text-xs text-warning-muted-foreground hover:text-warning border border-warning-border bg-warning-muted rounded px-2 py-1"
-                title={t('awp.missingRefsTitle')}
-              >
-                <AlertTriangle className="size-3" /> {t('awp.missingRefs')}
-              </button>
+              {/* Only for an award scoped to a DXCC entity. "In this award's
+                  scope but with no reference" needs a scope to be in: on a
+                  worldwide reference award — POTA, SOTA, IOTA, WWFF — every
+                  contact anywhere would qualify, so the answer is always none.
+                  The button used to be there regardless and could only ever
+                  open a window saying it had found nothing. */}
+              {awardList.find((a) => a.code === selected)?.scoped && (
+                <button
+                  onClick={() => setShowMissing(true)}
+                  className="flex items-center gap-1 text-xs text-warning-muted-foreground hover:text-warning border border-warning-border bg-warning-muted rounded px-2 py-1"
+                  title={t('awp.missingRefsTitle')}
+                >
+                  <AlertTriangle className="size-3" /> {t('awp.missingRefs')}
+                </button>
+              )}
               <div className="flex-1" />
               {/* Legend */}
               <div className="flex items-center gap-2 text-[10px] text-muted-foreground">
@@ -608,6 +618,10 @@ export function AwardsPanel({ onEditQSO, onAwardsChanged }: { onEditQSO?: (id: n
 // the missing reference (e.g. a department for DDFM).
 type MissingSortKey = 'qso_date' | 'callsign' | 'band' | 'mode' | 'country' | 'qth';
 
+// How many references the assign dropdown will put in the DOM at once. Above
+// this a search box appears beside it and the rest are held back, counted.
+const REF_MENU_MAX = 300;
+
 function MissingQSOModal({ code, name, onClose, onEditQSO }: { code: string; name: string; onClose: () => void; onEditQSO?: (id: number) => void }) {
   const { t } = useI18n();
   const [qsos, setQsos] = useState<any[]>([]);
@@ -616,6 +630,7 @@ function MissingQSOModal({ code, name, onClose, onEditQSO }: { code: string; nam
   const [sortKey, setSortKey] = useState<MissingSortKey>('callsign');
   const [sortDir, setSortDir] = useState<'asc' | 'desc'>('asc');
   const [refs, setRefs] = useState<Array<{ code: string; name: string }>>([]);
+  const [refSearch, setRefSearch] = useState('');
   const [assignRef, setAssignRef] = useState('');
   const [busy, setBusy] = useState(false);
   const [msg, setMsg] = useState('');
@@ -629,12 +644,38 @@ function MissingQSOModal({ code, name, onClose, onEditQSO }: { code: string; nam
       .finally(() => setLoading(false));
   };
   useEffect(() => { load(); }, [code]);
-  // The award's reference list drives the "assign" dropdown (e.g. China provinces).
+  // The award's reference list drives the "assign" dropdown (e.g. China
+  // provinces) — fetched ONLY once there is something to assign it to.
+  //
+  // Missing-reference detection needs a DXCC scope, so a worldwide award like
+  // POTA always has zero rows here and says so. It was still loading every
+  // reference behind that message: on a log with the full POTA park list
+  // imported, tens of thousands of them, feeding a dropdown that could not be
+  // used for anything. That is the window freezing on a screen with nothing to
+  // offer.
   useEffect(() => {
+    if (loading || qsos.length === 0) { setRefs([]); return; }
     ListAwardReferences(code)
       .then((r) => setRefs(((r ?? []) as any[]).map((x) => ({ code: String(x.code).toUpperCase(), name: String(x.name ?? '') }))))
       .catch(() => setRefs([]));
-  }, [code]);
+  }, [code, loading, qsos.length]);
+
+  // What the assign dropdown actually renders. Bounded, and it says how many it
+  // is holding back rather than silently showing the first few hundred.
+  const shownRefs = useMemo(() => {
+    const q = refSearch.trim().toUpperCase();
+    const rows = q
+      ? refs.filter((r) => r.code.includes(q) || r.name.toUpperCase().includes(q))
+      : refs;
+    return rows.slice(0, REF_MENU_MAX);
+  }, [refs, refSearch]);
+  const hiddenRefs = useMemo(() => {
+    const q = refSearch.trim().toUpperCase();
+    const total = q
+      ? refs.filter((r) => r.code.includes(q) || r.name.toUpperCase().includes(q)).length
+      : refs.length;
+    return Math.max(0, total - shownRefs.length);
+  }, [refs, refSearch, shownRefs.length]);
 
   const qthOf = (q: any) => String(q.qth || q.notes || '');
   const sorted = useMemo(() => {
@@ -711,14 +752,29 @@ function MissingQSOModal({ code, name, onClose, onEditQSO }: { code: string; nam
         {/* Bulk-assign toolbar */}
         <div className="flex items-center gap-2 px-4 py-2 border-b border-border/50 bg-muted/20">
           <span className="text-xs text-muted-foreground">{t('awp.selectedArrow', { n: sel.size })}</span>
+          {/* A search box beside the dropdown, and a bounded list inside it.
+              A reference list can hold tens of thousands of entries (POTA
+              parks, Russian districts); every one of them as a menu item is
+              hundreds of thousands of DOM nodes and a window that stops
+              answering. Nobody scrolls to K-4521 anyway — they type it. */}
+          {refs.length > REF_MENU_MAX && (
+            <Input className="h-7 w-40 text-xs font-mono" value={refSearch}
+              placeholder={t('awp.filterReferences')}
+              onChange={(e) => setRefSearch(e.target.value)} />
+          )}
           <Select value={assignRef} onValueChange={setAssignRef}>
             <SelectTrigger className="h-7 w-64 text-xs"><SelectValue placeholder={t('awp.chooseReference')} /></SelectTrigger>
             <SelectContent className="max-h-72">
-              {refs.map((r) => (
+              {shownRefs.map((r) => (
                 <SelectItem key={r.code} value={r.code}>
                   <span className="font-mono font-semibold">{r.code}</span>{r.name ? <span className="text-muted-foreground"> · {r.name}</span> : ''}
                 </SelectItem>
               ))}
+              {hiddenRefs > 0 && (
+                <div className="px-2 py-1.5 text-[11px] text-muted-foreground">
+                  {t('awp.refsNarrow', { n: hiddenRefs })}
+                </div>
+              )}
             </SelectContent>
           </Select>
           <Button size="sm" disabled={!assignRef || sel.size === 0 || busy} onClick={applyAssign}>

frontend/src/components/QSOEditModal.tsx

@@ -374,7 +374,10 @@ export function QSOEditModal({ qso, onSave, onDelete, onClose, countries = [], b
       // refreshes it. A cached answer from a thinner QRZ subscription (or any
       // stale row) otherwise stayed for its whole 30-day life and the button
       // appeared to do nothing.
-      const r: any = await LookupCallsignFresh(call);
+      // The QSO's OWN date: a ClubLog exception is resolved as of when the
+      // contact happened, not as of today. Re-looking-up a DXpedition contact
+      // months later must not move it to whatever the prefix means now.
+      const r: any = await LookupCallsignFresh(call, (dateOn || '').slice(0, 10));
       // The lookup WINS over what is in the record — that is the point of asking
       // for it. But an EMPTY result must never blank a good value: `??` only
       // guards against null, and Go marshals an unset string as "", so a QRZ

frontend/src/components/SettingsModal.tsx

@@ -51,6 +51,7 @@ import {
   GetUIPref, SetUIPref,
   GetFlexState, GetFlexBandAntennas, SaveFlexBandAntennas, GetFlexBandPower, SaveFlexBandPower,
   GetADIFMonitor, SaveADIFMonitor, PickADIFMonitorFile,
+  GetFolderSync, SaveFolderSync, PickFolderSyncFolder, GetFolderSyncStatus, SyncFolderNow,
   GetRelayAuto, SaveRelayAuto, GetStationDevices,
   GetAwardDefs, GetTrackedAwards, SaveTrackedAwards,
   GetBandOpenSettings, SaveBandOpenSettings, GetPSKReporterStatus, GetChaseNewGrids, SetChaseNewGrids, GetChaseNew, SetChaseNew, GetGridCacheStatus, GetLinkedAmps, SetLinkedAmps, GetSpotTTLMinutes, SetSpotTTLMinutes,
@@ -188,6 +189,7 @@ type SectionId =
   | 'external-services'
   | 'udp'
   | 'adifmon'
+  | 'foldersync'
   | 'webpublish'
   | 'lookup'
   | 'lists-bands'
@@ -242,6 +244,29 @@ function VendorMark({ vendor }: { vendor: 'o3a' }) {
 	);
 }
 
+// PanelHost renders the selected settings panel, and exists so a panel may hold
+// hooks of its own.
+//
+// The panels are nested inside SettingsModal, closing over its state — so they
+// cannot be rendered as <Panel />: a nested function is a NEW component type on
+// every parent render, which would unmount and remount the panel on each
+// keystroke. They were therefore CALLED, `PANELS[selected]()`, and a call runs
+// any hook inside them in SettingsModal's own hook list — conditionally, since
+// only the open section is called. React counts those, and the window stopped
+// drawing the moment such a section was opened (error #310, "rendered more
+// hooks than during the previous render"). It had happened once and was headed
+// off by a comment; the comment did not survive contact with the next panel.
+//
+// This host is module-scope, so its identity is stable and calling `render()`
+// inside it puts those hooks in a component context that persists. `key` is the
+// section, so switching sections REMOUNTS it — a fresh, consistent hook list per
+// section, and no panel state leaking into the next one. Within a section the
+// render prop is a new closure each parent render, which is how the panel keeps
+// seeing current values.
+function PanelHost({ render }: { render?: () => JSX.Element }) {
+  return render ? render() : null;
+}
+
 // buildTree returns the settings sidebar. The FlexRadio item only appears when
 // the active CAT backend is a Flex (per-band antenna config is Flex-specific).
 function buildTree(flexAvailable: boolean, t: (k: string) => string): TreeNode[] {
@@ -282,6 +307,7 @@ function buildTree(flexAvailable: boolean, t: (k: string) => string): TreeNode[]
       { kind: 'item', label: t('sec.cluster'),     id: 'cluster' },
       { kind: 'item', label: t('sec.udp'), id: 'udp' },
       { kind: 'item', label: t('sec.adifmon'), id: 'adifmon' },
+      { kind: 'item', label: t('sec.foldersync'), id: 'foldersync' },
       { kind: 'item', label: t('sec.webpublish'), id: 'webpublish' },
       { kind: 'item', label: t('sec.uscounties'), id: 'uscounties' },
       { kind: 'item', label: t('sec.database'), id: 'database' },
@@ -300,6 +326,7 @@ const SECTION_KEY: Partial<Record<SectionId, string>> = {
   'external-services': 'sec.external', appearance: 'sec.appearance', lookup: 'sec.lookup', 'lists-bands': 'sec.bands', 'lists-modes': 'sec.modes',
   cluster: 'sec.cluster', backup: 'sec.backup', database: 'sec.database', autostart: 'sec.autostart', udp: 'sec.udp',
   adifmon: 'sec.adifmon',
+  foldersync: 'sec.foldersync',
   webpublish: 'sec.webpublish',
   uscounties: 'sec.uscounties',
   awards: 'sec.awards', cat: 'sec.cat', rotator: 'sec.rotator', winkeyer: 'sec.winkeyer', antenna: 'sec.antenna',
@@ -682,6 +709,120 @@ function ADIFMonitorPanel() {
   );
 }
 
+// FolderSyncPanel: one operator, several PCs, one logbook through a folder they
+// already synchronise (Seafile, OneDrive, Dropbox, a NAS share).
+//
+// The status half is the point of the panel. Every part of this runs on someone
+// else's machine and on a sync client OpsLog cannot see, so "it is not working"
+// has to be answerable from here: which other PCs have written to the folder,
+// when each last did, and whether anything is sitting there unread.
+function FolderSyncPanel() {
+  const { t } = useI18n();
+  const [cfg, setCfg] = useState<{ enabled: boolean; folder: string; machine: string }>({ enabled: false, folder: '', machine: '' });
+  const [st, setSt] = useState<any>(null);
+  const [loaded, setLoaded] = useState(false);
+  const [err, setErr] = useState('');
+  const [msg, setMsg] = useState('');
+  const [busy, setBusy] = useState(false);
+  useEffect(() => {
+    GetFolderSync()
+      .then((c: any) => { if (c) setCfg({ enabled: !!c.enabled, folder: c.folder ?? '', machine: c.machine ?? '' }); })
+      .catch(() => {})
+      .finally(() => setLoaded(true));
+  }, []);
+  // Polled while the panel is open: a first setup is verified by watching the
+  // other PC appear in this list, and that happens on the sync client's clock,
+  // not on any action taken here.
+  useEffect(() => {
+    let alive = true;
+    const tick = () => GetFolderSyncStatus().then((s: any) => { if (alive) setSt(s); }).catch(() => {});
+    tick();
+    const h = window.setInterval(tick, 3000);
+    return () => { alive = false; window.clearInterval(h); };
+  }, []);
+  const save = async (next: typeof cfg) => {
+    setCfg(next);
+    setErr(''); setMsg('');
+    try {
+      await SaveFolderSync(next as any);
+      setMsg(t('sync.saved'));
+      GetFolderSyncStatus().then(setSt).catch(() => {});
+    } catch (e: any) {
+      setErr(String(e?.message ?? e));
+      // The switch goes back off rather than sitting on while nothing is
+      // written: a folder that refused the write test would lose every contact
+      // in silence, which is the one outcome this feature must never have.
+      setCfg({ ...next, enabled: false });
+    }
+  };
+  const pick = async () => {
+    try {
+      const p = await PickFolderSyncFolder();
+      if (p) save({ ...cfg, folder: p });
+    } catch { /* dialog cancelled */ }
+  };
+  const syncNow = async () => {
+    setBusy(true); setErr(''); setMsg('');
+    try {
+      const n = await SyncFolderNow();
+      setMsg(t('sync.applied').replace('{n}', String(n)));
+      GetFolderSyncStatus().then(setSt).catch(() => {});
+    } catch (e: any) {
+      setErr(String(e?.message ?? e));
+    } finally { setBusy(false); }
+  };
+  const when = (iso: string) => (iso ? new Date(iso).toLocaleString() : t('sync.never'));
+  return (
+    <div className="space-y-4 max-w-2xl">
+      <p className="text-xs text-muted-foreground">{t('sync.hint')}</p>
+      <label className="flex items-center gap-2 text-sm cursor-pointer">
+        <Checkbox checked={cfg.enabled} disabled={!loaded} onCheckedChange={(c) => save({ ...cfg, enabled: !!c })} />
+        {t('sync.enable')}
+      </label>
+
+      <div className="grid grid-cols-[130px_1fr] items-center gap-2">
+        <span className="text-sm">{t('sync.machine')}</span>
+        <Input value={cfg.machine} placeholder="shack" className="h-8"
+          onChange={(e) => setCfg({ ...cfg, machine: e.target.value })}
+          onBlur={() => save(cfg)} />
+        <span className="text-sm">{t('sync.folder')}</span>
+        <div className="flex items-center gap-2 min-w-0">
+          <span className="flex-1 font-mono text-xs truncate" title={cfg.folder}>{cfg.folder || '—'}</span>
+          <Button variant="outline" size="sm" onClick={pick}><FolderOpen className="size-3.5 mr-1" /> {t('sync.choose')}</Button>
+        </div>
+      </div>
+
+      {err && <p className="text-xs text-destructive">{err}</p>}
+      {msg && <p className="text-xs text-success">{msg}</p>}
+
+      <SectionHeader title={t('sync.state')} />
+      <div className="rounded-md border border-border bg-muted/20 p-3 space-y-2 text-xs">
+        <div className="flex flex-wrap gap-x-6 gap-y-1 text-muted-foreground">
+          <span>{t('sync.thisPc')}: <span className="font-mono text-foreground">{st?.machine_id || '—'}</span></span>
+          <span>{t('sync.lastSync')}: <span className="text-foreground">{when(st?.last_sync ?? '')}</span></span>
+          <span>{t('sync.sent')}: <span className="text-foreground">{st?.sent ?? 0}</span></span>
+          <span>{t('sync.received')}: <span className="text-foreground">{st?.received ?? 0}</span></span>
+        </div>
+        {st?.error && <p className="text-destructive">{st.error}</p>}
+        <div className="space-y-1">
+          {(st?.peers ?? []).length === 0 && <p className="italic text-muted-foreground">{t('sync.noPeers')}</p>}
+          {(st?.peers ?? []).map((p: any) => (
+            <div key={p.machine} className="flex items-center gap-2">
+              <span className="font-mono truncate">{p.machine}</span>
+              <span className="text-muted-foreground">{when(p.last_change)}</span>
+              {p.behind > 0 && <span className="text-warning">{t('sync.behind')}</span>}
+            </div>
+          ))}
+        </div>
+      </div>
+      <Button variant="outline" size="sm" onClick={syncNow} disabled={busy || !cfg.enabled}>
+        {busy ? <Loader2 className="size-3.5 animate-spin mr-1.5" /> : null}
+        {t('sync.now')}
+      </Button>
+    </div>
+  );
+}
+
 // AmpUI mirrors the backend AmpConfig — one configured amplifier.
 type AmpUI = { id: string; name: string; enabled: boolean; type: string; transport: string; host: string; port: number; com_port: string; baud: number;
   // Band-follow (ACOM): a SECOND serial port on which OpsLog answers the amp's
@@ -1262,7 +1403,7 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
     yaesu_port: '', yaesu_baud: 38400, yaesu_low_lines: false, kenwood_low_lines: false, kenwood_port: '', kenwood_baud: 9600, kenwood_host: '', kenwood_data_mode: 'usb', xiegu_port: '', xiegu_baud: 19200, xiegu_addr: 0x70, xiegu_ptt_line: '',
     icom_port: '', icom_baud: 115200, icom_addr: 0x98, icom_net_host: '', icom_net_user: '', icom_net_pass: '', icom_net_audio: false,
     tci_host: '', tci_port: 40001, tci_spots: false, poll_ms: 250, delay_ms: 0,
-    digital_default: 'FT8', share_enabled: false, share_port: 4532,
+    digital_default: 'FT8', share_enabled: false, share_port: 4532, share_proto: 'rigctl', share_tci_port: 40001,
     ptt_hotkey_enabled: false, ptt_hotkey: '', ptt_hotkey_toggle: false,
   });
   // While true, the next key press is captured as the PTT hotkey.
@@ -1534,8 +1675,10 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
   const [eqslTest, setEqslTest] = useState<{ ok: boolean; msg: string } | null>(null);
   const [eqslTesting, setEqslTesting] = useState(false);
   const [stationLocations, setStationLocations] = useState<string[]>([]);
-  // Active tab in the External Services panel — lifted here because
-  // PANELS[selected]() is called as a function, so panels can't hold hooks.
+  // Active tab in the External Services panel. Lifted here back when a panel
+  // could not hold hooks at all; PanelHost lifted that restriction, and this
+  // stays put because moving it down would reset the tab on every reopen —
+  // a choice now, not a workaround.
   const [extSvcTab, setExtSvcTab] = useState<'qrz' | 'clublog' | 'hrdlog' | 'eqsl' | 'lotw' | 'cloudlog' | 'pota'>('qrz');
   // POTA hunter-log sync (stamps pota_ref on local QSOs from your pota.app log).
   const [potaToken, setPotaToken] = useState('');
@@ -1632,9 +1775,9 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
     return () => { unsub?.(); };
   }, []);
   const [profiles, setProfiles] = useState<Profile[]>([]);
-  // State for ProfilesPanel — lifted here because PANELS[selected]() calls
-  // the panel as a plain function, not as a JSX element, so any useState
-  // inside the panel function would violate the Rules of Hooks.
+  // State for ProfilesPanel. Lifted here back when a panel could not hold hooks
+  // — PanelHost lifted that restriction — and left here because the selection
+  // then survives switching sections, which is what an operator expects of it.
   const [profileSelectedId, setProfileSelectedId] = useState<number>(0);
   const [profileNameDraft, setProfileNameDraft] = useState<string>('');
 
@@ -2225,6 +2368,11 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
   }
 
   function LookupPanel() {
+    // The cache lifetime as TYPED, so the box can be emptied and re-filled.
+    // Re-seeded when the settings arrive from the backend — which is after the
+    // first render, so it cannot simply be the initial value.
+    const [ttlText, setTtlText] = useState(String(lookup.cache_ttl_days));
+    useEffect(() => { setTtlText(String(lookup.cache_ttl_days)); }, [lookup.cache_ttl_days]);
     // Per-row provider editor — kept inline because it's only used twice
     // and needs closure access to the parent state.
     const row = (
@@ -2361,16 +2509,31 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
           <div className="flex gap-3 items-end">
             <div className="space-y-1 w-40">
               <Label>{t('lk.ttl')}</Label>
+              {/* Raw text, not the stored number. Deriving the value from the
+                  number on every keystroke made the box impossible to empty —
+                  and "0" itself unreachable, since parseInt('0') || 30 is 30.
+                  Zero is now a real setting, so it has to be typeable. */}
               <Input
-                type="number" min={1} max={3650}
-                value={lookup.cache_ttl_days}
-                onChange={(e) => setLookup((s) => ({ ...s, cache_ttl_days: parseInt(e.target.value) || 30 }))}
+                type="number" min={0} max={3650}
+                value={ttlText}
+                onChange={(e) => {
+                  const raw = e.target.value;
+                  setTtlText(raw);
+                  const n = parseInt(raw, 10);
+                  if (Number.isFinite(n) && n >= 0) {
+                    setLookup((s) => ({ ...s, cache_ttl_days: Math.min(n, 3650) }));
+                  }
+                }}
+                onBlur={() => setTtlText(String(lookup.cache_ttl_days))}
               />
             </div>
             <Button variant="outline" onClick={clearCache} disabled={clearing}>
               {clearing ? t('lk.clearing') : t('lk.clearCache')}
             </Button>
           </div>
+          {lookup.cache_ttl_days === 0 && (
+            <p className="text-[11px] text-warning mt-2">{t('lk.cacheOff')}</p>
+          )}
         </div>
       </>
     );
@@ -3006,15 +3169,47 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
             </label>
             <p className="text-[11px] text-muted-foreground">{t('cat.shareHint')}</p>
             {catCfg.share_enabled && (
-              <div className="space-y-1 max-w-[200px]">
-                <Label>{t('cat.sharePort')}</Label>
-                <PortInput
-                  value={catCfg.share_port || 4532}
-                  fallback={4532}
-                  onChange={(n) => setCatCfg((s) => ({ ...s, share_port: n }))}
-                />
+              <div className="flex flex-wrap items-end gap-4">
+                {/* One protocol or the other. They answer the same questions about
+                    the same radio, and no program speaks both — so this is a
+                    choice, not two switches. */}
+                <div className="space-y-1">
+                  <Label>{t('cat.shareProto')}</Label>
+                  <Select
+                    value={(catCfg as any).share_proto === 'tci' ? 'tci' : 'rigctl'}
+                    onValueChange={(v) => setCatCfg((s) => ({ ...s, share_proto: v } as any))}
+                  >
+                    <SelectTrigger className="h-8 w-[240px]"><SelectValue /></SelectTrigger>
+                    <SelectContent>
+                      <SelectItem value="rigctl">{t('cat.shareRigctl')}</SelectItem>
+                      <SelectItem value="tci">{t('cat.shareTci')}</SelectItem>
+                    </SelectContent>
+                  </Select>
+                </div>
+                <div className="space-y-1 max-w-[200px]">
+                  <Label>{t('cat.sharePort')}</Label>
+                  {(catCfg as any).share_proto === 'tci' ? (
+                    <PortInput
+                      value={(catCfg as any).share_tci_port || 40001}
+                      fallback={40001}
+                      onChange={(n) => setCatCfg((s) => ({ ...s, share_tci_port: n } as any))}
+                    />
+                  ) : (
+                    <PortInput
+                      value={catCfg.share_port || 4532}
+                      fallback={4532}
+                      onChange={(n) => setCatCfg((s) => ({ ...s, share_port: n }))}
+                    />
+                  )}
+                </div>
               </div>
             )}
+            {catCfg.share_enabled && (catCfg as any).share_proto === 'tci' && catCfg.backend === 'tci' && (
+              // Both ends TCI: ExpertSDR is almost certainly already holding
+              // 40001 on this machine, and our server would fail to bind. Worth
+              // saying here rather than leaving it in the log.
+              <p className="text-[11px] text-warning">{t('cat.shareTciClash')}</p>
+            )}
           </div>
           {/* PTT hotkey — a keyboard key that keys the rig while OpsLog is focused.
               Uses the Audio → PTT method (CAT / RTS / DTR), falling back to CAT. */}
@@ -3406,13 +3601,9 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
   // register — the output on/off. The voltage and current SET points are shown
   // because they are worth seeing, and are not editable here: they belong to the
   // supply's front panel, and a logbook that can set them can set them wrong.
-  // NO HOOKS IN HERE. This panel is called as a plain function, like its
-  // neighbours — PANELS[x]() — so a useState of its own lands in SettingsModal's
-  // hook list and only while this section is open. React counts those, and it
-  // stopped drawing the moment the section was clicked (error #310, "rendered
-  // more hooks than during the previous render"). The COM ports come from the
-  // list SettingsModal already loads for the Winkeyer panel: one machine, one
-  // set of serial ports, loaded once.
+  // The COM ports come from the list SettingsModal already loads for the
+  // Winkeyer panel — one machine, one set of serial ports, fetched once. A hook
+  // of its own would be legal now (see PanelHost) and would fetch them twice.
   function PSUPanelSettings() {
     const ports = wkPorts;
     const setPorts = setWkPorts;
@@ -4720,10 +4911,8 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
     }
     return (
       <>
-        <SectionHeader
-          title={t('sec.backup')}
-          hint={mysqlCfg.enabled ? t('bk.hintMysql') : t('bk.hint')}
-        />
+        {/* No hint: everyone knows what a backup is. */}
+        <SectionHeader title={t('sec.backup')} />
         <div className="space-y-5 max-w-2xl">
           <label className="flex items-center gap-2 text-sm cursor-pointer">
             <Checkbox
@@ -5641,7 +5830,9 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
             <Button variant="outline" size="sm" onClick={openExisting}><Database className="size-3.5" /> {t('db.openExisting')}</Button>
             <Button variant="outline" size="sm" onClick={saveCopy}><Copy className="size-3.5" /> {t('db.saveCopy')}</Button>
             <Button variant="outline" size="sm" onClick={renameDb} title={t('db.renameTip')}><Pencil className="size-3.5" /> {t('db.rename')}</Button>
-            <Button variant="outline" size="sm" onClick={revealFolder}><FolderOpen className="size-3.5" /> {t('db.openFolder')}</Button>
+            {/* Pushed right: these two act on the file that is already there,
+                while the four on the left change WHICH file is in use. */}
+            <Button variant="outline" size="sm" className="ml-auto" onClick={revealFolder}><FolderOpen className="size-3.5" /> {t('db.openFolder')}</Button>
             {dbSettings.is_custom && <Button variant="ghost" size="sm" onClick={resetDefault}>{t('db.resetDefault')}</Button>}
           </div>
           {/* The DB pointer is only read at startup, so offer the restart inline. */}
@@ -5674,7 +5865,6 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
             </SelectContent>
           </Select>
         </div>
-        <p className="text-[11px] text-muted-foreground max-w-2xl mb-3">{t('db.profileHint')}</p>
 
         {/* Compact active-backend confirmation / MySQL-fallback warning. */}
         {backendStatus && (
@@ -5715,7 +5905,6 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
         {/* MySQL: shared logbook connection (multi-operator) */}
         {mysqlCfg.enabled && (
         <div className="space-y-3 max-w-2xl">
-          <div className="text-[11px] text-muted-foreground leading-relaxed">{t('db.mysqlHint')}</div>
           <div className="grid grid-cols-[130px_1fr] gap-2 items-center">
             <Label className="text-sm">{t('db.host')}</Label>
             <Input className="h-8" placeholder="192.168.1.10 or db.example.com" value={mysqlCfg.host} onChange={(e) => setMysqlField({ host: e.target.value })} />
@@ -6206,12 +6395,8 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
                 }}
                 className="mt-0.5"
               />
-              <span>
-                {t('gen.mwShow')}
-                <span className="block text-xs text-muted-foreground mt-0.5">
-                  {t('gen.mwDesc')}
-                </span>
-              </span>
+              {/* No explanatory line: every operator knows what Most Wanted is. */}
+              <span>{t('gen.mwShow')}</span>
             </label>
             <div className="flex items-center gap-3 pl-6">
               <Button variant="outline" size="sm" className="h-8" disabled={mwBusy}
@@ -6407,10 +6592,11 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
     'lists-modes':  ModesPanel,
     cluster:        ClusterPanel,
     udp:            UDPIntegrationsPanelWrapper,
-    // Rendered as a real element (not called as a bare function) so its own hooks
-    // — useState/useEffect/useI18n — get a proper component context; PANELS[x]()
-    // is a plain call and hook-holding panels must go through JSX like this.
+    // Module-scope components, wrapped so their props can be passed. The nested
+    // panels below go through PanelHost instead — which is what now lets either
+    // kind hold hooks.
     adifmon:        () => <ADIFMonitorPanel />,
+    foldersync:     () => <FolderSyncPanel />,
     webpublish:     () => <WebPublishPanel />,
     relayauto:      () => <RelayAutoPanel />,
     backup:         BackupPanel,
@@ -6452,7 +6638,7 @@ export function SettingsModal({ onClose, onSaved, initialSection, onMainPaneChan
               <div className="text-[10px] uppercase tracking-wider text-muted-foreground mb-3 font-semibold">
                 {breadcrumb}
               </div>
-              {PANELS[selected]?.()}
+              <PanelHost key={selected} render={PANELS[selected]} />
 
               {err && (
                 <div className="mt-6 text-xs text-destructive bg-destructive/10 border border-destructive/30 rounded-md px-3 py-2 max-w-2xl">

frontend/src/components/StationControlPanel.tsx

@@ -3,6 +3,7 @@ import { Plus, Pencil, Trash2, Power, PlugZap, Loader2, Check, X, Compass, Squar
 import { Button } from '@/components/ui/button';
 import { Input } from '@/components/ui/input';
 import { Label } from '@/components/ui/label';
+import { Checkbox } from '@/components/ui/checkbox';
 import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from '@/components/ui/select';
 import { cn } from '@/lib/utils';
 import { useI18n } from '@/lib/i18n';
@@ -82,7 +83,7 @@ type Device = {
   id: string; type: string; name: string; host: string;
   user?: string; pass?: string; channels?: number; labels: string[];
   // Generic HTTP board only. The per-relay URLs win over the patterns.
-  on_urls?: string[]; off_urls?: string[]; on_pattern?: string; off_pattern?: string;
+  on_urls?: string[]; off_urls?: string[]; on_pattern?: string; off_pattern?: string; insecure_tls?: boolean;
 };
 type Relay = { number: number; label: string; on: boolean };
 type DevStatus = { id: string; name: string; type: string; connected: boolean; error?: string; relays: Relay[] };
@@ -787,6 +788,12 @@ function DeviceEditor({ device, onChange, onSave, onCancel, t }: {
     && [...(device.on_urls ?? []), ...(device.off_urls ?? []), device.on_pattern ?? '', device.off_pattern ?? '']
       .some((s) => (s ?? '').includes('{value}'))
     && device.labels.some((l) => !l.trim());
+  // Any https:// among this board's URLs. A relay box on the LAN signs its own
+  // certificate, so HTTPS to one cannot be verified — the operator has to say
+  // whether to accept that, and the question only arises once they type https.
+  const usesHTTPS = isHTTPGen
+    && [...(device.on_urls ?? []), ...(device.off_urls ?? []), device.on_pattern ?? '', device.off_pattern ?? '']
+      .some((u) => (u ?? '').trim().toLowerCase().startsWith('https://'));
   // COM ports for the generic USB-serial relay picker.
   const [serialPorts, setSerialPorts] = useState<string[]>([]);
   useEffect(() => {
@@ -959,6 +966,19 @@ function DeviceEditor({ device, onChange, onSave, onCancel, t }: {
             </div>
           </div>
           <div className="text-[10px] text-muted-foreground">{t('station.patternHint')}</div>
+          {/* Shown only once an https:// URL is actually in use. A board on
+              plain HTTP has no certificate to argue about, and an option that
+              cannot matter yet is one more thing to wonder about. */}
+          {usesHTTPS && (
+            <label className="flex items-start gap-2 text-xs cursor-pointer">
+              <Checkbox className="mt-0.5" checked={!!device.insecure_tls}
+                onCheckedChange={(c) => onChange({ ...device, insecure_tls: !!c })} />
+              <span>
+                {t('station.insecureTls')}
+                <span className="block text-[10px] text-muted-foreground">{t('station.insecureTlsHint')}</span>
+              </span>
+            </label>
+          )}
           <div className="space-y-1">
             <Label>{t('station.perRelayUrls')}</Label>
             <div className="space-y-1">

frontend/src/lib/awardRefs.ts

@@ -192,3 +192,27 @@ export function spotRefList(byCode: Record<string, string>, fieldOf: Record<stri
   }
   return out.sort((a, b) => a.length - b.length || a.localeCompare(b));
 }
+
+// withIOTARef adds an island reference from the callbook to an entry's award
+// references, unless the operator has already set one.
+//
+// QRZ sends <iota> for an operator on an island and OpsLog used to read past it.
+// It matters more here than it would for another award: unlike POTA there is no
+// live "who is on an island right now" feed anywhere, so the callbook record is
+// the practical source — and it is known before the contact is logged, which is
+// when it is useful.
+//
+// A reference the operator typed or picked wins. A callbook entry can be years
+// old, and the operator in front of the radio has just been told where the
+// station is.
+export function withIOTARef(awardRefs: string, iota: string): string {
+  const ref = (iota || '').trim().toUpperCase();
+  // EU-048: two letters, a hyphen, three digits. Anything else is not an IOTA
+  // reference, and writing it into the award would make a reference that no
+  // list contains — which counts for nothing and has to be found by hand later.
+  if (!/^[A-Z]{2}-\d{3}$/.test(ref)) return awardRefs;
+  const byCode = parseAwardRefs(awardRefs);
+  if ((byCode['IOTA'] ?? '').trim() !== '') return awardRefs; // already set: leave it
+  const sep = awardRefs.trim() === '' ? '' : ';';
+  return awardRefs + sep + 'IOTA@' + ref;
+}

frontend/src/lib/rotorHeading.ts

@@ -36,6 +36,14 @@ let timer: number | undefined;
 let inFlight = false;
 let lastAz: number | null = null;
 let lastMoveAt = 0;
+// The last heading anyone received, replayed to whoever subscribes next.
+//
+// Without it, opening Station Control left the compass blank for one or two
+// seconds while every other panel filled at once. Nothing was slow: the status
+// bar already had the loop running with a tick pending, so a component mounting
+// halfway through an idle interval simply waited out the rest of it. The
+// heading was known the whole time — it just had nowhere to be read from.
+let last: RotorHeading | null = null;
 
 function schedule(delay: number) {
   if (timer !== undefined) window.clearTimeout(timer);
@@ -54,6 +62,7 @@ async function tick() {
       if (lastAz !== null && h.azimuth !== lastAz) lastMoveAt = Date.now();
       lastAz = h.azimuth;
     }
+    last = h;
     subs.forEach((fn) => { try { fn(h); } catch { /* a subscriber must not stop the loop */ } });
   } catch {
     // Leave the last heading alone: a single failed poll on a shared serial port
@@ -69,6 +78,12 @@ async function tick() {
 // unsubscribe. The loop stops when the last subscriber leaves.
 export function subscribeRotorHeading(fn: (h: RotorHeading) => void): () => void {
   subs.add(fn);
+  // Hand over what is already known, at once. An Alpha SPID poll is an open,
+  // a read at 600 baud and a close, so even an immediate one takes a moment —
+  // the cached heading is what makes the compass appear with the panel rather
+  // than after it. In a microtask, so a subscriber is never called back before
+  // subscribeRotorHeading has returned to it.
+  if (last) { const h = last; queueMicrotask(() => { if (subs.has(fn)) fn(h); }); }
   if (timer === undefined && !inFlight) void tick();
   return () => {
     subs.delete(fn);

frontend/src/version.ts

@@ -1,6 +1,6 @@
 // Single source of truth for the app version shown in the UI (header + About).
 // Bump this on a release (the release script updates it alongside telemetry.go).
-export const APP_VERSION = '0.25.6';
+export const APP_VERSION = '0.25.8';
 
 // Author / credits, shown in Help -> About.
 export const APP_AUTHOR = 'F4BPO';

frontend/wailsjs/go/main/App.d.ts

@@ -451,6 +451,10 @@ export function GetFlexBandPower():Promise<Record<string, main.FlexBandPower>>;
 
 export function GetFlexState():Promise<cat.FlexTXState>;
 
+export function GetFolderSync():Promise<main.FolderSyncConfig>;
+
+export function GetFolderSyncStatus():Promise<main.FolderSyncStatus>;
+
 export function GetGridCacheStatus():Promise<main.GridCacheStatus>;
 
 export function GetIcomState():Promise<cat.IcomTXState>;
@@ -703,9 +707,9 @@ export function LogUDPLoggedADIF(arg1:string):Promise<number>;
 
 export function LogUIError(arg1:string,arg2:string,arg3:string):Promise<void>;
 
-export function LookupCallsign(arg1:string):Promise<lookup.Result>;
+export function LookupCallsign(arg1:string,arg2:string):Promise<lookup.Result>;
 
-export function LookupCallsignFresh(arg1:string):Promise<lookup.Result>;
+export function LookupCallsignFresh(arg1:string,arg2:string):Promise<lookup.Result>;
 
 export function MotorNudgeKHz(arg1:number):Promise<void>;
 
@@ -775,6 +779,8 @@ export function PickAudioFolder():Promise<string>;
 
 export function PickBackupFolder():Promise<string>;
 
+export function PickFolderSyncFolder():Promise<string>;
+
 export function PickOpenDatabase():Promise<string>;
 
 export function PickSaveDatabase():Promise<string>;
@@ -855,6 +861,8 @@ export function ReloadUDPIntegrations():Promise<Array<string>>;
 
 export function RemovePassphrase(arg1:string):Promise<void>;
 
+export function RenameAwardReference(arg1:string,arg2:string,arg3:string):Promise<void>;
+
 export function RenameDatabase(arg1:string):Promise<void>;
 
 export function RenameLogbook(arg1:string):Promise<void>;
@@ -935,6 +943,8 @@ export function SaveFlexBandAntennas(arg1:Record<string, main.FlexBandAnt>):Prom
 
 export function SaveFlexBandPower(arg1:Record<string, main.FlexBandPower>):Promise<void>;
 
+export function SaveFolderSync(arg1:main.FolderSyncConfig):Promise<void>;
+
 export function SaveListsSettings(arg1:main.ListsSettings):Promise<void>;
 
 export function SaveLookupSettings(arg1:main.LookupSettings):Promise<void>;
@@ -1095,6 +1105,8 @@ export function StopCWDecoder():Promise<void>;
 
 export function SwitchCATRig(arg1:number):Promise<void>;
 
+export function SyncFolderNow():Promise<number>;
+
 export function SyncPOTAHunterLog(arg1:boolean,arg2:boolean):Promise<main.POTASyncResult>;
 
 export function TailLogFile(arg1:number):Promise<string>;

frontend/wailsjs/go/main/App.js

@@ -842,6 +842,14 @@ export function GetFlexState() {
   return window['go']['main']['App']['GetFlexState']();
 }
 
+export function GetFolderSync() {
+  return window['go']['main']['App']['GetFolderSync']();
+}
+
+export function GetFolderSyncStatus() {
+  return window['go']['main']['App']['GetFolderSyncStatus']();
+}
+
 export function GetGridCacheStatus() {
   return window['go']['main']['App']['GetGridCacheStatus']();
 }
@@ -1346,12 +1354,12 @@ export function LogUIError(arg1, arg2, arg3) {
   return window['go']['main']['App']['LogUIError'](arg1, arg2, arg3);
 }
 
-export function LookupCallsign(arg1) {
-  return window['go']['main']['App']['LookupCallsign'](arg1);
+export function LookupCallsign(arg1, arg2) {
+  return window['go']['main']['App']['LookupCallsign'](arg1, arg2);
 }
 
-export function LookupCallsignFresh(arg1) {
-  return window['go']['main']['App']['LookupCallsignFresh'](arg1);
+export function LookupCallsignFresh(arg1, arg2) {
+  return window['go']['main']['App']['LookupCallsignFresh'](arg1, arg2);
 }
 
 export function MotorNudgeKHz(arg1) {
@@ -1490,6 +1498,10 @@ export function PickBackupFolder() {
   return window['go']['main']['App']['PickBackupFolder']();
 }
 
+export function PickFolderSyncFolder() {
+  return window['go']['main']['App']['PickFolderSyncFolder']();
+}
+
 export function PickOpenDatabase() {
   return window['go']['main']['App']['PickOpenDatabase']();
 }
@@ -1650,6 +1662,10 @@ export function RemovePassphrase(arg1) {
   return window['go']['main']['App']['RemovePassphrase'](arg1);
 }
 
+export function RenameAwardReference(arg1, arg2, arg3) {
+  return window['go']['main']['App']['RenameAwardReference'](arg1, arg2, arg3);
+}
+
 export function RenameDatabase(arg1) {
   return window['go']['main']['App']['RenameDatabase'](arg1);
 }
@@ -1810,6 +1826,10 @@ export function SaveFlexBandPower(arg1) {
   return window['go']['main']['App']['SaveFlexBandPower'](arg1);
 }
 
+export function SaveFolderSync(arg1) {
+  return window['go']['main']['App']['SaveFolderSync'](arg1);
+}
+
 export function SaveListsSettings(arg1) {
   return window['go']['main']['App']['SaveListsSettings'](arg1);
 }
@@ -2130,6 +2150,10 @@ export function SwitchCATRig(arg1) {
   return window['go']['main']['App']['SwitchCATRig'](arg1);
 }
 
+export function SyncFolderNow() {
+  return window['go']['main']['App']['SyncFolderNow']();
+}
+
 export function SyncPOTAHunterLog(arg1, arg2) {
   return window['go']['main']['App']['SyncPOTAHunterLog'](arg1, arg2);
 }

frontend/wailsjs/go/models.ts

@@ -1444,6 +1444,7 @@ export namespace lookup {
 	    email?: string;
 	    qsl_via?: string;
 	    web?: string;
+	    iota?: string;
 	    zip?: string;
 	    image_url?: string;
 	    source: string;
@@ -1473,6 +1474,7 @@ export namespace lookup {
 	        this.email = source["email"];
 	        this.qsl_via = source["qsl_via"];
 	        this.web = source["web"];
+	        this.iota = source["iota"];
 	        this.zip = source["zip"];
 	        this.image_url = source["image_url"];
 	        this.source = source["source"];
@@ -2048,6 +2050,8 @@ export namespace main {
 	    digital_default: string;
 	    share_enabled: boolean;
 	    share_port: number;
+	    share_proto: string;
+	    share_tci_port: number;
 	    ptt_hotkey_enabled: boolean;
 	    ptt_hotkey: string;
 	    ptt_hotkey_toggle: boolean;
@@ -2094,6 +2098,8 @@ export namespace main {
 	        this.digital_default = source["digital_default"];
 	        this.share_enabled = source["share_enabled"];
 	        this.share_port = source["share_port"];
+	        this.share_proto = source["share_proto"];
+	        this.share_tci_port = source["share_tci_port"];
 	        this.ptt_hotkey_enabled = source["ptt_hotkey_enabled"];
 	        this.ptt_hotkey = source["ptt_hotkey"];
 	        this.ptt_hotkey_toggle = source["ptt_hotkey_toggle"];
@@ -2463,6 +2469,82 @@ export namespace main {
 	        this.body = source["body"];
 	    }
 	}
+	export class FolderSyncConfig {
+	    enabled: boolean;
+	    folder: string;
+	    machine: string;
+	
+	    static createFrom(source: any = {}) {
+	        return new FolderSyncConfig(source);
+	    }
+	
+	    constructor(source: any = {}) {
+	        if ('string' === typeof source) source = JSON.parse(source);
+	        this.enabled = source["enabled"];
+	        this.folder = source["folder"];
+	        this.machine = source["machine"];
+	    }
+	}
+	export class FolderSyncPeer {
+	    machine: string;
+	    last_change: string;
+	    behind: number;
+	
+	    static createFrom(source: any = {}) {
+	        return new FolderSyncPeer(source);
+	    }
+	
+	    constructor(source: any = {}) {
+	        if ('string' === typeof source) source = JSON.parse(source);
+	        this.machine = source["machine"];
+	        this.last_change = source["last_change"];
+	        this.behind = source["behind"];
+	    }
+	}
+	export class FolderSyncStatus {
+	    enabled: boolean;
+	    folder: string;
+	    machine_id: string;
+	    peers: FolderSyncPeer[];
+	    last_sync: string;
+	    sent: number;
+	    received: number;
+	    error: string;
+	
+	    static createFrom(source: any = {}) {
+	        return new FolderSyncStatus(source);
+	    }
+	
+	    constructor(source: any = {}) {
+	        if ('string' === typeof source) source = JSON.parse(source);
+	        this.enabled = source["enabled"];
+	        this.folder = source["folder"];
+	        this.machine_id = source["machine_id"];
+	        this.peers = this.convertValues(source["peers"], FolderSyncPeer);
+	        this.last_sync = source["last_sync"];
+	        this.sent = source["sent"];
+	        this.received = source["received"];
+	        this.error = source["error"];
+	    }
+	
+		convertValues(a: any, classs: any, asMap: boolean = false): any {
+		    if (!a) {
+		        return a;
+		    }
+		    if (a.slice && a.map) {
+		        return (a as any[]).map(elem => this.convertValues(elem, classs));
+		    } else if ("object" === typeof a) {
+		        if (asMap) {
+		            for (const key of Object.keys(a)) {
+		                a[key] = new classs(a[key]);
+		            }
+		            return a;
+		        }
+		        return new classs(a);
+		    }
+		    return a;
+		}
+	}
 	export class GridCacheStatus {
 	    enabled: boolean;
 	    known: number;
@@ -3297,6 +3379,7 @@ export namespace main {
 	    off_urls?: string[];
 	    on_pattern?: string;
 	    off_pattern?: string;
+	    insecure_tls?: boolean;
 	
 	    static createFrom(source: any = {}) {
 	        return new StationDevice(source);
@@ -3316,6 +3399,7 @@ export namespace main {
 	        this.off_urls = source["off_urls"];
 	        this.on_pattern = source["on_pattern"];
 	        this.off_pattern = source["off_pattern"];
+	        this.insecure_tls = source["insecure_tls"];
 	    }
 	}
 	export class StationRelay {
@@ -4546,6 +4630,7 @@ export namespace qso {
 	    my_vucc_grids?: string;
 	    extras?: Record<string, string>;
 	    award_refs?: string;
+	    sync_uid?: string;
 	    // Go type: time
 	    created_at: any;
 	    // Go type: time
@@ -4687,6 +4772,7 @@ export namespace qso {
 	        this.my_vucc_grids = source["my_vucc_grids"];
 	        this.extras = source["extras"];
 	        this.award_refs = source["award_refs"];
+	        this.sync_uid = source["sync_uid"];
 	        this.created_at = this.convertValues(source["created_at"], null);
 	        this.updated_at = this.convertValues(source["updated_at"], null);
 	    }

i18n_test.go

@@ -0,0 +1,59 @@
+package main
+
+import (
+	"os"
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// Every user-visible string ships in BOTH languages. That is a project rule,
+// and until now it was only a rule: seventeen keys had drifted into English
+// only, among them the whole update panel and two DX-cluster settings, found by
+// a French operator photographing his own screen.
+//
+// A key present in one dictionary and not the other falls back to the key
+// itself, so the interface shows "clu.spotTtl" where a label belongs — or, as
+// here, the English text, which reads as deliberate and is not.
+func TestEveryStringIsInBothLanguages(t *testing.T) {
+	src, err := os.ReadFile("frontend/src/lib/i18n.tsx")
+	if err != nil {
+		t.Fatalf("read i18n.tsx: %v", err)
+	}
+	en := dictKeys(t, string(src), "const en: Dict = {")
+	fr := dictKeys(t, string(src), "const fr: Dict = {")
+
+	if len(en) == 0 || len(fr) == 0 {
+		t.Fatal("one of the dictionaries came back empty — this test has stopped checking anything")
+	}
+	for k := range en {
+		if !fr[k] {
+			t.Errorf("%q has no French — a French operator sees the English string, or the key itself", k)
+		}
+	}
+	for k := range fr {
+		if !en[k] {
+			t.Errorf("%q exists only in French — an English operator sees the key", k)
+		}
+	}
+}
+
+// dictKeys collects the keys of one dictionary literal: from its opening line
+// to the closing brace in column 0.
+func dictKeys(t *testing.T, src, opening string) map[string]bool {
+	t.Helper()
+	i := strings.Index(src, opening)
+	if i < 0 {
+		t.Fatalf("%q not found — the dictionaries have been renamed and this test needs updating", opening)
+	}
+	body := src[i+len(opening):]
+	if j := strings.Index(body, "\n};"); j >= 0 {
+		body = body[:j]
+	}
+	keyRe := regexp.MustCompile(`'([a-zA-Z0-9_.]+)'\s*:`)
+	out := map[string]bool{}
+	for _, m := range keyRe.FindAllStringSubmatch(body, -1) {
+		out[m[1]] = true
+	}
+	return out
+}

internal/award/catalog/waja.json

@@ -46,11 +46,12 @@
           "qsl"
         ],
         "total": 0,
-        "builtin": true
+        "builtin": true,
+        "version": 2
       },
       "references": [
         {
-          "code": "1",
+          "code": "01",
           "name": "Hokkaido",
           "dxcc": 0,
           "group": "",
@@ -58,16 +59,81 @@
           "valid": true
         },
         {
-          "code": "10",
-          "name": "Gunma",
+          "code": "02",
+          "name": "Aomori",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
           "valid": true
         },
+        {
+          "code": "03",
+          "name": "Iwate",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "04",
+          "name": "Akita",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "05",
+          "name": "Yamagata",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "06",
+          "name": "Miyagi",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "07",
+          "name": "Fukushima",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "08",
+          "name": "Niigata",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "09",
+          "name": "Nagano",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "10",
+          "name": "Tokyo",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "pattern": "\\bTok[iy]o\\b",
+          "valid": true
+        },
         {
           "code": "11",
-          "name": "Saitama",
+          "name": "Kanagawa",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -83,16 +149,15 @@
         },
         {
           "code": "13",
-          "name": "Tokyo",
+          "name": "Saitama",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
-          "pattern": "\\bTok[iy]o\\b",
           "valid": true
         },
         {
           "code": "14",
-          "name": "Kanagawa",
+          "name": "Ibaraki",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -100,7 +165,7 @@
         },
         {
           "code": "15",
-          "name": "Niigata",
+          "name": "Tochigi",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -108,7 +173,7 @@
         },
         {
           "code": "16",
-          "name": "Toyama",
+          "name": "Gunma",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -116,22 +181,6 @@
         },
         {
           "code": "17",
-          "name": "Ishikawa",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "18",
-          "name": "Fukui",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "19",
           "name": "Yamanashi",
           "dxcc": 0,
           "group": "",
@@ -139,31 +188,7 @@
           "valid": true
         },
         {
-          "code": "2",
-          "name": "Aomori",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "20",
-          "name": "Nagano",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "21",
-          "name": "Gifu",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "22",
+          "code": "18",
           "name": "Shizuoka",
           "dxcc": 0,
           "group": "",
@@ -171,7 +196,15 @@
           "valid": true
         },
         {
-          "code": "23",
+          "code": "19",
+          "name": "Gifu",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "20",
           "name": "Aichi",
           "dxcc": 0,
           "group": "",
@@ -179,7 +212,7 @@
           "valid": true
         },
         {
-          "code": "24",
+          "code": "21",
           "name": "Mie",
           "dxcc": 0,
           "group": "",
@@ -187,15 +220,7 @@
           "valid": true
         },
         {
-          "code": "25",
-          "name": "Shiga",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "26",
+          "code": "22",
           "name": "Kyoto",
           "dxcc": 0,
           "group": "",
@@ -203,23 +228,15 @@
           "valid": true
         },
         {
-          "code": "27",
-          "name": "Osaka",
+          "code": "23",
+          "name": "Shiga",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
           "valid": true
         },
         {
-          "code": "28",
-          "name": "Hyogo",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "29",
+          "code": "24",
           "name": "Nara",
           "dxcc": 0,
           "group": "",
@@ -227,24 +244,56 @@
           "valid": true
         },
         {
-          "code": "3",
-          "name": "Iwate",
+          "code": "25",
+          "name": "Osaka",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
           "valid": true
         },
         {
-          "code": "30",
+          "code": "26",
           "name": "Wakayama",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
           "valid": true
         },
+        {
+          "code": "27",
+          "name": "Hyogo",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "28",
+          "name": "Toyama",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "29",
+          "name": "Fukui",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "30",
+          "name": "Ishikawa",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
         {
           "code": "31",
-          "name": "Tottori",
+          "name": "Okayama",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -260,31 +309,31 @@
         },
         {
           "code": "33",
-          "name": "Okayama",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "34",
-          "name": "Hiroshima",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "35",
           "name": "Yamaguchi",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
           "valid": true
         },
+        {
+          "code": "34",
+          "name": "Tottori",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
+        {
+          "code": "35",
+          "name": "Hiroshima",
+          "dxcc": 0,
+          "group": "",
+          "subgrp": "",
+          "valid": true
+        },
         {
           "code": "36",
-          "name": "Tokushima",
+          "name": "Kagawa",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -292,7 +341,7 @@
         },
         {
           "code": "37",
-          "name": "Kagawa",
+          "name": "Tokushima",
           "dxcc": 0,
           "group": "",
           "subgrp": "",
@@ -314,14 +363,6 @@
           "subgrp": "",
           "valid": true
         },
-        {
-          "code": "4",
-          "name": "Miyagi",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
         {
           "code": "40",
           "name": "Fukuoka",
@@ -385,48 +426,8 @@
           "group": "",
           "subgrp": "",
           "valid": true
-        },
-        {
-          "code": "5",
-          "name": "Akita",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "6",
-          "name": "Yamagata",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "7",
-          "name": "Fukushima",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "8",
-          "name": "Ibaraki",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
-        },
-        {
-          "code": "9",
-          "name": "Tochigi",
-          "dxcc": 0,
-          "group": "",
-          "subgrp": "",
-          "valid": true
         }
       ]
     }
   ]
-}
+}

internal/award/waja_test.go

@@ -0,0 +1,101 @@
+package award
+
+import (
+	"encoding/json"
+	"testing"
+)
+
+// WAJA is numbered by the JARL, and the numbering is NOT Japan's ordinary
+// prefecture code.
+//
+// The catalog shipped with the government's JIS numbering instead — 01
+// Hokkaido, 02 Aomori, 03 Iwate, 04 Miyagi… — which agrees with the JARL's for
+// the first three prefectures and then diverges for thirty-five of the
+// remaining forty-four. The names were right throughout, so the award still
+// counted the right contacts; every reference simply carried the wrong number,
+// which is what an operator sends to the JARL when they claim it.
+//
+// The two schemes agree often enough to look correct at a glance, so this pins
+// the places they differ rather than a count. Each pair below is one the old
+// list got wrong, and the comment is what the old list said.
+func TestCatalogWAJAUsesTheJARLNumbering(t *testing.T) {
+	raw, ok := CatalogRefs("WAJA")
+	if !ok {
+		t.Fatal("WAJA has no reference list in the embedded catalog")
+	}
+	var refs []struct {
+		Code    string `json:"code"`
+		Name    string `json:"name"`
+		Pattern string `json:"pattern"`
+	}
+	if err := json.Unmarshal(raw, &refs); err != nil {
+		t.Fatalf("WAJA references: %v", err)
+	}
+	if len(refs) != 47 {
+		t.Fatalf("WAJA has %d prefectures, want exactly 47", len(refs))
+	}
+
+	byName := map[string]string{}
+	byCode := map[string]string{}
+	for _, r := range refs {
+		byName[r.Name] = r.Code
+		if prev, dup := byCode[r.Code]; dup {
+			t.Errorf("number %s is on both %s and %s", r.Code, prev, r.Name)
+		}
+		byCode[r.Code] = r.Name
+	}
+
+	for _, c := range []struct{ name, code string }{
+		{"Hokkaido", "01"},  // the one both schemes agree on, and the anchor
+		{"Miyagi", "06"},    // was 04
+		{"Akita", "04"},     // was 05
+		{"Niigata", "08"},   // was 15 — the JIS number
+		{"Nagano", "09"},    // was 20
+		{"Tokyo", "10"},     // was 13, the JIS number everyone recognises
+		{"Kanagawa", "11"},  // was 14
+		{"Saitama", "13"},   // was 11
+		{"Ibaraki", "14"},   // was 8
+		{"Gunma", "16"},     // was 10
+		{"Yamanashi", "17"}, // was 19
+		{"Kyoto", "22"},     // was 26
+		{"Osaka", "25"},     // was 27
+		{"Toyama", "28"},    // was 16
+		{"Ishikawa", "30"},  // was 17
+		{"Okayama", "31"},   // was 33
+		{"Tottori", "34"},   // was 31
+		{"Kagawa", "36"},    // was 37
+		{"Tokushima", "37"}, // was 36
+		{"Okinawa", "47"},   // unchanged: the far end of the list was already right
+	} {
+		if got := byName[c.name]; got != c.code {
+			t.Errorf("%s is numbered %q, want %q on the JARL list", c.name, got, c.code)
+		}
+	}
+
+	// Two digits throughout, as the JARL prints them. Not cosmetic: the codes
+	// are strings, so "1" sorts between "09" and "10" and the panel showed the
+	// prefectures in an order no list anywhere uses.
+	for _, r := range refs {
+		if len(r.Code) != 2 {
+			t.Errorf("%s is numbered %q — the JARL list is two digits throughout", r.Name, r.Code)
+		}
+	}
+
+	// The Tokyo spelling rule has to sit on Tokyo, and Tokyo moved. Left behind
+	// on the old number it would be matching QTHs for Saitama.
+	for _, r := range refs {
+		if r.Pattern == "" {
+			continue
+		}
+		if r.Name != "Tokyo" {
+			t.Errorf("%s (%s) carries the pattern %q, which belongs to Tokyo", r.Name, r.Code, r.Pattern)
+		}
+	}
+	if byName["Tokyo"] != "" {
+		for _, r := range refs {
+			if r.Name == "Tokyo" && r.Pattern == "" {
+				t.Error("Tokyo lost its spelling pattern in the renumbering — Tokio would stop counting")
+			}
+		}
+	}
+}

internal/awardref/awardref.go

@@ -266,6 +266,48 @@ func (r *Repo) Upsert(ctx context.Context, awardCode string, ref Ref) error {
 	return err
 }
 
+// Rename changes a reference's CODE, keeping everything else about it.
+//
+// Wanted because a shipped list can simply be wrong: WAJA went out numbered by
+// the Japanese state instead of by the JARL, and the only way to correct it was
+// to delete all 47 references and import a new list — losing anything the
+// operator had adjusted. The number is the one field an editor could not touch.
+//
+// A rename, not a delete plus an insert: everything the reference carries — its
+// pattern, its DXCC list, its validity window — travels with it, which is the
+// whole point of correcting a number rather than replacing an entry.
+func (r *Repo) Rename(ctx context.Context, awardCode, oldCode, newCode string) error {
+	ac := strings.ToUpper(strings.TrimSpace(awardCode))
+	from := strings.ToUpper(strings.TrimSpace(oldCode))
+	to := strings.ToUpper(strings.TrimSpace(newCode))
+	if ac == "" || from == "" || to == "" {
+		return fmt.Errorf("empty award or reference code")
+	}
+	if from == to {
+		return nil
+	}
+	// A collision would REPLACE the other reference and take its name, pattern
+	// and dates with it — one silently swallowing another, discovered much later
+	// as a reference that has quietly gone missing.
+	var n int
+	if err := r.db.QueryRowContext(ctx,
+		`SELECT COUNT(*) FROM award_references WHERE award_code = ? AND ref_code = ?`, ac, to).Scan(&n); err != nil {
+		return err
+	}
+	if n > 0 {
+		return fmt.Errorf("%s already has a reference %s", ac, to)
+	}
+	res, err := r.db.ExecContext(ctx,
+		`UPDATE award_references SET ref_code = ? WHERE award_code = ? AND ref_code = ?`, to, ac, from)
+	if err != nil {
+		return err
+	}
+	if rows, _ := res.RowsAffected(); rows == 0 {
+		return fmt.Errorf("%s has no reference %s", ac, from)
+	}
+	return nil
+}
+
 // Delete removes one reference from an award.
 func (r *Repo) Delete(ctx context.Context, awardCode, refCode string) error {
 	_, err := r.db.ExecContext(ctx,

internal/awardref/dxcciota.go

@@ -0,0 +1,129 @@
+package awardref
+
+// One-IOTA DXCC entities: a table of the entities that ARE a single IOTA group.
+//
+// WHY IT EXISTS. QRZ.com carries <iota> for an operator who fills it in, and
+// most do not. But for a great many entities the island reference follows from
+// the entity alone — a station in Ascension Island is on AF-003, there is
+// nothing else it could be — and the entity is known for every callsign, from
+// cty.dat, without any callbook at all. So the reference can be filled for an
+// operator with no QRZ subscription, on a station that has never touched a
+// callbook, before the contact is logged.
+//
+// Only entities that map to EXACTLY ONE reference are here. France is not: a
+// French station is usually on the mainland and on no island at all, and
+// guessing would put a reference on hundreds of contacts that earn none.
+//
+// SOURCE: dxcc_matches_one_iota.json from the IOTA programme
+// (www.iota-world.org/islands-on-the-air/downloads/), fetched 2026-08-17,
+// 99 entities. It changes only when an entity appears or IOTA re-maps one, so
+// it is a table here rather than a download: it then works offline, which is
+// where a portable station usually is. To refresh, fetch that file again and
+// re-emit this map, sorted by entity number.
+//
+// Entity names are comments only, joined from internal/dxcc for readability.
+var iotaByDXCC = map[int]string{
+	5:   "EU-002", // Aland Islands
+	10:  "AF-002", // Amsterdam & St. Paul Is.
+	12:  "NA-022", // Anguilla
+	17:  "NA-020", // Aves Island
+	20:  "OC-089", // Baker & Howland Islands
+	21:  "EU-004", // Balearic Islands
+	24:  "AN-002", // Bouvet
+	29:  "AF-004", // Canary Islands
+	34:  "OC-038", // Chatham Islands
+	35:  "OC-002", // Christmas Island
+	36:  "NA-011", // Clipperton Island
+	37:  "NA-012", // Cocos Island
+	38:  "OC-003", // Cocos (Keeling) Islands
+	41:  "AF-008", // Crozet Island
+	43:  "NA-095", // Desecheo Island
+	45:  "EU-001", // Dodecanese
+	62:  "NA-021", // Barbados
+	64:  "NA-005", // Bermuda
+	65:  "NA-023", // British Virgin Islands
+	69:  "NA-016", // Cayman Islands
+	71:  "SA-004", // Galapagos Islands
+	82:  "NA-097", // Jamaica
+	84:  "NA-107", // Martinique
+	91:  "SA-036", // Aruba
+	94:  "NA-100", // Antigua & Barbuda
+	95:  "NA-101", // Dominica
+	96:  "NA-103", // Montserrat
+	97:  "NA-108", // St. Lucia
+	99:  "AF-011", // Glorioso Islands
+	103: "OC-026", // Guam
+	105: "NA-015", // Guantanamo Bay
+	106: "EU-114", // Guernsey
+	111: "AN-003", // Heard Island
+	114: "EU-116", // Isle Of Man
+	118: "EU-022", // Jan Mayen
+	123: "OC-023", // Johnston Island
+	131: "AF-048", // Kerguelen Islands
+	133: "OC-039", // Kermadec Islands
+	138: "OC-020", // Kure Island
+	141: "SA-002", // Falkland Islands
+	147: "OC-004", // Lord Howe Island
+	153: "AN-005", // Macquarie Island
+	157: "OC-031", // Nauru
+	159: "AS-013", // Maldives
+	161: "SA-007", // Malpelo Island
+	165: "AF-049", // Mauritius
+	166: "OC-086", // Mariana Islands
+	167: "EU-053", // Market Reef
+	169: "AF-027", // Mayotte
+	171: "OC-072", // Mellish Reef
+	174: "OC-030", // Midway Island
+	177: "OC-073", // Minami Torishima
+	182: "NA-098", // Navassa Island
+	188: "OC-040", // Niue
+	189: "OC-005", // Norfolk Island
+	190: "OC-097", // Samoa
+	195: "AF-039", // Annobon Island
+	199: "AN-004", // Peter 1 Island
+	201: "AF-021", // Pr. Edward & Marion Is.
+	205: "AF-003", // Ascension Island
+	207: "AF-017", // Rodriguez Island
+	211: "NA-063", // Sable Island
+	217: "SA-013", // San Felix & San Ambrosio
+	222: "EU-018", // Faroe Islands
+	238: "AN-008", // South Orkney Islands
+	240: "AN-009", // South Sandwich Islands
+	241: "AN-010", // South Shetland Islands
+	247: "AS-051", // Spratly Islands
+	249: "NA-104", // St. Kitts & Nevis
+	250: "AF-022", // St. Helena
+	252: "NA-094", // St. Paul Island
+	253: "SA-014", // St. Peter & St. Paul
+	257: "EU-023", // Malta
+	270: "OC-048", // Tokelau Islands
+	273: "SA-010", // Trindade & Martim Vaz
+	276: "AF-031", // Tromelin Island
+	277: "NA-032", // St. Pierre & Miquelon
+	282: "OC-015", // Tuvalu
+	283: "AS-004", // Uk Base Areas On Cyprus
+	285: "NA-106", // Us Virgin Islands
+	297: "OC-053", // Wake Island
+	301: "OC-017", // Western Kiribati
+	303: "OC-007", // Willis Island
+	381: "AS-019", // Singapore
+	411: "AF-007", // Comoros
+	453: "AF-016", // Reunion Island
+	460: "OC-060", // Rotuma Island
+	489: "OC-112", // Conway Reef
+	490: "OC-018", // Banaba Island
+	505: "AS-110", // Pratas Island
+	506: "AS-116", // Scarborough Reef
+	509: "OC-027", // Marquesas Islands
+	512: "OC-176", // Chesterfield Islands
+	513: "OC-182", // Ducie Island
+	515: "OC-200", // Swains Island
+	516: "NA-146", // St. Barthelemy
+	517: "SA-099", // Curacao
+	519: "NA-145", // Saba & St. Eustatius
+	520: "SA-006", // Bonaire
+}
+
+// IOTAForDXCC returns the island reference of a DXCC entity that is a single
+// IOTA group, or "" for an entity that holds several islands or none.
+func IOTAForDXCC(dxcc int) string { return iotaByDXCC[dxcc] }

internal/awardref/dxcciota_test.go

@@ -0,0 +1,51 @@
+package awardref
+
+import (
+	"regexp"
+	"testing"
+)
+
+// The entities an operator meets: an island group that is its own DXCC, and a
+// mainland country that is not.
+func TestIOTAForDXCC(t *testing.T) {
+	cases := []struct {
+		dxcc int
+		want string
+		why  string
+	}{
+		{205, "AF-003", "Ascension Island is one island and one reference"},
+		{5, "EU-002", "Aland Islands"},
+		{12, "NA-022", "Anguilla"},
+		{24, "AN-002", "Bouvet — the entity whose entry is checked most often and worked least"},
+		// France holds hundreds of islands and, far more to the point, a
+		// mainland. Filling a reference here would earn nothing and would put
+		// one on nearly every European contact in the log.
+		{227, "", "France is not one IOTA"},
+		{291, "", "the United States is not one IOTA"},
+		{0, "", "no entity resolved"},
+		{99999, "", "not an entity at all"},
+	}
+	for _, c := range cases {
+		if got := IOTAForDXCC(c.dxcc); got != c.want {
+			t.Errorf("IOTAForDXCC(%d) = %q, want %q — %s", c.dxcc, got, c.want, c.why)
+		}
+	}
+}
+
+// Every reference in the table must be a well-formed IOTA reference, because
+// one that is not would be written onto a contact's award references and count
+// for nothing — and would then have to be found by hand, one QSO at a time.
+func TestEveryOneIOTAEntryIsWellFormed(t *testing.T) {
+	ref := regexp.MustCompile(`^(AF|AN|AS|EU|NA|OC|SA)-\d{3}$`)
+	if len(iotaByDXCC) < 50 {
+		t.Fatalf("the table holds %d entities — it has been truncated", len(iotaByDXCC))
+	}
+	for dxcc, r := range iotaByDXCC {
+		if dxcc <= 0 {
+			t.Errorf("entity number %d is not a DXCC entity", dxcc)
+		}
+		if !ref.MatchString(r) {
+			t.Errorf("entity %d maps to %q, which is not an IOTA reference", dxcc, r)
+		}
+	}
+}

internal/awardref/rename_test.go

@@ -0,0 +1,112 @@
+package awardref
+
+import (
+	"context"
+	"path/filepath"
+	"testing"
+
+	"hamlog/internal/db"
+)
+
+func renameRepo(t *testing.T) *Repo {
+	t.Helper()
+	conn, err := db.Open(filepath.Join(t.TempDir(), "a.db"))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { conn.Close() })
+	return NewRepo(conn)
+}
+
+// Correcting a reference's number must keep the reference.
+//
+// WAJA shipped numbered by the Japanese state instead of by the JARL, and until
+// now the only way to fix that was to delete all 47 references and import a new
+// list — losing anything the operator had adjusted. A rename keeps the pattern,
+// the entity list and the validity window, because a wrong NUMBER is all that
+// was wrong.
+func TestRenameKeepsEverythingButTheCode(t *testing.T) {
+	r := renameRepo(t)
+	ctx := context.Background()
+	if err := r.Upsert(ctx, "WAJA", Ref{
+		Code: "13", Name: "Tokyo", Pattern: `\bTok[iy]o\b`, Valid: true,
+		DXCCList: []int{339}, ValidFrom: "1970-01-01",
+	}); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := r.Rename(ctx, "WAJA", "13", "10"); err != nil {
+		t.Fatalf("rename: %v", err)
+	}
+
+	refs, err := r.List(ctx, "WAJA")
+	if err != nil {
+		t.Fatalf("list: %v", err)
+	}
+	if len(refs) != 1 {
+		t.Fatalf("WAJA holds %d references after a rename, want 1 — it was copied, not renamed", len(refs))
+	}
+	got := refs[0]
+	if got.Code != "10" {
+		t.Errorf("code = %q, want 10", got.Code)
+	}
+	if got.Name != "Tokyo" || got.Pattern != `\bTok[iy]o\b` {
+		t.Errorf("the reference lost what it carried: name=%q pattern=%q", got.Name, got.Pattern)
+	}
+	if len(got.DXCCList) != 1 || got.DXCCList[0] != 339 || got.ValidFrom != "1970-01-01" {
+		t.Errorf("the reference lost its entity list or dates: %+v", got)
+	}
+}
+
+// A number already in use must be refused. Left to REPLACE, the rename would
+// take the other reference's name, pattern and dates with it — one entry
+// silently swallowing another, found much later as a prefecture that has
+// quietly gone missing from the list.
+func TestRenameRefusesANumberAlreadyTaken(t *testing.T) {
+	r := renameRepo(t)
+	ctx := context.Background()
+	if err := r.Upsert(ctx, "WAJA", Ref{Code: "10", Name: "Gunma", Valid: true}); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+	if err := r.Upsert(ctx, "WAJA", Ref{Code: "13", Name: "Tokyo", Valid: true}); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	if err := r.Rename(ctx, "WAJA", "13", "10"); err == nil {
+		t.Fatal("renaming onto an existing number was accepted — one reference would have eaten the other")
+	}
+	refs, _ := r.List(ctx, "WAJA")
+	if len(refs) != 2 {
+		t.Fatalf("WAJA holds %d references, want both still there", len(refs))
+	}
+}
+
+// Renaming something that is not there is an error, not a silent no-op: it
+// means the editor and the store disagree about what the award holds.
+func TestRenameAnUnknownReferenceFails(t *testing.T) {
+	r := renameRepo(t)
+	if err := r.Rename(context.Background(), "WAJA", "99", "10"); err == nil {
+		t.Error("renaming a reference the award does not have was accepted")
+	}
+}
+
+// Codes are stored upper-cased, so a rename must compare the same way — else
+// "eu-048" onto "EU-048" looks like a move and is really the same reference,
+// which the collision check has to catch.
+func TestRenameIsCaseInsensitive(t *testing.T) {
+	r := renameRepo(t)
+	ctx := context.Background()
+	if err := r.Upsert(ctx, "IOTA", Ref{Code: "EU-048", Name: "Belle-Ile", Valid: true}); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+	if err := r.Rename(ctx, "iota", "eu-048", "eu-048"); err != nil {
+		t.Errorf("renaming a reference to itself in another case failed: %v", err)
+	}
+	if err := r.Rename(ctx, "IOTA", "eu-048", "eu-049"); err != nil {
+		t.Fatalf("rename: %v", err)
+	}
+	refs, _ := r.List(ctx, "IOTA")
+	if len(refs) != 1 || refs[0].Code != "EU-049" {
+		t.Errorf("references = %+v, want the one renamed to EU-049 and upper-cased", refs)
+	}
+}

internal/cat/icomserial.go

@@ -1,6 +1,7 @@
 package cat
 
 import (
+	"errors"
 	"fmt"
 	"strings"
 	"sync"
@@ -535,11 +536,36 @@ func (b *IcomSerial) SetMode(mode string) error {
 	return nil
 }
 
+// errIcomAckLost is "the reply never came" — as opposed to a reply that said no
+// (NG) or a dead port. Only this one is worth repeating: the command may well
+// have been carried out and only its acknowledgement lost. Sentinel rather than
+// a formatted string so callers can tell the two apart.
+var errIcomAckLost = errors.New("icom: timeout waiting for response")
+
+// SetPTT keys or unkeys the transmitter (CI-V 0x1C 0x00), retrying ONCE when the
+// acknowledgement is lost.
+//
+// A missing FB is not a missing command — the rig acts on the frame as soon as it
+// decodes it, and what expires is our wait for the answer on a bus shared with
+// the rig's own transceive updates. JTDX in "Split Operating: Fake It" moves the
+// dial immediately before every key-down, so the PTT ack queues behind that
+// traffic, and one lost ack was fatal: rigctld answered RPRT -9, JTDX read that
+// as losing rig control and tore the connection down mid-over, reopening it a
+// moment later (an operator's log shows exactly that, twice, a new rigctld client
+// within 300 ms of each failure). The same session over TCI never failed, because
+// TCI carries no CI-V and needs no Fake It.
+//
+// Re-sending is safe: asking for a state the rig is already in changes nothing.
 func (b *IcomSerial) SetPTT(on bool) error {
 	state := byte(0)
 	if on {
 		state = 1
 	}
+	err := b.exec(civ.CmdPTT, civ.SubPTT, state)
+	if err == nil || !errors.Is(err, errIcomAckLost) {
+		return err
+	}
+	applog.Printf("icom: PTT %v — no acknowledgement in %s, sending it once more", on, icomCmdTimeout)
 	return b.exec(civ.CmdPTT, civ.SubPTT, state)
 }
 
@@ -619,7 +645,7 @@ func (b *IcomSerial) recv(timeout time.Duration, match func(civ.Decoded) bool) (
 		case <-cancel:
 			return civ.Decoded{}, fmt.Errorf("icom: interrupted")
 		case <-deadline:
-			return civ.Decoded{}, fmt.Errorf("icom: timeout waiting for response")
+			return civ.Decoded{}, errIcomAckLost
 		}
 	}
 }

internal/cat/tci.go

@@ -46,8 +46,33 @@ type TCI struct {
 	mode   string
 	split  bool
 	tx     bool
+	// txAllowed is what the radio last said about TRANSMIT PERMISSION.
+	//
+	// TX_ENABLE is sent by ExpertSDR when a client connects and again whenever
+	// the band changes, "in case transmitter permission was changed" (§4.3). When
+	// it is false the radio silently ignores TRX — which is exactly what an
+	// operator sees as "PTT does nothing", with no error anywhere to explain it.
+	//
+	// txAllowedKnown keeps an OLDER ExpertSDR, or a TCI-compatible program that
+	// never sends TX_ENABLE at all, from being treated as refusing: without a
+	// word from the radio we key and let it decide.
+	txAllowed      bool
+	txAllowedKnown bool
 
 	lastSig string // last logged state signature (log only on change)
+
+	// spotFreq is the frequency of the marker currently on the panorama for each
+	// callsign — the panadapter's own state, which TCI never reports back. It is
+	// what makes one spot per call possible: without it there is no way to know
+	// there is an older marker to delete.
+	spotFreq map[string]int64
+}
+
+func absInt64(v int64) int64 {
+	if v < 0 {
+		return -v
+	}
+	return v
 }
 
 const tciDefaultPort = 40001
@@ -96,18 +121,73 @@ func (t *TCI) Connect() error {
 	t.mu.Lock()
 	t.conn = conn
 	t.ready = false
+	// Forget the previous session's transmit permission: the radio announces it
+	// again on connect, and a refusal remembered from a band we have since left
+	// would block PTT until it did.
+	t.txAllowed, t.txAllowedKnown = false, false
 	t.mu.Unlock()
 	debugLog.Printf("TCI: connected to %s", url)
 	go t.reader(conn)
 	if t.spotsEnabled {
+		// Forget what we thought was on the panorama at the same moment the radio
+		// is told to drop it. Kept, the memory would suppress the next spot for
+		// each of those calls as "already drawn" onto a panorama now empty.
+		t.mu.Lock()
+		t.spotFreq = map[string]int64{}
+		t.mu.Unlock()
 		_ = t.send("spot_clear;") // drop any leftover spots from a previous session
 	}
 	return nil
 }
 
+// spotFreqTolHz is how far a re-spot of the same callsign may sit from the one
+// already on the panorama before it is treated as a move rather than the same
+// spot said again.
+//
+// Two spotters hearing the same CW station rarely agree to better than a couple
+// of hundred hertz, and every one of them produces a cluster line. Below this
+// they are the same spot and nothing is sent at all; above it the marker is
+// deleted and redrawn where the station now is.
+const spotFreqTolHz = 500
+
+// noteSpot records what the panorama is about to hold for a callsign and says
+// what has to be sent: whether to draw at all, and whether an older marker for
+// the same call must be deleted first.
+//
+// Separate from SendSpot so the rule can be tested without a radio — and
+// because the lock must be released before anything is sent: t.send takes t.mu
+// itself, Go mutexes are not reentrant, and sending while holding it would
+// deadlock the backend and take the rig offline.
+func (t *TCI) noteSpot(call string, freqHz int64) (draw, deletePrev bool) {
+	key := strings.ToUpper(strings.TrimSpace(call))
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	prev, had := t.spotFreq[key]
+	if had && absInt64(prev-freqHz) <= spotFreqTolHz {
+		return false, false
+	}
+	if t.spotFreq == nil {
+		t.spotFreq = map[string]int64{}
+	}
+	if len(t.spotFreq) > 4000 {
+		t.spotFreq = map[string]int64{} // bound memory on a long session
+		had = false                     // forgotten: nothing left to delete by name
+	}
+	t.spotFreq[key] = freqHz
+	return true, had
+}
+
 // SendSpot mirrors a cluster spot onto the TCI panorama (implements Spotter).
-// The radio replaces a spot that has the same callsign, so re-spotting updates
-// it in place. No-op when spot mirroring is disabled.
+// No-op when spot mirroring is disabled.
+//
+// ONE MARKER PER CALLSIGN. This code assumed the radio replaced a spot carrying
+// a callsign it already had; it does not. ExpertSDR keys a spot on its
+// frequency too, so a DX station spotted by three operators — 14025.00,
+// 14025.12, 14024.90, which is an ordinary minute on a cluster — was drawn
+// three times, a few pixels apart, and stayed that way.
+//
+// So the previous spot for the call is deleted before the new one is sent,
+// which is what the FlexRadio backend has always done (spot remove / spot add).
 func (t *TCI) SendSpot(s SpotInfo) error {
 	if !t.spotsEnabled {
 		return nil
@@ -116,6 +196,17 @@ func (t *TCI) SendSpot(s SpotInfo) error {
 	if call == "" || s.FreqHz <= 0 {
 		return nil
 	}
+	draw, deletePrev := t.noteSpot(call, s.FreqHz)
+	if !draw {
+		return nil // the same station said again by another spotter
+	}
+	if deletePrev {
+		// SPOT_DELETE takes the callsign alone. Not in the protocol PDF this
+		// backend was written from; confirmed against ars-ka0s/eesdr-tci, which
+		// lists SPOT (5 arguments), SPOT_DELETE (1) and SPOT_CLEAR (0) — the
+		// other two matching what already works here.
+		_ = t.send(fmt.Sprintf("spot_delete:%s;", call))
+	}
 	// TCI's SPOT command wants the colour as a signed 32-bit DECIMAL integer in
 	// 0xAARRGGBB order — NOT a "0x…" hex string (e.g. "spot:UN7GK,cw,14025000,
 	// -16776961,test;"). ExpertSDR silently drops a spot whose colour field it
@@ -223,7 +314,22 @@ func (t *TCI) SetMode(mode string) error {
 }
 
 // SetPTT keys or unkeys the transmitter (VFO 0).
+//
+// A refusal by the radio is reported rather than swallowed. ExpertSDR announces
+// transmit permission with TX_ENABLE and then simply IGNORES trx when it is
+// false — out-of-band frequency, TX disabled in the program, no PA. The command
+// went out, nothing happened, and nothing anywhere said why. Now the operator
+// is told, and the message names the place to look.
 func (t *TCI) SetPTT(on bool) error {
+	if on {
+		t.mu.Lock()
+		known, allowed := t.txAllowedKnown, t.txAllowed
+		t.mu.Unlock()
+		if known && !allowed {
+			return fmt.Errorf("the radio is refusing to transmit (TCI reports TX disabled) — " +
+				"check the frequency is inside a transmit band and that TX is enabled in ExpertSDR")
+		}
+	}
 	return t.send(fmt.Sprintf("trx:0,%t;", on))
 }
 
@@ -317,6 +423,14 @@ func (t *TCI) handle(msg string) {
 		if get(0) == "0" {
 			t.tx = get(1) == "true"
 		}
+	case "tx_enable":
+		if get(0) == "0" {
+			allowed := get(1) == "true"
+			if !t.txAllowedKnown || t.txAllowed != allowed {
+				debugLog.Printf("TCI: the radio %s transmitting", map[bool]string{true: "allows", false: "REFUSES"}[allowed])
+			}
+			t.txAllowed, t.txAllowedKnown = allowed, true
+		}
 	default:
 		lname := strings.ToLower(name)
 		// A click on one of our panorama spots comes back as

internal/cat/tci_ptt_test.go

@@ -0,0 +1,80 @@
+//go:build windows
+
+package cat
+
+import "testing"
+
+// feed pushes messages at the backend the way the radio would.
+func feed(t *TCI, msgs ...string) {
+	for _, m := range msgs {
+		t.handle(m)
+	}
+}
+
+// "PTT via CAT does nothing on TCI."
+//
+// ExpertSDR announces transmit permission with TX_ENABLE — on connect, and
+// again whenever the band changes "in case transmitter permission was changed"
+// (§4.3 of the protocol document). When it is false the radio simply IGNORES
+// trx. OpsLog sent the documented command, the radio discarded it, and nothing
+// anywhere said why: the operator pressed a dead key.
+func TestPTTIsRefusedOutLoudWhenTheRadioForbidsTransmitting(t *testing.T) {
+	tci := NewTCI("localhost", 40001, "FT8", false)
+	feed(tci, "tx_enable:0,false")
+
+	err := tci.SetPTT(true)
+	if err == nil {
+		t.Fatal("keying was accepted while the radio forbids transmitting — the operator gets no reason at all")
+	}
+	// The message has to name where to look; "PTT failed" sends nobody anywhere.
+	for _, want := range []string{"transmit", "ExpertSDR"} {
+		if !contains(err.Error(), want) {
+			t.Errorf("the refusal reads %q, which does not mention %q", err.Error(), want)
+		}
+	}
+
+	// Unkeying is never blocked. Whatever the radio thinks about permission, a
+	// request to STOP transmitting must always reach it.
+	if err := tci.SetPTT(false); err != nil && contains(err.Error(), "refusing") {
+		t.Errorf("unkeying was refused: %v", err)
+	}
+}
+
+// Permission comes back when the operator returns to a band they may use, and
+// PTT has to come back with it — not stay blocked until OpsLog is restarted.
+func TestPermissionGrantedAgainRestoresPTT(t *testing.T) {
+	tci := NewTCI("localhost", 40001, "FT8", false)
+	feed(tci, "tx_enable:0,false")
+	if err := tci.SetPTT(true); err == nil {
+		t.Fatal("keying was accepted while forbidden")
+	}
+	feed(tci, "tx_enable:0,true")
+	// No connection here, so the send fails — but it must fail as a TRANSPORT
+	// error, never as a refusal.
+	if err := tci.SetPTT(true); err != nil && contains(err.Error(), "refusing") {
+		t.Errorf("still refusing after permission was granted: %v", err)
+	}
+}
+
+// A radio that never mentions TX_ENABLE — an older ExpertSDR, or one of the
+// other programs that speak TCI — must not be treated as refusing. Silence is
+// not a "no": we key, and let the radio decide.
+func TestSilenceAboutPermissionIsNotARefusal(t *testing.T) {
+	tci := NewTCI("localhost", 40001, "FT8", false)
+	if err := tci.SetPTT(true); err != nil && contains(err.Error(), "refusing") {
+		t.Errorf("a radio that never sent TX_ENABLE was treated as forbidding transmit: %v", err)
+	}
+}
+
+func contains(s, sub string) bool {
+	return len(sub) == 0 || (len(s) >= len(sub) && indexOf(s, sub) >= 0)
+}
+
+func indexOf(s, sub string) int {
+	for i := 0; i+len(sub) <= len(s); i++ {
+		if s[i:i+len(sub)] == sub {
+			return i
+		}
+	}
+	return -1
+}

internal/cat/tci_spot_test.go

@@ -0,0 +1,87 @@
+//go:build windows
+
+package cat
+
+import "testing"
+
+// The reported symptom: with spot mirroring on, the same station appeared two
+// or three times on the panorama.
+//
+// Its cause is not in OpsLog's spot pipeline — one cluster line produces one
+// SendSpot. It is that a popular DX station IS spotted two or three times, by
+// different operators within the same minute, and no two of them agree on the
+// frequency to better than a few tens of hertz. The backend assumed ExpertSDR
+// replaced a spot bearing a callsign it already had; it keys on the frequency
+// too, so each of those became its own marker.
+func TestSameStationSpottedBySeveralOperatorsIsDrawnOnce(t *testing.T) {
+	tci := &TCI{spotsEnabled: true}
+
+	draw, del := tci.noteSpot("UN7GK", 14025000)
+	if !draw || del {
+		t.Fatalf("first spot: draw=%v delete=%v, want draw and nothing to delete", draw, del)
+	}
+	// The same station, two more spotters, a few tens of hertz apart.
+	for _, hz := range []int64{14025120, 14024900} {
+		if draw, del := tci.noteSpot("UN7GK", hz); draw || del {
+			t.Errorf("re-spot at %d Hz: draw=%v delete=%v, want nothing sent — this is the duplicate marker", hz, draw, del)
+		}
+	}
+}
+
+// A station that really moves must still move on the panorama, and the marker
+// left where it was must go. Deleting first is the whole difference between
+// "the spot follows the station" and "the station collects markers".
+func TestAStationThatMovesReplacesItsMarker(t *testing.T) {
+	tci := &TCI{spotsEnabled: true}
+	tci.noteSpot("UN7GK", 14025000)
+
+	draw, del := tci.noteSpot("UN7GK", 14032000) // 7 kHz up: a real QSY
+	if !draw || !del {
+		t.Fatalf("after a QSY: draw=%v delete=%v, want the old marker deleted and a new one drawn", draw, del)
+	}
+	// And the new position becomes the reference, so spotters agreeing with it
+	// are quiet again.
+	if draw, _ := tci.noteSpot("UN7GK", 14032100); draw {
+		t.Error("a spot at the station's new frequency was drawn again")
+	}
+}
+
+// Case matters nowhere in ham radio, and the cluster is not consistent about it.
+func TestSpotMemoryIgnoresCase(t *testing.T) {
+	tci := &TCI{spotsEnabled: true}
+	tci.noteSpot("un7gk", 14025000)
+	if draw, _ := tci.noteSpot("UN7GK", 14025000); draw {
+		t.Error("the same call in another case was treated as a different station")
+	}
+}
+
+// Two different stations are two markers — the whole point of the panorama.
+func TestDifferentStationsEachGetAMarker(t *testing.T) {
+	tci := &TCI{spotsEnabled: true}
+	tci.noteSpot("UN7GK", 14025000)
+	draw, del := tci.noteSpot("ZD7BG", 14025050) // 50 Hz away, a different operator
+	if !draw {
+		t.Error("a second station near the first was swallowed as a duplicate")
+	}
+	if del {
+		t.Error("deleting by callsign would have removed a spot this station never had")
+	}
+}
+
+// The connection drops and comes back: Connect sends spot_clear, so the
+// panorama is empty. If the memory survived that, the next spot for each of
+// those calls would be suppressed as "already drawn" onto an empty panorama —
+// the operator's spots would simply stop appearing until they changed
+// frequency.
+func TestReconnectingForgetsWhatWasDrawn(t *testing.T) {
+	tci := &TCI{spotsEnabled: true}
+	tci.noteSpot("UN7GK", 14025000)
+
+	tci.mu.Lock()
+	tci.spotFreq = map[string]int64{} // what Connect does alongside spot_clear
+	tci.mu.Unlock()
+
+	if draw, del := tci.noteSpot("UN7GK", 14025000); !draw || del {
+		t.Errorf("after a reconnect: draw=%v delete=%v, want it drawn again and nothing deleted", draw, del)
+	}
+}

internal/cluster/cluster.go

@@ -18,6 +18,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"hamlog/internal/applog"
@@ -355,6 +356,18 @@ func (s *session) run() {
 // that gets skipped.
 var idleTick = 30 * time.Second
 
+// promptTick is the read deadline used until the login handshake is finished.
+// Short, because a node's "login:" / "password:" carries no newline and is only
+// visible when the read times out — see the read loop. It costs a few wake-ups
+// during the first seconds of a connection and nothing afterwards.
+const promptTick = 700 * time.Millisecond
+
+// handshakeWindow bounds how long the fast promptTick applies. A node that has
+// a password configured but never asks for one would otherwise keep the loop
+// waking every 700 ms for the life of the connection, for a prompt that is never
+// coming. Any real login exchange is over in a second or two.
+const handshakeWindow = 20 * time.Second
+
 // quietNotice is the silence after which the log says so, once.
 const quietNotice = 10 * time.Minute
 
@@ -398,16 +411,37 @@ func (s *session) runOnce() (time.Time, error) {
 
 	// Login: send on first prompt OR blindly after 1.5s. Many DXSpider
 	// nodes accept the callsign without re-prompting.
-	loginSent := false
+	//
+	// Atomic because the blind-login timer below and the read loop both touch
+	// these. loginSent used to be a plain bool that the timer NEVER SET: the
+	// callsign went out and nothing recorded it, so the password branch — gated on
+	// loginSent — was dead code, and the session could only reach "connected" by
+	// recognising a welcome banner. On a node whose entire greeting is a bare
+	// "login:" and which then demands a password (f5mzn.org:9000), that left the
+	// server stuck at "connecting" for ever while telnet logged in by hand fine.
+	var loginSent, pwdSent atomic.Bool
+	// CompareAndSwap, not a plain store: the timer and the loop can reach these at
+	// the same moment, and the callsign must be written exactly once.
+	sendLogin := func() {
+		if s.login != "" && loginSent.CompareAndSwap(false, true) {
+			_, _ = conn.Write([]byte(s.login + "\r\n"))
+		}
+	}
+	// Sent ONCE per connection. A node that re-prompts is refusing the password,
+	// and answering with the same one again only loops — better to let the
+	// refusal show in the console than to hide it behind a retry.
+	sendPassword := func() {
+		if s.cfg.Password != "" && loginSent.Load() && pwdSent.CompareAndSwap(false, true) {
+			_, _ = conn.Write([]byte(s.cfg.Password + "\r\n"))
+		}
+	}
 	if s.login != "" {
 		go func() {
 			select {
 			case <-s.stopCh:
 				return
 			case <-time.After(1500 * time.Millisecond):
-				if !loginSent {
-					_, _ = conn.Write([]byte(s.login + "\r\n"))
-				}
+				sendLogin()
 			}
 		}()
 	}
@@ -453,7 +487,21 @@ func (s *session) runOnce() (time.Time, error) {
 
 	var connectedAt time.Time
 	var quiet time.Duration // how long the node has said nothing
+	var quietNoticed bool   // the long-silence line is said once
 	var pending string      // a line cut in half by a read deadline
+	markConnected := func() {
+		if s.snapshot().State == StateConnected {
+			return
+		}
+		connectedAt = time.Now()
+		s.mu.Lock()
+		s.status.State = StateConnected
+		s.status.ConnectedAt = connectedAt
+		s.status.Error = ""
+		s.mu.Unlock()
+		s.emitStatus()
+		fireInitCommands()
+	}
 	rd := bufio.NewReader(conn)
 	for {
 		select {
@@ -471,7 +519,21 @@ func (s *session) runOnce() (time.Time, error) {
 		// Only a REAL error ends the session. A dead peer still gets caught:
 		// TCP keepalive probes an idle connection and its failure arrives here
 		// as an error, not as a timeout.
-		_ = conn.SetReadDeadline(time.Now().Add(idleTick))
+		// SHORT deadline until the handshake is done, the long idle tick after.
+		//
+		// A cluster writes its prompts WITHOUT a trailing newline, and ReadString
+		// only returns on one — so a prompt is never a "line" at all, it is whatever
+		// is sitting in the buffer when the read deadline expires. At the ordinary
+		// 30 s tick that made a bare "login:" invisible for half a minute and a
+		// following "password:" invisible for another, which is long enough for the
+		// node to give up on us. Only the handshake needs the fast tick; once logged
+		// in, a long deadline is exactly what we want (see idleTick).
+		tick := idleTick
+		if time.Since(linkUpAt) < handshakeWindow &&
+			(!loginSent.Load() || (s.cfg.Password != "" && !pwdSent.Load())) {
+			tick = promptTick
+		}
+		_ = conn.SetReadDeadline(time.Now().Add(tick))
 		chunk, err := rd.ReadString('\n')
 		if err != nil {
 			var ne net.Error
@@ -480,16 +542,48 @@ func (s *session) runOnce() (time.Time, error) {
 				// Keep it: a spot line straddling the deadline would otherwise lose
 				// its first half and arrive as nonsense, or vanish entirely.
 				pending += chunk
-				quiet += idleTick
+				// …but a newline-less PROMPT is not half a line, it is a question,
+				// and this is the only place it can ever be seen. Answer it, show it
+				// in the console (an operator watching a stuck server has a right to
+				// see what the node actually asked), and drop it so it is not glued
+				// onto the front of the next real line.
+				if p := strings.TrimSpace(pending); p != "" {
+					switch {
+					case !loginSent.Load() && s.login != "" && isLoginPrompt(p):
+						s.emitLine(p, false)
+						pending = ""
+						sendLogin()
+						if s.cfg.Password == "" {
+							markConnected()
+						}
+						continue
+					case !pwdSent.Load() && isPasswordPrompt(p):
+						s.emitLine(p, false)
+						pending = ""
+						if s.cfg.Password == "" {
+							// Nothing to answer with. Shown in the console rather than
+							// swallowed: an unanswered "password:" sitting there IS the
+							// explanation for a server that never finishes connecting,
+							// and it is something the operator can act on.
+							continue
+						}
+						sendPassword()
+						markConnected()
+						continue
+					}
+				}
+				quiet += tick
 				// Said once at the first long silence, so a genuinely mute node is
 				// visible without a line every tick.
-				if quiet == quietNotice {
+				if !quietNoticed && quiet >= quietNotice {
+					quietNoticed = true
 					applog.Printf("cluster[%s] no traffic for %s — still connected", s.cfg.Name, quiet)
 				}
 				continue
 			}
 			return connectedAt, fmt.Errorf("read: %w", err)
 		}
+		quietNoticed = false
 		quiet = 0
 		line := pending + chunk
 		pending = ""
@@ -519,28 +613,23 @@ func (s *session) runOnce() (time.Time, error) {
 			}
 		}
 
-		// Login on explicit prompt.
-		if !loginSent && s.login != "" && isLoginPrompt(line) {
-			_, _ = conn.Write([]byte(s.login + "\r\n"))
-			loginSent = true
+		// Login on explicit prompt — the case where the node DID terminate it with
+		// a newline. The newline-less form is handled on the timeout path above.
+		if !loginSent.Load() && s.login != "" && isLoginPrompt(line) {
+			s.emitLine(line, false)
+			sendLogin()
 			continue
 		}
-		// Password on prompt (rare).
-		if loginSent && s.cfg.Password != "" && isPasswordPrompt(line) {
-			_, _ = conn.Write([]byte(s.cfg.Password + "\r\n"))
+		// Password on prompt.
+		if !pwdSent.Load() && isPasswordPrompt(line) {
+			s.emitLine(line, false)
+			sendPassword()
 			continue
 		}
 
 		// Mark connected once we've sent login OR seen a welcome banner.
-		if s.snapshot().State != StateConnected && (loginSent || isWelcome(line)) {
-			connectedAt = time.Now()
-			s.mu.Lock()
-			s.status.State = StateConnected
-			s.status.ConnectedAt = connectedAt
-			s.status.Error = ""
-			s.mu.Unlock()
-			s.emitStatus()
-			fireInitCommands()
+		if loginSent.Load() || isWelcome(line) {
+			markConnected()
 		}
 
 		// EVERY line goes to the console — spot or not. This is the whole point:
@@ -701,6 +790,38 @@ func parseSpot(line string) (Spot, bool) {
 	}, true
 }
 
+// NewLocalSpot builds the Spot for a DX announcement WE just sent, so it lands
+// in the operator's own list at once.
+//
+// A node does not necessarily broadcast a spot back to the station that sent it:
+// DXSpider suppresses the echo to the originator, and a node-side filter can eat
+// it too. So an operator spotted a station, watched their own spot list stay
+// empty, and concluded the spot had never gone out — when it had. This is the
+// spot the echo would have carried, built from what we sent, deliberately the
+// same shape so the UI's call+band de-dupe folds the two into one row on the
+// nodes that DO echo.
+func NewLocalSpot(srv ServerConfig, spotter, dxCall string, freqKHz float64, comment string) Spot {
+	freqHz := int64(freqKHz*1000 + 0.5)
+	now := time.Now()
+	sp := Spot{
+		SourceID:   srv.ID,
+		SourceName: srv.Name,
+		Spotter:    strings.ToUpper(strings.TrimSpace(spotter)),
+		DXCall:     strings.ToUpper(strings.TrimSpace(dxCall)),
+		FreqKHz:    freqKHz,
+		FreqHz:     freqHz,
+		Band:       bandFromHz(freqHz),
+		Comment:    strings.TrimSpace(comment),
+		TimeUTC:    now.UTC().Format("1504") + "Z",
+		ReceivedAt: now,
+	}
+	// Raw reads like the node's own broadcast — it is what anything showing the
+	// source line expects, and it keeps a local spot legible in the log.
+	sp.Raw = fmt.Sprintf("DX de %s: %9.1f  %-12s %-30s %s",
+		sp.Spotter, sp.FreqKHz, sp.DXCall, sp.Comment, sp.TimeUTC)
+	return sp
+}
+
 func isLoginPrompt(s string) bool {
 	low := strings.ToLower(s)
 	return strings.Contains(low, "login:") ||

internal/cluster/login_test.go

@@ -0,0 +1,105 @@
+package cluster
+
+import (
+	"bufio"
+	"net"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+)
+
+// A node whose whole greeting is a bare "login:" — no newline, no banner — and
+// which then demands a password must still log in.
+//
+// Reported on f5mzn.org:9000: telnet by hand worked, OpsLog sat at "connecting"
+// for ever. Two faults met there. The prompts carry no newline, and the read
+// loop only ever looked at complete LINES, so neither prompt was seen at all;
+// and the blind 1.5 s login never recorded that it had sent the callsign, which
+// left the password branch — gated on that flag — permanently switched off.
+//
+// The test speaks the node's side literally: "login:" with no newline, then
+// "password:" with no newline, then a spot. It asserts both answers arrive and
+// that the session reaches Connected without any welcome banner to lean on.
+func TestBareLoginAndPasswordPromptsWithoutNewlines(t *testing.T) {
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer ln.Close()
+
+	host, portStr, _ := net.SplitHostPort(ln.Addr().String())
+	port, err := strconv.Atoi(portStr)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	type answers struct{ login, pwd string }
+	got := make(chan answers, 1)
+	go func() {
+		c, err := ln.Accept()
+		if err != nil {
+			return
+		}
+		defer c.Close()
+		rd := bufio.NewReader(c)
+		// No newline, exactly as the node sends it.
+		if _, err := c.Write([]byte("login: ")); err != nil {
+			return
+		}
+		call, err := rd.ReadString('\n')
+		if err != nil {
+			return
+		}
+		if _, err := c.Write([]byte("password: ")); err != nil {
+			return
+		}
+		pwd, err := rd.ReadString('\n')
+		if err != nil {
+			return
+		}
+		got <- answers{strings.TrimSpace(call), strings.TrimSpace(pwd)}
+		// Something to prove the link is live and parsing again afterwards.
+		_, _ = c.Write([]byte("DX de F4BPO:     14074.0  OY1CT        FT8            1234Z\r\n"))
+		time.Sleep(time.Second)
+	}()
+
+	spots := make(chan Spot, 4)
+	s := &session{
+		cfg:      ServerConfig{Name: "pwd node", Host: host, Port: port, Password: "s3cret"},
+		login:    "F4BPO",
+		onSpot:   func(sp Spot) { spots <- sp },
+		onLine:   func(Line) {},
+		onStatus: func() {},
+		stopCh:   make(chan struct{}),
+	}
+	done := make(chan error, 1)
+	go func() { _, err := s.runOnce(); done <- err }()
+	defer func() { close(s.stopCh); <-done }()
+
+	select {
+	case a := <-got:
+		if a.login != "F4BPO" {
+			t.Errorf("callsign sent = %q, want F4BPO", a.login)
+		}
+		if a.pwd != "s3cret" {
+			t.Errorf("password sent = %q, want s3cret — the prompt carried no newline", a.pwd)
+		}
+	case <-time.After(10 * time.Second):
+		t.Fatal("the node's newline-less prompts were never answered")
+	}
+
+	select {
+	case sp := <-spots:
+		if sp.DXCall != "OY1CT" {
+			t.Errorf("spot from the wrong station: %+v", sp)
+		}
+	case <-time.After(5 * time.Second):
+		t.Fatal("no spot after the login — the stream is not being parsed")
+	}
+
+	// Connected without a welcome banner: the handshake alone must be enough.
+	if st := s.snapshot().State; st != StateConnected {
+		t.Errorf("state = %q, want %q — the server would still show as connecting", st, StateConnected)
+	}
+}

internal/db/indexedcolumns_test.go

@@ -0,0 +1,87 @@
+package db
+
+import (
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// A TEXT column that an index is built on must be listed in varcharColumns.
+//
+// MySQL cannot index a TEXT column without a prefix length: the migration dies
+// with error 1170 — and it dies again on every startup afterwards, leaving the
+// operator with a logbook that will not connect and no way forward from the
+// interface. The translator only emits VARCHAR for the names in varcharColumns,
+// and that list is maintained by hand.
+//
+// So: read the migrations, work out which columns are TEXT, find which are
+// indexed, and insist the list covers the overlap. Integer columns are indexed
+// perfectly well and are none of this test's business.
+func TestIndexedTextColumnsAreVarchar(t *testing.T) {
+	files, err := migrationsFS.ReadDir("migrations")
+	if err != nil {
+		t.Fatalf("read migrations: %v", err)
+	}
+
+	// name → declared type, from "ADD COLUMN name TYPE" and from the column
+	// lines inside a CREATE TABLE body.
+	declared := map[string]string{}
+	reAdd := regexp.MustCompile(`(?i)ADD\s+COLUMN\s+` + "`" + `?(\w+)` + "`" + `?\s+(\w+)`)
+	reCol := regexp.MustCompile(`(?im)^\s*` + "`" + `?(\w+)` + "`" + `?\s+(TEXT|VARCHAR|INTEGER|INT|REAL|BLOB|DATETIME|BOOLEAN)\b`)
+	reIdx := regexp.MustCompile(`(?i)CREATE\s+INDEX\s+(?:IF\s+NOT\s+EXISTS\s+)?\S+\s+ON\s+\S+\s*\(([^)]*)\)`)
+
+	var indexed []struct{ file, col string }
+	for _, f := range files {
+		if f.IsDir() || !strings.HasSuffix(f.Name(), ".sql") {
+			continue
+		}
+		body, err := migrationsFS.ReadFile("migrations/" + f.Name())
+		if err != nil {
+			t.Fatalf("read %s: %v", f.Name(), err)
+		}
+		src := string(body)
+		for _, m := range reAdd.FindAllStringSubmatch(src, -1) {
+			declared[strings.ToLower(m[1])] = strings.ToUpper(m[2])
+		}
+		for _, m := range reCol.FindAllStringSubmatch(src, -1) {
+			if _, seen := declared[strings.ToLower(m[1])]; !seen {
+				declared[strings.ToLower(m[1])] = strings.ToUpper(m[2])
+			}
+		}
+		for _, m := range reIdx.FindAllStringSubmatch(src, -1) {
+			for _, col := range strings.Split(m[1], ",") {
+				col = strings.TrimSpace(col)
+				if col == "" || strings.Contains(col, "(") {
+					continue // a prefix length or an expression is MySQL-safe already
+				}
+				col = strings.Trim(col, "`\"")
+				if i := strings.IndexAny(col, " \t"); i > 0 {
+					col = col[:i] // "col DESC"
+				}
+				indexed = append(indexed, struct{ file, col string }{f.Name(), strings.ToLower(col)})
+			}
+		}
+	}
+
+	if len(indexed) == 0 || len(declared) == 0 {
+		t.Fatal("nothing parsed out of the migrations — this test has stopped checking anything")
+	}
+
+	textIndexed := 0
+	for _, ix := range indexed {
+		typ, known := declared[ix.col]
+		if !known || typ != "TEXT" {
+			continue // an integer index, or a column this test could not type
+		}
+		textIndexed++
+		if !varcharColumns[ix.col] {
+			t.Errorf("%s indexes the TEXT column %q, which is not in varcharColumns — "+
+				"on MySQL that migration fails with error 1170 and the logbook stops connecting for good",
+				ix.file, ix.col)
+		}
+	}
+	if textIndexed == 0 {
+		t.Fatal("no indexed TEXT column found — the parsing has drifted and this test checks nothing")
+	}
+	t.Logf("%d indexed TEXT column(s) checked against varcharColumns", textIndexed)
+}

internal/db/migrations/0029_sync_uid.sql

@@ -0,0 +1,24 @@
+-- A stable identity per contact, for folder-based synchronisation.
+--
+-- Two PCs exchanging changes through a shared folder need to name the SAME
+-- contact in both logs. "the QSO with M0ABC at 14:32" is a guess, and the two
+-- machines can disagree about which row that is — so an edit or a deletion
+-- cannot be addressed at all without an identity that travels with the record.
+--
+-- A REAL COLUMN, indexed, rather than a key inside extras_json. The identity is
+-- looked up once per incoming change, and on a 120 000-QSO logbook scanning
+-- JSON for it would turn every sync into a full table read. A column costs one
+-- migration; the JSON would cost a scan every time.
+--
+-- sync_uid is listed in varcharColumns (internal/db/mysql.go): MySQL cannot
+-- index a TEXT column without a prefix length, and a migration that tries dies
+-- with error 1170 on every startup thereafter, with no way out from the UI.
+--
+-- Empty on every existing row, and it STAYS empty on most of them. The change
+-- log starts empty too, so contacts logged before synchronisation was switched
+-- on are never exchanged: nothing has to be copied across, and an identity is
+-- stamped only on a contact that is actually logged, edited or deleted from
+-- then on. Seeding a second machine with the existing log is a one-time copy of
+-- the database or an ADIF import, not something synchronisation does.
+ALTER TABLE qso ADD COLUMN sync_uid TEXT NOT NULL DEFAULT '';
+CREATE INDEX IF NOT EXISTS idx_qso_sync_uid ON qso (sync_uid);

internal/db/migrations/0030_lookup_iota.sql

@@ -0,0 +1,8 @@
+-- The lookup cache gains the IOTA reference QRZ was already sending.
+--
+-- QRZ's XML carries <iota>EU-048</iota> for an operator on an island, and
+-- nothing read it — so the one place an IOTA reference can be known BEFORE the
+-- contact is logged was thrown away on every lookup. There is no live "who is
+-- on an island right now" feed anywhere (unlike POTA), which makes the callbook
+-- record the practical source.
+ALTER TABLE callsign_cache ADD COLUMN iota TEXT;

internal/db/mysql.go

@@ -67,6 +67,10 @@ var varcharColumns = map[string]bool{
 	"callsign": true, "qso_date": true, "band": true, "mode": true,
 	"grid": true, "station_callsign": true, "state": true, "contest_id": true,
 	"sat_name": true, "prop_mode": true, "sig": true, "wwff_ref": true, "skcc": true,
+	"sync_uid": true, // qso index (0029) — folder-sync identity, looked up per record
+	// A new indexed TEXT column MUST be added here. TestIndexedTextColumnsAreVarchar
+	// reads the migrations and fails if one is missing, because the alternative is
+	// a MySQL logbook that stops connecting for good on error 1170.
 	// integrations_udp index (0011)
 	"direction": true,
 	// award_references composite primary key (0017)

internal/extsvc/clublog.go

@@ -117,6 +117,13 @@ func UploadClublogADIF(ctx context.Context, client *http.Client, cfg ServiceConf
 	if api == "" {
 		api = clublogAppAPIKey
 	}
+	// putlogs.php reads the upload as an ADIF *file*, so it needs a header.
+	// Callers that already build a full document (the QSL Manager) pass one;
+	// callers that only have <EOR>-terminated records (the on-close flush) do
+	// not, and a headerless file is rejected. Same rule as the LoTW writer.
+	if !strings.Contains(strings.ToUpper(adifDoc), "<EOH>") {
+		adifDoc = "OpsLog Club Log upload\n<PROGRAMID:6>OpsLog <EOH>\n" + adifDoc
+	}
 
 	var buf bytes.Buffer
 	mw := multipart.NewWriter(&buf)

internal/extsvc/eqsl.go

@@ -267,6 +267,78 @@ func UploadEQSL(ctx context.Context, client *http.Client, user, pswd, qthNick, a
 	return UploadResult{OK: false, Message: reason}, fmt.Errorf("eqsl: upload failed: %s", reason)
 }
 
+// eqslBatchMax is the largest number of records eQSL asks a single upload to
+// carry ("upload only files smaller than about 1000 records at a time", eQSL's
+// own ImportADIF interface notes). Callers chunk to this.
+const eqslBatchMax = 1000
+
+// UploadEQSLBatch pushes MANY ADIF records to eQSL.cc in ONE request.
+//
+// ImportADIF.cfm is a file importer, not a per-QSO endpoint: it takes one *or
+// more* QSOs and answers "Result: X out of Y records added" — the plural in its
+// own reply. So an on-close sweep or a bulk upload is one request, not one per
+// contact. No ADIF header is prepended: the single-record path has always posted
+// bare <EOR> records and eQSL accepts them (per ADIF, a file starting with '<'
+// has no header), and there is no reason to change what is known to work.
+//
+// A PARTIAL result ("97 out of 100") sets Ignored so the caller can say so.
+// eQSL does not identify which records it left out, and in practice they are
+// QSOs it already holds — the same duplicate that UploadEQSL reports as success.
+func UploadEQSLBatch(ctx context.Context, client *http.Client, user, pswd, qthNick string, records []string) (UploadResult, error) {
+	user = strings.ToUpper(strings.TrimSpace(user))
+	if user == "" {
+		return UploadResult{}, fmt.Errorf("eqsl: username (callsign) not set")
+	}
+	if strings.TrimSpace(pswd) == "" {
+		return UploadResult{}, fmt.Errorf("eqsl: password not set")
+	}
+	docs := make([]string, 0, len(records))
+	for _, r := range records {
+		if strings.TrimSpace(r) == "" {
+			continue
+		}
+		docs = append(docs, eqslRecordWithNickname(strings.TrimRight(r, "\r\n"), qthNick))
+	}
+	if len(docs) == 0 {
+		return UploadResult{}, fmt.Errorf("eqsl: empty adif batch")
+	}
+	if len(docs) > eqslBatchMax {
+		return UploadResult{}, fmt.Errorf("eqsl: batch of %d exceeds the %d-record limit", len(docs), eqslBatchMax)
+	}
+
+	body, err := eqslPost(ctx, client, user, pswd, strings.Join(docs, "\n"))
+	if err != nil {
+		return UploadResult{OK: false, Message: body}, err
+	}
+	if reason := authErrEQSL(body); reason != "" {
+		return UploadResult{OK: false, Message: reason}, fmt.Errorf("eqsl: %s", reason)
+	}
+
+	// The counted result is read FIRST here, unlike the single-record path: a
+	// batch reply routinely carries both "Result: 97 out of 100 records added"
+	// and a "Bad record: Duplicate" line for the other three, and matching the
+	// duplicate first would throw away the count that says the rest went in.
+	if m := eqslResultRe.FindStringSubmatch(body); m != nil {
+		added, _ := strconv.Atoi(m[1])
+		total, _ := strconv.Atoi(m[2])
+		if added >= 1 {
+			return UploadResult{OK: true, Message: strings.TrimSpace(m[0]), Ignored: added < total}, nil
+		}
+		// "0 out of N" — nothing added. A re-upload of QSOs eQSL already holds
+		// lands here, and that is not a failure.
+		if strings.Contains(strings.ToLower(body), "duplicate") {
+			return UploadResult{OK: true, Message: "already in logbook"}, nil
+		}
+		reason := eqslReason(body)
+		return UploadResult{OK: false, Message: reason}, fmt.Errorf("eqsl: batch upload failed: %s", reason)
+	}
+	if strings.Contains(strings.ToLower(body), "duplicate") {
+		return UploadResult{OK: true, Message: "already in logbook"}, nil
+	}
+	reason := eqslReason(body)
+	return UploadResult{OK: false, Message: reason}, fmt.Errorf("eqsl: batch upload failed: %s", reason)
+}
+
 // eqslReason trims an eQSL reply to a short human-readable reason: the first
 // "Error:" / "Warning:" / "Bad record:" line if present, else the whole body
 // (capped), else a generic phrase.

internal/extsvc/manager.go

@@ -295,9 +295,10 @@ func (m *Manager) CloseUploadCount() int {
 }
 
 // FlushOnClose uploads every QSO due for an on-close push, scanning the whole
-// logbook (not just this session). Called from the shutdown sequence. QRZ/Club
-// Log go one-by-one (fast HTTP); LoTW is signed and uploaded as a single TQSL
-// batch. Returns the number of QSOs uploaded successfully.
+// logbook (not just this session). Called from the shutdown sequence. QRZ and
+// the rest go one-by-one (fast HTTP, no batch API); LoTW is signed and uploaded
+// as a single TQSL batch, and Club Log goes through its batch endpoint.
+// Returns the number of QSOs uploaded successfully.
 func (m *Manager) FlushOnClose() int {
 	if m.deps.CloseUploadIDs == nil {
 		return 0
@@ -312,41 +313,190 @@ func (m *Manager) FlushOnClose() int {
 		switch svc {
 		case ServiceLoTW:
 			uploaded += m.flushLoTWBatch(ids, cfg.LoTW)
-		case ServiceQRZ:
-			for _, id := range ids {
-				if ok, _ := m.upload(svc, id, cfg.QRZ); ok {
-					uploaded++
-				}
-			}
 		case ServiceClublog:
-			for _, id := range ids {
-				if ok, _ := m.upload(svc, id, cfg.Clublog); ok {
-					uploaded++
-				}
-			}
-		case ServiceHRDLog:
-			for _, id := range ids {
-				if ok, _ := m.upload(svc, id, cfg.HRDLog); ok {
-					uploaded++
-				}
-			}
+			uploaded += m.flushClublogBatch(ids, cfg.Clublog)
 		case ServiceEQSL:
-			for _, id := range ids {
-				if ok, _ := m.upload(svc, id, cfg.EQSL); ok {
-					uploaded++
-				}
-			}
+			uploaded += m.flushEQSLBatch(ids, cfg.EQSL)
+		case ServiceQRZ:
+			uploaded += m.flushOneByOne(svc, ids, cfg.QRZ)
+		case ServiceHRDLog:
+			uploaded += m.flushOneByOne(svc, ids, cfg.HRDLog)
 		case ServiceCloudlog:
-			for _, id := range ids {
-				if ok, _ := m.upload(svc, id, cfg.Cloudlog); ok {
-					uploaded++
-				}
-			}
+			uploaded += m.flushOneByOne(svc, ids, cfg.Cloudlog)
 		}
 	}
 	return uploaded
 }
 
+// uploadPace is the shortest gap between two consecutive single-QSO uploads in
+// an on-close sweep. QRZ, HRDLog and Cloudlog have no batch endpoint — HRDLog's
+// NewEntry.aspx keeps only the first record of a multi-record ADIF — so a sweep
+// of a freshly imported log is unavoidably one request per contact. It does not
+// have to arrive as fast as the link allows, though: that burst is what a
+// service reads as a robot, and what got an operator's IP threatened at Club Log
+// (see flushClublogBatch). The gap costs nothing in practice, since a round trip
+// to any of these already takes longer than it.
+const uploadPace = 200 * time.Millisecond
+
+// flushOneByOne uploads ids one request at a time, paced. For the services that
+// have no batch API; everything else has its own flush<Service>Batch.
+func (m *Manager) flushOneByOne(svc Service, ids []int64, cfg ServiceConfig) int {
+	uploaded := 0
+	for i, id := range ids {
+		if i > 0 {
+			time.Sleep(uploadPace)
+		}
+		if ok, _ := m.upload(svc, id, cfg); ok {
+			uploaded++
+		}
+	}
+	return uploaded
+}
+
+// eqslBatchChunk is how many QSOs go into one ImportADIF.cfm request. eQSL's own
+// limit is ten times this (eqslBatchMax); the smaller chunk keeps one refused
+// record from taking a thousand others down with it, and keeps the form body
+// small enough to be unremarkable.
+const eqslBatchChunk = 100
+
+// flushEQSLBatch uploads the on-close eQSL QSOs through ImportADIF.cfm in
+// batches instead of one request per contact. Same reasoning as
+// flushClublogBatch — eQSL's import endpoint has always taken a whole file, so
+// the one-at-a-time loop was making hundreds of requests it never needed to.
+func (m *Manager) flushEQSLBatch(ids []int64, cfg ServiceConfig) int {
+	uploaded := 0
+	var records []string
+	var kept []int64
+
+	send := func() {
+		if len(records) == 0 {
+			return
+		}
+		// nil client: UploadEQSLBatch then builds one with a 30 s timeout rather
+		// than reusing the 20 s budget of a single realtime QSO.
+		res, err := UploadEQSLBatch(context.Background(), nil, cfg.Username, cfg.Password, cfg.QTHNickname, records)
+		if err != nil || !res.OK {
+			if err == nil {
+				err = errFromResult(res)
+			}
+			m.logf("extsvc: eqsl batch upload (%d QSOs) failed: %v", len(kept), err)
+			if m.deps.NotifyError != nil {
+				m.deps.NotifyError(ServiceEQSL, 0, err)
+			}
+		} else {
+			// res.Ignored means eQSL took the file but left records out. Say the
+			// count out loud: the whole chunk is still marked sent (eQSL never
+			// says WHICH it dropped, and in practice they are QSOs it already
+			// had), so the log line is the only trace of the shortfall.
+			if res.Ignored {
+				m.logf("extsvc: eqsl batch upload PARTIAL (%d QSOs sent) %s", len(kept), res.Message)
+			} else {
+				m.logf("extsvc: eqsl batch upload OK (%d QSOs) %s", len(kept), res.Message)
+			}
+			if m.deps.MarkUploaded != nil {
+				for _, id := range kept {
+					m.deps.MarkUploaded(ServiceEQSL, id, res.LogID)
+				}
+			}
+			uploaded += len(kept)
+		}
+		records = records[:0]
+		kept = kept[:0]
+	}
+
+	for _, id := range ids {
+		if m.deps.ShouldUpload != nil && !m.deps.ShouldUpload(ServiceEQSL, id) {
+			continue
+		}
+		// eQSL keeps the QSO's own station call; the account is identified by the
+		// credentials and the optional QTH nickname — as in upload().
+		rec, ok := m.deps.BuildADIF(id, "")
+		if !ok {
+			continue
+		}
+		records = append(records, rec)
+		kept = append(kept, id)
+		if len(records) >= eqslBatchChunk {
+			send()
+		}
+	}
+	send()
+	return uploaded
+}
+
+// clublogBatchChunk is how many QSOs go into one putlogs.php request. Club Log
+// dedupes server-side, so chunking is not about correctness — it keeps a single
+// malformed record from failing a whole ten-thousand-QSO document, and matches
+// what the QSL Manager's bulk upload already uses.
+const clublogBatchChunk = 100
+
+// flushClublogBatch uploads the on-close Club Log QSOs through the BATCH
+// endpoint (putlogs.php) rather than one realtime.php call each.
+//
+// It used to walk the ids and call UploadClublog per QSO. On-close upload sweeps
+// the WHOLE logbook, so importing an ADIF — or simply switching Club Log on over
+// an existing log — turned one app close into hundreds of realtime.php posts.
+// That endpoint is reserved for an operator logging contacts as they work them,
+// and Club Log blocks the IP of anything that batches through it: an OpsLog user
+// was flagged by G7VJR for 185 QSOs in four minutes, which is this loop, not a
+// pile-up. Batch upload is the mechanism Club Log provides for exactly this.
+func (m *Manager) flushClublogBatch(ids []int64, cfg ServiceConfig) int {
+	uploaded := 0
+	var records []string
+	var kept []int64
+
+	send := func() {
+		if len(records) == 0 {
+			return
+		}
+		// nil client on purpose: UploadClublogADIF then builds one with a 120 s
+		// timeout. m.deps.Client is the 20 s budget of a single realtime QSO,
+		// which a hundred-QSO document on a slow link would blow through.
+		res, err := UploadClublogADIF(context.Background(), nil, cfg, strings.Join(records, "\n"))
+		if err != nil || !res.OK {
+			if err == nil {
+				err = errFromResult(res)
+			}
+			m.logf("extsvc: clublog batch upload (%d QSOs) failed: %v", len(kept), err)
+			if m.deps.NotifyError != nil {
+				m.deps.NotifyError(ServiceClublog, 0, err)
+			}
+		} else {
+			m.logf("extsvc: clublog batch upload OK (%d QSOs) %s", len(kept), res.Message)
+			if m.deps.MarkUploaded != nil {
+				for _, id := range kept {
+					m.deps.MarkUploaded(ServiceClublog, id, res.LogID)
+				}
+			}
+			uploaded += len(kept)
+		}
+		records = records[:0]
+		kept = kept[:0]
+	}
+
+	for _, id := range ids {
+		// Skip QSOs not eligible (already sent). The wrong-logbook guard that
+		// upload() applies per QSO is not repeated here: closeUploadIDs has
+		// already filtered the sweep down to this logbook's callsign.
+		if m.deps.ShouldUpload != nil && !m.deps.ShouldUpload(ServiceClublog, id) {
+			continue
+		}
+		// Club Log takes the logbook callsign as its own form field, so the ADIF
+		// keeps the QSO's own station call (no override) — as in upload().
+		rec, ok := m.deps.BuildADIF(id, "")
+		if !ok {
+			continue
+		}
+		records = append(records, rec)
+		kept = append(kept, id)
+		if len(records) >= clublogBatchChunk {
+			send()
+		}
+	}
+	send()
+	return uploaded
+}
+
 // flushLoTWBatch signs+uploads all queued LoTW QSOs in one TQSL run, then
 // stamps each as uploaded on success.
 func (m *Manager) flushLoTWBatch(ids []int64, cfg ServiceConfig) int {

internal/lookup/QRZ-XML-FIELDS.md

@@ -46,6 +46,51 @@ street, no image.
 </Callsign>
 ```
 
+```
+<QRZDatabase xmlns="http://xmldata.qrz.com" version="1.36">
+<div id="in-page-channel-node-id" data-channel-name="in_page_channel_ogW01U"/>
+<Callsign>
+<call>F5IRH</call>
+<dxcc>227</dxcc>
+<fname>AVRILLON</fname>
+<name>Max</name>
+<addr1>La Grand Prairie</addr1>
+<addr2>Le Palais BELLE-ILE-EN-MER</addr2>
+<zip>56360</zip>
+<country>France</country>
+<lat>47.339686</lat>
+<lon>-3.156500</lon>
+<grid>IN87ki</grid>
+<ccode>97</ccode>
+<land>France</land>
+<codes>TP</codes>
+<qslmgr>VIA BURO</qslmgr>
+<email>f5irh@free.fr</email>
+<u_views>8005</u_views>
+<bio>2463</bio>
+<biodate>2015-07-16 00:29:49</biodate>
+<image>https://cdn-xml.qrz.com/h/f5irh/qsl_F5IRH_111-3.JPG</image>
+<imageinfo>518:799:105108</imageinfo>
+<moddate>2010-08-05 01:05:37</moddate>
+<eqsl>0</eqsl>
+<mqsl>0</mqsl>
+<cqzone>14</cqzone>
+<iota>EU-048</iota>
+<lotw>0</lotw>
+<geoloc>user</geoloc>
+<name_fmt>AVRILLON Max</name_fmt>
+<serial>1570698</serial>
+</Callsign>
+<Session>
+<Key>e5ca5b3e7f88d733408ab7677e605270</Key>
+<Count>161639</Count>
+<SubExp>Sat Jul 3 21:45:38 2027</SubExp>
+<GMTime>Sun Aug 16 21:14:40 2026</GMTime>
+<Remark>cpu: 0.073s</Remark>
+</Session>
+</QRZDatabase>
+```
+
 ## Consequences
 
 - **A free QRZ account can never fill the locator, the coordinates, the zones

internal/lookup/cache_iota_test.go

@@ -0,0 +1,54 @@
+package lookup
+
+import (
+	"context"
+	"path/filepath"
+	"testing"
+
+	"hamlog/internal/db"
+)
+
+func testCache(t *testing.T) *Cache {
+	t.Helper()
+	conn, err := db.Open(filepath.Join(t.TempDir(), "c.db"))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { conn.Close() })
+	return NewCache(conn, 0)
+}
+
+// Adding a field to the cache leaves every row already in it without that field,
+// and the cache lasts thirty days. So a callsign looked up before the change
+// would go a MONTH without its island reference — which is exactly what the
+// first test of the feature ran into: a QRZ record plainly carrying
+// <iota>EU-048</iota>, and no IOTA on the entry.
+//
+// A row that predates the column is therefore treated as stale and refetched
+// once. NULL and "" mean different things here, and that is the whole mechanism.
+func TestCacheRefetchesRowsWrittenBeforeTheIOTAColumn(t *testing.T) {
+	c := testCache(t)
+	ctx := context.Background()
+
+	// An operator with an island: stored and returned.
+	if err := c.Put(ctx, Result{Callsign: "F5IRH", Name: "Max", IOTA: "EU-048", Source: "qrz"}); err != nil {
+		t.Fatalf("put: %v", err)
+	}
+	got, ok := c.Get(ctx, "F5IRH")
+	if !ok || got.IOTA != "EU-048" {
+		t.Fatalf("Get = (%+v,%v), want the island back", got, ok)
+	}
+
+	// An operator with NO island: an empty string is stored, and the row stays
+	// usable. If this wrote NULL, every ordinary callsign would refetch for ever.
+	if err := c.Put(ctx, Result{Callsign: "M0ABC", Name: "Ann", Source: "qrz"}); err != nil {
+		t.Fatalf("put: %v", err)
+	}
+	got, ok = c.Get(ctx, "M0ABC")
+	if !ok {
+		t.Fatal("a callsign with no island was treated as stale — every lookup would repeat for ever")
+	}
+	if got.IOTA != "" {
+		t.Errorf("IOTA = %q for an operator with no island", got.IOTA)
+	}
+}

internal/lookup/cache_off_test.go

@@ -0,0 +1,74 @@
+package lookup
+
+import (
+	"context"
+	"testing"
+	"time"
+)
+
+// A TTL of zero means no cache: nothing is read from it, and nothing is written
+// to it either.
+//
+// It is a real thing to want. An operator correcting their own QRZ record — or
+// chasing a DXpedition whose page changes during the operation — otherwise
+// waits out thirty days before OpsLog will ask again. Clearing the cache by
+// hand works once; switching it off is the setting for a session where the
+// answers are moving.
+func TestATTLOfZeroSwitchesTheCacheOff(t *testing.T) {
+	c := testCache(t)
+	ctx := context.Background()
+
+	if err := c.Put(ctx, Result{Callsign: "M0ABC", Name: "Ann", Source: "qrz"}); err != nil {
+		t.Fatalf("put: %v", err)
+	}
+	if _, ok := c.Get(ctx, "M0ABC"); !ok {
+		t.Fatal("the cache did not hold a fresh entry while switched on")
+	}
+
+	c.SetTTL(0)
+	if c.Enabled() {
+		t.Error("Enabled() is true with a zero TTL")
+	}
+	if _, ok := c.Get(ctx, "M0ABC"); ok {
+		t.Error("a cached entry was still returned with the cache off — the provider would never be asked again")
+	}
+	// And nothing new is stored: those rows would only sit there going stale,
+	// waiting for the day the cache is switched back on.
+	if err := c.Put(ctx, Result{Callsign: "M0XYZ", Name: "Bob", Source: "qrz"}); err != nil {
+		t.Fatalf("put with the cache off: %v", err)
+	}
+	c.SetTTL(30 * 24 * time.Hour)
+	if _, ok := c.Get(ctx, "M0XYZ"); ok {
+		t.Error("a lookup made while the cache was off was written to it anyway")
+	}
+	// The entry from before it was switched off is still there — switching off
+	// is not the same as clearing, and the Clear cache button remains the way to
+	// throw the contents away.
+	if _, ok := c.Get(ctx, "M0ABC"); !ok {
+		t.Error("switching the cache off discarded what it already held")
+	}
+}
+
+// A negative lifetime is meaningless, and rounding it into either "off" or a
+// default would be a guess. It is ignored instead.
+func TestANegativeTTLIsIgnored(t *testing.T) {
+	c := testCache(t)
+	c.SetTTL(7 * 24 * time.Hour)
+	c.SetTTL(-1)
+	if !c.Enabled() {
+		t.Fatal("a negative TTL switched the cache off")
+	}
+	if c.ttl != 7*24*time.Hour {
+		t.Errorf("ttl = %v after a negative value, want the 7 days it already had", c.ttl)
+	}
+}
+
+// The constructor's zero is the DEFAULT, not "off": at startup the settings
+// have not been read, and beginning with no cache would hammer the provider for
+// the first seconds of every launch.
+func TestNewCacheWithZeroStillCaches(t *testing.T) {
+	c := testCache(t) // built with NewCache(conn, 0)
+	if !c.Enabled() {
+		t.Error("a cache built with a zero TTL started switched off")
+	}
+}

internal/lookup/lookup.go

@@ -41,6 +41,10 @@ type Result struct {
 	// along and nothing ever filled it, because no provider mapping read the
 	// field.
 	Web string `json:"web,omitempty"`
+	// IOTA is the island reference (EU-048) for an operator on one. QRZ sends it
+	// and nothing used to read it — and since no live activation feed exists for
+	// IOTA the way it does for POTA, the callbook record is the practical source.
+	IOTA string `json:"iota,omitempty"`
 	// Zip is the postal code. HamQTH and QRZ both send one.
 	Zip       string    `json:"zip,omitempty"`
 	ImageURL  string    `json:"image_url,omitempty"` // profile picture URL
@@ -428,11 +432,21 @@ func fillFromDXCC(r *Result, dxcc DXCCResolver) bool {
 // ----- Cache -----
 
 // Cache is a SQLite-backed cache of lookup results with a TTL.
+//
+// A ttl of zero means NO CACHE: every lookup goes to the provider. That is a
+// real thing to want — an operator correcting their own QRZ record, or chasing
+// a DXpedition whose page changes during the operation, otherwise waits out the
+// cache before OpsLog will look again.
 type Cache struct {
 	db  *sql.DB
 	ttl time.Duration
 }
 
+// NewCache builds the cache. A ttl of zero here is the CONSTRUCTOR default
+// (thirty days), not "off": at startup the settings have not been read yet, and
+// starting with no cache would hammer the provider for the first seconds of
+// every launch. Switching it off is a decision the operator makes, through
+// SetTTL, once their settings are known.
 func NewCache(db *sql.DB, ttl time.Duration) *Cache {
 	if ttl <= 0 {
 		ttl = 30 * 24 * time.Hour
@@ -440,25 +454,35 @@ func NewCache(db *sql.DB, ttl time.Duration) *Cache {
 	return &Cache{db: db, ttl: ttl}
 }
 
-// SetTTL updates the cache TTL (e.g. when user changes settings).
+// SetTTL updates the cache lifetime.
+//
+// ZERO switches the cache OFF — nothing is read from it and nothing is written
+// to it. A NEGATIVE value is meaningless and is ignored, rather than being
+// rounded into one of the two meanings above.
 func (c *Cache) SetTTL(ttl time.Duration) {
-	if ttl > 0 {
+	if ttl >= 0 {
 		c.ttl = ttl
 	}
 }
 
+// Enabled reports whether anything is being cached at all.
+func (c *Cache) Enabled() bool { return c != nil && c.ttl > 0 }
+
 // Get returns the cached result if present and not expired.
 func (c *Cache) Get(ctx context.Context, callsign string) (Result, bool) {
+	if !c.Enabled() {
+		return Result{}, false
+	}
 	row := c.db.QueryRowContext(ctx, `
         SELECT callsign, name, qth, address, state, cnty, country, grid,
                lat, lon, dxcc, cqz, ituz, cont, email, qsl_via, image_url,
-               web, zip, source, fetched_at
+               web, zip, iota, source, fetched_at
         FROM callsign_cache WHERE callsign = ?`, callsign)
 	var (
 		r                                         Result
 		name, qth, addr, state, cnty              sql.NullString
 		country, grid, cont, email, qslVia, image sql.NullString
-		web, zip                                  sql.NullString
+		web, zip, iotaRef                         sql.NullString
 		src                                       string
 		dxcc, cqz, ituz                           sql.NullInt64
 		lat, lon                                  sql.NullFloat64
@@ -466,7 +490,7 @@ func (c *Cache) Get(ctx context.Context, callsign string) (Result, bool) {
 	)
 	if err := row.Scan(&r.Callsign, &name, &qth, &addr, &state, &cnty,
 		&country, &grid, &lat, &lon,
-		&dxcc, &cqz, &ituz, &cont, &email, &qslVia, &image, &web, &zip,
+		&dxcc, &cqz, &ituz, &cont, &email, &qslVia, &image, &web, &zip, &iotaRef,
 		&src, &fetched); err != nil {
 		return Result{}, false
 	}
@@ -477,6 +501,17 @@ func (c *Cache) Get(ctx context.Context, callsign string) (Result, bool) {
 	if time.Since(t) > c.ttl {
 		return Result{}, false
 	}
+	// A row written before the iota column existed has NULL there, and the cache
+	// lasts thirty days — so without this every callsign already looked up would
+	// go a month without its island reference, which is exactly what the first
+	// test of the feature ran into.
+	//
+	// NULL and '' are deliberately different here: Put writes an empty string for
+	// an operator with no island, so only a row that predates the column reads as
+	// invalid. One refetch per such callsign, the next time it is used.
+	if !iotaRef.Valid {
+		return Result{}, false
+	}
 	r.Name = name.String
 	r.QTH = qth.String
 	r.Address = addr.String
@@ -490,6 +525,7 @@ func (c *Cache) Get(ctx context.Context, callsign string) (Result, bool) {
 	r.Email = email.String
 	r.Web = web.String
 	r.Zip = zip.String
+	r.IOTA = strings.ToUpper(iotaRef.String)
 	r.QSLVia = qslVia.String
 	r.ImageURL = image.String
 	r.DXCC = int(dxcc.Int64)
@@ -503,10 +539,15 @@ func (c *Cache) Get(ctx context.Context, callsign string) (Result, bool) {
 // Put upserts a lookup result. fetched_at is generated in Go (NowISO) so the
 // INSERT is backend-agnostic; the conflict tail is dialect-specific.
 func (c *Cache) Put(ctx context.Context, r Result) error {
+	if !c.Enabled() {
+		// Nothing reads it, so writing would only grow the table — and leave
+		// stale rows waiting for the day the cache is switched back on.
+		return nil
+	}
 	updateCols := []string{
 		"name", "qth", "address", "state", "cnty",
 		"country", "grid", "lat", "lon",
-		"dxcc", "cqz", "ituz", "cont", "email", "qsl_via", "image_url", "web", "zip",
+		"dxcc", "cqz", "ituz", "cont", "email", "qsl_via", "image_url", "web", "zip", "iota",
 		"source", "fetched_at",
 	}
 	// The lookup cache always lives in the local SQLite database, so SQLite
@@ -519,8 +560,8 @@ func (c *Cache) Put(ctx context.Context, r Result) error {
         INSERT INTO callsign_cache(callsign, name, qth, address, state, cnty,
                                    country, grid, lat, lon,
                                    dxcc, cqz, ituz, cont, email, qsl_via, image_url,
-                                   web, zip, source, fetched_at)
-        VALUES(?,?,?,?,?,?, ?,?,?,?, ?,?,?,?,?,?,?, ?,?, ?,?)
+                                   web, zip, iota, source, fetched_at)
+        VALUES(?,?,?,?,?,?, ?,?,?,?, ?,?,?,?,?,?,?, ?,?,?, ?,?)
         ON CONFLICT(callsign) DO UPDATE SET ` + strings.Join(sets, ", ")
 	_, err := c.db.ExecContext(ctx, q,
 		r.Callsign, nullable(r.Name), nullable(r.QTH), nullable(r.Address),
@@ -530,6 +571,9 @@ func (c *Cache) Put(ctx context.Context, r Result) error {
 		nullableInt(r.DXCC), nullableInt(r.CQZ), nullableInt(r.ITUZ),
 		nullable(r.Continent), nullable(r.Email), nullable(r.QSLVia),
 		nullable(r.ImageURL), nullable(r.Web), nullable(r.Zip),
+		// NOT nullable(): an operator with no island must store '', so that a NULL
+		// keeps its one meaning — a row written before the column existed.
+		r.IOTA,
 		r.Source, db.NowISO(),
 	)
 	return err

internal/lookup/qrz.go

@@ -134,6 +134,7 @@ func (q *QRZ) fetch(ctx context.Context, sessionKey, callsign string) (Result, e
 		Email:     c.Email,
 		QSLVia:    c.QSLMgr,
 		ImageURL:  strings.TrimSpace(c.Image),
+		IOTA:      strings.ToUpper(strings.TrimSpace(c.IOTA)),
 	}
 	r.Lat, _ = strconv.ParseFloat(c.Lat, 64)
 	r.Lon, _ = strconv.ParseFloat(c.Lon, 64)
@@ -193,6 +194,7 @@ type qrzCallsign struct {
 	Email     string `xml:"email"`
 	QSLMgr    string `xml:"qslmgr"`
 	Image     string `xml:"image"` // direct URL to the profile picture (subscribers only on QRZ)
+	IOTA      string `xml:"iota"`  // island reference for an operator on one, e.g. EU-048
 }
 
 // composeQRZAddress builds a multi-line postal address from QRZ's separate

internal/lookup/qrz_iota_test.go

@@ -0,0 +1,44 @@
+package lookup
+
+import (
+	"encoding/xml"
+	"strings"
+	"testing"
+)
+
+// QRZ sends the island reference and OpsLog read past it.
+//
+// It matters more for IOTA than it would for another award: unlike POTA there
+// is no live "who is on an island right now" feed anywhere, so the callbook
+// record is the practical source — and it is known BEFORE the contact is
+// logged, which is when it is useful.
+func TestQRZCallsignCarriesIOTA(t *testing.T) {
+	const body = `<?xml version="1.0" encoding="utf-8" ?>
+<QRZDatabase version="1.34">
+<Callsign>
+<call>F5IRH</call>
+<fname>Max</fname>
+<country>France</country>
+<grid>IN87ki</grid>
+<cqzone>14</cqzone>
+<iota>EU-048</iota>
+</Callsign>
+<Session><Key>abc</Key></Session>
+</QRZDatabase>`
+
+	var resp qrzDB
+	if err := xml.Unmarshal([]byte(body), &resp); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	if resp.Callsign.IOTA != "EU-048" {
+		t.Fatalf("the <iota> tag decoded as %q — the field is not being read", resp.Callsign.IOTA)
+	}
+	// And a record with no island must not invent one.
+	var none qrzDB
+	if err := xml.Unmarshal([]byte(strings.Replace(body, "<iota>EU-048</iota>", "", 1)), &none); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	if none.Callsign.IOTA != "" {
+		t.Errorf("IOTA = %q for a record without one", none.Callsign.IOTA)
+	}
+}

internal/qso/callmatch_test.go

@@ -2,6 +2,32 @@ package qso
 
 import "testing"
 
+// baseCall picks the operator's OWN callsign out of a portable form. The slash
+// carries a qualifier on either side and which side depends on what it is, so
+// this is the whole difficulty.
+func TestBaseCall(t *testing.T) {
+	for in, want := range map[string]string{
+		"RK3DWA":      "RK3DWA",
+		"RK3DWA/3":    "RK3DWA", // call-area digit
+		"RK3DWA/P":    "RK3DWA", // portable
+		"RK3DWA/QRP":  "RK3DWA",
+		"RK3DWA/MM":   "RK3DWA",
+		"ZA/OE8NDR":   "OE8NDR", // an Austrian in Albania — the case that prompted this
+		"ZA/IZ2DPX":   "IZ2DPX",
+		"F/DL1ABC":    "DL1ABC",
+		"F/DL1ABC/P":  "DL1ABC",
+		"KH6/K6ABC":   "K6ABC",
+		"VP2E/W1ABC":  "W1ABC",
+		"3DA0/ZS1ABC": "ZS1ABC",
+		"9A/S51AB":    "S51AB",
+		"/RK3DWA":     "RK3DWA",
+	} {
+		if got := baseCall(in); got != want {
+			t.Errorf("baseCall(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
 // The predicate behind "Worked before". Exact when folding is off; with it on,
 // a station's portable forms are one operator — and the fold has to work from
 // either end, because you may type the base call or the portable one.
@@ -10,25 +36,61 @@ func TestCallMatch(t *testing.T) {
 		t.Errorf("exact: got %q %v", pred, args)
 	}
 
-	// Typing the base call: match it and everything suffixed off it.
-	pred, args := callMatch("RK3DWA", true)
-	if pred != "(callsign = ? OR callsign LIKE ?)" {
-		t.Errorf("variants predicate = %q", pred)
-	}
-	if len(args) != 2 || args[0] != "RK3DWA" || args[1] != "RK3DWA/%" {
-		t.Errorf("variants args = %v, want [RK3DWA RK3DWA/%%]", args)
-	}
+	// Typing the base call: match it, everything suffixed off it, and every
+	// prefixed form of it.
+	_, args := callMatch("RK3DWA", true)
+	want := []any{"RK3DWA", "RK3DWA/%", "%/RK3DWA", "%/RK3DWA/%"}
+	assertArgs(t, "base call", args, want)
 
-	// Typing a portable form must reach the plain call too — the suffix is
+	// Typing a portable form must reach the plain call too — the qualifier is
 	// stripped from the INPUT, not just matched in the column.
 	_, args = callMatch("RK3DWA/3", true)
-	if len(args) != 2 || args[0] != "RK3DWA" || args[1] != "RK3DWA/%" {
-		t.Errorf("portable input args = %v, want [RK3DWA RK3DWA/%%]", args)
+	assertArgs(t, "portable input", args, want)
+}
+
+// A prefixed call is the same operator, and NOT every other visitor to that
+// country.
+//
+// Matching on the part before the slash made ZA/OE8NDR the station "ZA", so the
+// worked-before list for one Austrian operating from Albania showed every other
+// ZA/ guest — two Italians — and counted them as four contacts "with this
+// call". Reported from a screenshot of exactly that.
+func TestCallMatchPrefixedIsTheOperatorNotTheCountry(t *testing.T) {
+	_, args := callMatch("ZA/OE8NDR", true)
+	assertArgs(t, "ZA/OE8NDR", args, []any{"OE8NDR", "OE8NDR/%", "%/OE8NDR", "%/OE8NDR/%"})
+
+	for _, a := range args {
+		if s, _ := a.(string); s == "ZA" || s == "ZA/%" {
+			t.Fatalf("the country prefix is still being matched as a station: %v", args)
+		}
 	}
 
-	// A leading slash is not a suffix marker — dropping to "" there would match
-	// the entire logbook.
-	if _, args := callMatch("/RK3DWA", true); args[0] != "/RK3DWA" {
-		t.Errorf("leading slash: args[0] = %v, want the call unchanged", args[0])
+	// And the two are not each other: nothing in one operator's predicate can
+	// select the other's call.
+	_, other := callMatch("ZA/IZ2DPX", true)
+	assertArgs(t, "ZA/IZ2DPX", other, []any{"IZ2DPX", "IZ2DPX/%", "%/IZ2DPX", "%/IZ2DPX/%"})
+}
+
+// A base too short to be a callsign falls back to an exact match. "F/DL1ABC"
+// resolves fine, but a malformed entry must never produce a LIKE that selects
+// half the logbook.
+func TestCallMatchRefusesAShortBase(t *testing.T) {
+	for _, call := range []string{"F/", "9A", "/P", "K/M"} {
+		pred, args := callMatch(call, true)
+		if pred != "callsign = ?" || len(args) != 1 {
+			t.Errorf("%q produced %q %v — want an exact match", call, pred, args)
+		}
+	}
+}
+
+func assertArgs(t *testing.T, what string, got, want []any) {
+	t.Helper()
+	if len(got) != len(want) {
+		t.Fatalf("%s: args %v, want %v", what, got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("%s: args %v, want %v", what, got, want)
+		}
 	}
 }

internal/qso/qso.go

@@ -215,6 +215,12 @@ type QSO struct {
 	// that doesn't know about it can't clobber it.
 	AwardRefs string `json:"award_refs,omitempty"`
 
+	// SyncUID is this contact's stable identity for folder synchronisation —
+	// what lets another machine name the same QSO when it edits or deletes it.
+	// Like AwardRefs it is read here but NOT in columnList, so an ordinary edit
+	// can never clobber it; it is written only by SetSyncUID / BackfillSyncUIDs.
+	SyncUID string `json:"sync_uid,omitempty"`
+
 	CreatedAt time.Time `json:"created_at"`
 	UpdatedAt time.Time `json:"updated_at"`
 }
@@ -271,7 +277,7 @@ const columnList = `callsign, qso_date, qso_date_off, band, band_rx, mode, submo
 // award_refs is read here but is NOT part of columnList (the insert/update
 // write path) — it is a derived cache written only via SetAwardRefs, so a
 // normal QSO write can never clobber it.
-const selectCols = `id, ` + columnList + `, award_refs, created_at, updated_at`
+const selectCols = `id, ` + columnList + `, award_refs, sync_uid, created_at, updated_at`
 
 // columnCount is derived from columnList at init so they can never drift.
 var columnCount = countColumns(columnList)
@@ -1723,26 +1729,77 @@ type BandMode struct {
 // rendering a recent-contacts mini-list.
 const maxWorkedEntries = 50
 
+// operatorSuffixes are the appendages that qualify a callsign without changing
+// whose it is. A bare digit (RK3DWA/3) counts too, handled separately.
+var operatorSuffixes = map[string]bool{
+	"P": true, "M": true, "MM": true, "AM": true, "QRP": true,
+	"A": true, "B": true, "J": true, "LH": true, "R": true, "T": true,
+}
+
+// baseCall extracts the operator's OWN callsign from a portable form.
+//
+// The slash carries the qualifier on either side, and which side depends on
+// what it is: RK3DWA/3 and RK3DWA/P append to the call, while ZA/OE8NDR and
+// F/DL1ABC put a country prefix in front of it. Taking the part before the
+// slash — which is what this used to do — reads ZA/OE8NDR as the station "ZA".
+//
+// So: drop the known qualifiers, and of what is left take the longest part. A
+// prefix is short by nature (ZA, F, KH6, VP2E) and a callsign is not.
+func baseCall(call string) string {
+	if !strings.Contains(call, "/") {
+		return call
+	}
+	best := ""
+	for _, p := range strings.Split(call, "/") {
+		if p == "" || operatorSuffixes[p] {
+			continue
+		}
+		if len(p) == 1 && p[0] >= '0' && p[0] <= '9' {
+			continue // the call-area digit
+		}
+		if len(p) > len(best) {
+			best = p
+		}
+	}
+	if best == "" {
+		return call
+	}
+	return best
+}
+
 // callMatch builds the WHERE fragment that selects one station's QSOs.
 //
 // Exact by default. With variants on, an operator's portable forms count as the
 // same station: RK3DWA, RK3DWA/3, RK3DWA/P and RK3DWA/QRP are one person, and
 // someone asking "have I worked RK3DWA?" means the person, not the string.
 // Typing 21 QSOs' worth of history only when you happen to add "/3" is the
-// behaviour this replaces. The suffix is stripped from what was TYPED too, so
-// it matches both ways round — RK3DWA/3 also finds the plain RK3DWA contacts.
+// behaviour this replaces.
+//
+// PREFIXED FORMS MATCH TOO, and getting that wrong is what prompted this.
+// Matching on the part before the slash turned ZA/OE8NDR into the station "ZA",
+// so the worked-before list for one Austrian operating from Albania showed
+// every OTHER visitor to Albania — two Italians and himself — and counted them
+// as four contacts "with this call".
 //
 // Deliberately NOT a bare "starts with": LIKE 'RK3DWA%' would also drag in
 // RK3DWAB, which is a different station. The '/' is what makes it the same one.
+// And the variant match is only used on a base of three characters or more —
+// below that a prefix could match half the log.
 func callMatch(call string, variants bool) (string, []any) {
 	if !variants {
 		return "callsign = ?", []any{call}
 	}
-	base := call
-	if i := strings.IndexByte(base, '/'); i > 0 {
-		base = base[:i]
+	base := baseCall(call)
+	if len(base) < 3 {
+		return "callsign = ?", []any{call}
 	}
-	return "(callsign = ? OR callsign LIKE ?)", []any{base, base + "/%"}
+	return "(callsign = ? OR callsign LIKE ? OR callsign LIKE ? OR callsign LIKE ?)",
+		[]any{
+			base,               // OE8NDR
+			base + "/%",        // OE8NDR/P
+			"%/" + base,        // ZA/OE8NDR
+			"%/" + base + "/%", // ZA/OE8NDR/P
+		}
 }
 
 // WorkedBefore returns aggregated history at both callsign and DXCC level.
@@ -3025,6 +3082,7 @@ func scanQSO(s scanner) (QSO, error) {
 		myARRLSect, myVUCCGrids                sql.NullString
 		extrasJSON                             sql.NullString
 		awardRefs                              sql.NullString
+		syncUID                                sql.NullString
 		createdStr, updatedStr                 string
 	)
 	if err := s.Scan(
@@ -3052,7 +3110,7 @@ func scanQSO(s scanner) (QSO, error) {
 		&skcc, &fists, &tenTen, &contactedOp, &eqCall, &pfx, &myName, &class,
 		&darcDOK, &myDarcDOK, &region, &silentKey, &swl, &qsoComplete, &qsoRandom,
 		&creditGranted, &creditSubmitted, &myARRLSect, &myVUCCGrids,
-		&extrasJSON, &awardRefs, &createdStr, &updatedStr,
+		&extrasJSON, &awardRefs, &syncUID, &createdStr, &updatedStr,
 	); err != nil {
 		return QSO{}, fmt.Errorf("scan qso: %w", err)
 	}
@@ -3252,6 +3310,7 @@ func scanQSO(s scanner) (QSO, error) {
 	q.MyVUCCGrids = myVUCCGrids.String
 	q.Extras = decodeExtras(extrasJSON.String)
 	q.AwardRefs = awardRefs.String
+	q.SyncUID = syncUID.String
 	return q, nil
 }
 
@@ -3362,3 +3421,67 @@ func (r *Repo) OrderedIDs(ctx context.Context) ([]int64, time.Time, error) {
 	// Parsed once, for the newest row only — the ordering came from SQL.
 	return out, parseTimeLoose(lastDate), nil
 }
+
+// --- Folder synchronisation identity -----------------------------------------
+
+// SetSyncUID stamps a contact's sync identity. Targeted UPDATE: it must never
+// go through the normal write path, or an ordinary edit would clobber it.
+func (r *Repo) SetSyncUID(ctx context.Context, id int64, uid string) error {
+	_, err := r.db.ExecContext(ctx, `UPDATE qso SET sync_uid = ? WHERE id = ?`, uid, id)
+	return err
+}
+
+// IDBySyncUID resolves an incoming change to a local row. Indexed, so this is
+// the one lookup a sync performs per record and it stays a key hit rather than
+// a scan — which is the whole reason sync_uid is a column and not a JSON key.
+func (r *Repo) IDBySyncUID(ctx context.Context, uid string) (int64, bool, error) {
+	if uid == "" {
+		return 0, false, nil
+	}
+	var id int64
+	err := r.db.QueryRowContext(ctx, `SELECT id FROM qso WHERE sync_uid = ? LIMIT 1`, uid).Scan(&id)
+	if err == sql.ErrNoRows {
+		return 0, false, nil
+	}
+	if err != nil {
+		return 0, false, err
+	}
+	return id, true, nil
+}
+
+// IDByDedupeKey finds the local row a change refers to when its identity is
+// unknown here.
+//
+// The change log starts EMPTY, so contacts logged before synchronisation was
+// switched on are never exchanged — nothing has to be copied across, and
+// identities are stamped lazily on the contacts that are actually touched.
+//
+// One case still needs this. Two machines can already hold the SAME old contact
+// — the second was seeded by copying the database or importing an ADIF — with
+// different row ids and no identity on either. The day one of them edits that
+// contact, it stamps an identity and sends a change naming it; the other has
+// never seen that identity and would insert a duplicate. Matching on the
+// contact itself (callsign + minute + band + mode, the importer's own dedupe
+// key) recognises it instead.
+//
+// Looked up only when an identity is unknown, which is rare, and it goes
+// through idx_qso_callsign rather than scanning — a full map of every contact,
+// rebuilt each pass, was the wrong shape for something this occasional.
+func (r *Repo) IDByDedupeKey(ctx context.Context, callsign, qsoDateMinute, band, mode string) (id int64, uid string, found bool, err error) {
+	if callsign == "" || qsoDateMinute == "" {
+		return 0, "", false, nil
+	}
+	var u sql.NullString
+	err = r.db.QueryRowContext(ctx, `
+		SELECT id, sync_uid FROM qso
+		 WHERE callsign = ? AND substr(qso_date, 1, 16) = ? AND band = ? AND mode = ?
+		 LIMIT 1`,
+		strings.ToUpper(callsign), qsoDateMinute, band, mode).Scan(&id, &u)
+	if err == sql.ErrNoRows {
+		return 0, "", false, nil
+	}
+	if err != nil {
+		return 0, "", false, err
+	}
+	return id, u.String, true, nil
+}

internal/qso/stats_test.go

@@ -70,12 +70,12 @@ func TestStatsNoNilSlices(t *testing.T) {
 }
 
 // Contest metrics over a window. The two traps:
-//   1. "Best hour" must be the best ROLLING 60 minutes, not the best clock hour —
-//      a run straddling 13:45–14:45 is invisible to clock-hour bucketing, and the
-//      rolling figure is the one contesters quote.
-//   2. Both rates must be reported: QSOs ÷ whole window (honest, breaks included)
-//      AND QSOs ÷ hours actually operated. Quoting only the latter is how an
-//      8-hour effort gets sold as a 48-hour score.
+//  1. "Best hour" must be the best ROLLING 60 minutes, not the best clock hour —
+//     a run straddling 13:45–14:45 is invisible to clock-hour bucketing, and the
+//     rolling figure is the one contesters quote.
+//  2. Both rates must be reported: QSOs ÷ whole window (honest, breaks included)
+//     AND QSOs ÷ hours actually operated. Quoting only the latter is how an
+//     8-hour effort gets sold as a 48-hour score.
 func TestContestPeriodMetrics(t *testing.T) {
 	base := time.Date(2026, 5, 30, 12, 0, 0, 0, time.UTC)
 	at := func(min int) time.Time { return base.Add(time.Duration(min) * time.Minute) }
@@ -90,8 +90,8 @@ func TestContestPeriodMetrics(t *testing.T) {
 		times = append(times, entry{t: at(240 + i*5), op: "F5XYZ"}) // 16:00 …
 	}
 
-	from := base                       // 12:00
-	to := base.Add(6 * time.Hour)      // 18:00 → a 6-hour window
+	from := base                  // 12:00
+	to := base.Add(6 * time.Hour) // 18:00 → a 6-hour window
 	var s Stats
 	s.periodMetrics(times, from, to, time.Time{}, time.Time{})
 

internal/qso/syncuid_test.go

@@ -0,0 +1,145 @@
+package qso
+
+import (
+	"context"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"hamlog/internal/db"
+)
+
+// openRepo gives a migrated, empty logbook on disk.
+func openRepo(t *testing.T) *Repo {
+	t.Helper()
+	conn, err := db.Open(filepath.Join(t.TempDir(), "test.db"))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { conn.Close() })
+	return NewRepo(conn)
+}
+
+func addQSO(t *testing.T, r *Repo, call string, when time.Time) int64 {
+	t.Helper()
+	id, err := r.Add(context.Background(), QSO{
+		Callsign: call, QSODate: when, Band: "20m", Mode: "CW",
+	})
+	if err != nil {
+		t.Fatalf("insert %s: %v", call, err)
+	}
+	return id
+}
+
+// A QSO written and read back must come out whole. selectCols and scanQSO are
+// two hand-maintained lists that must line up column for column, and adding
+// sync_uid touched both — a drift there fails every read at runtime, which no
+// compiler catches.
+func TestSyncUIDRoundTrips(t *testing.T) {
+	r := openRepo(t)
+	ctx := context.Background()
+	id := addQSO(t, r, "M0ABC", time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC))
+
+	got, err := r.GetByID(ctx, id)
+	if err != nil {
+		t.Fatalf("read back: %v", err)
+	}
+	if got.Callsign != "M0ABC" {
+		t.Fatalf("read back %+v", got)
+	}
+	if got.SyncUID != "" {
+		t.Errorf("a fresh QSO has identity %q — it should have none until sync is switched on", got.SyncUID)
+	}
+
+	if err := r.SetSyncUID(ctx, id, "abc123"); err != nil {
+		t.Fatalf("SetSyncUID: %v", err)
+	}
+	got, _ = r.GetByID(ctx, id)
+	if got.SyncUID != "abc123" {
+		t.Errorf("SyncUID = %q after stamping", got.SyncUID)
+	}
+}
+
+// An ordinary edit must NEVER clobber the identity. sync_uid is deliberately
+// outside columnList for this reason: another machine that has already seen the
+// contact addresses it by that id, and losing it makes the same QSO arrive
+// again as a new one.
+func TestAnEditDoesNotClobberTheIdentity(t *testing.T) {
+	r := openRepo(t)
+	ctx := context.Background()
+	id := addQSO(t, r, "M0ABC", time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC))
+	if err := r.SetSyncUID(ctx, id, "keepme"); err != nil {
+		t.Fatal(err)
+	}
+
+	q, _ := r.GetByID(ctx, id)
+	q.Name = "Edited"
+	q.SyncUID = "" // exactly what a caller that knows nothing about sync sends
+	if err := r.Update(ctx, q); err != nil {
+		t.Fatalf("update: %v", err)
+	}
+
+	got, _ := r.GetByID(ctx, id)
+	if got.Name != "Edited" {
+		t.Errorf("the edit did not take: %+v", got)
+	}
+	if got.SyncUID != "keepme" {
+		t.Errorf("SyncUID = %q — an edit wiped the sync identity", got.SyncUID)
+	}
+}
+
+// The lookup an incoming change performs, once per record.
+func TestIDBySyncUID(t *testing.T) {
+	r := openRepo(t)
+	ctx := context.Background()
+	id := addQSO(t, r, "M0ABC", time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC))
+	_ = r.SetSyncUID(ctx, id, "u-1")
+
+	got, ok, err := r.IDBySyncUID(ctx, "u-1")
+	if err != nil || !ok || got != id {
+		t.Fatalf("IDBySyncUID = (%d,%v,%v), want (%d,true,nil)", got, ok, err, id)
+	}
+	if _, ok, _ := r.IDBySyncUID(ctx, "nope"); ok {
+		t.Error("an unknown identity was resolved")
+	}
+	// An empty id must never match the rows that have none.
+	if _, ok, _ := r.IDBySyncUID(ctx, ""); ok {
+		t.Error("the empty identity matched a row — every un-stamped QSO would be that row")
+	}
+}
+
+// Two machines can already hold the SAME old contact — the second was seeded by
+// copying the database or importing an ADIF — with different row ids and no
+// identity on either. The day one of them edits it, it stamps an identity and
+// sends a change naming it; the other has never seen that identity and would
+// insert a duplicate. Matching on the contact itself recognises it instead.
+func TestIDByDedupeKeyRecognisesTheSameContact(t *testing.T) {
+	r := openRepo(t)
+	ctx := context.Background()
+	when := time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC)
+	id := addQSO(t, r, "M0ABC", when)
+
+	minute := when.UTC().Format("2006-01-02T15:04")
+	gotID, uid, found, err := r.IDByDedupeKey(ctx, "M0ABC", minute, "20m", "CW")
+	if err != nil || !found {
+		t.Fatalf("IDByDedupeKey = (%d,%q,%v,%v), want it found", gotID, uid, found, err)
+	}
+	if gotID != id {
+		t.Errorf("resolved to id %d, want %d", gotID, id)
+	}
+	if uid != "" {
+		t.Errorf("uid = %q, want empty until something stamps it", uid)
+	}
+
+	// A contact this machine does not have must NOT match something else.
+	if _, _, found, _ := r.IDByDedupeKey(ctx, "M0ABC", minute, "40m", "CW"); found {
+		t.Error("a different band matched — the key must be all four parts")
+	}
+	if _, _, found, _ := r.IDByDedupeKey(ctx, "M0XYZ", minute, "20m", "CW"); found {
+		t.Error("a different callsign matched")
+	}
+	// An empty key must never match anything.
+	if _, _, found, _ := r.IDByDedupeKey(ctx, "", "", "", ""); found {
+		t.Error("an empty key matched a row")
+	}
+}

internal/relaydev/denkovi_other.go

@@ -20,8 +20,8 @@ func NewDenkovi(serial string, count int) Device {
 	return denkoviStub{count: count}
 }
 
-func (s denkoviStub) Count() int     { return s.count }
-func (denkoviStub) Close() error     { return nil }
+func (s denkoviStub) Count() int { return s.count }
+func (denkoviStub) Close() error { return nil }
 func (denkoviStub) Status(context.Context) ([]bool, error) {
 	return nil, fmt.Errorf("Denkovi USB relay board is only supported on Windows")
 }

internal/relaydev/httpgen.go

@@ -56,6 +56,9 @@ type httpGen struct {
 	user    string
 	pass    string
 	count   int
+	// insecure accepts a certificate nothing can verify — the self-signed one a
+	// relay board on the LAN presents. Per board, and the operator's choice.
+	insecure bool
 
 	mu    sync.Mutex
 	state []bool
@@ -64,7 +67,7 @@ type httpGen struct {
 // NewHTTPGeneric builds the driver. onURLs/offURLs are per relay (index 0 =
 // relay 1) and may be short or hold empty entries; onPat/offPat are the
 // fallback patterns; labels are the relay names {value} substitutes.
-func NewHTTPGeneric(onURLs, offURLs []string, onPat, offPat, user, pass string, count int, labels []string) Device {
+func NewHTTPGeneric(onURLs, offURLs []string, onPat, offPat, user, pass string, count int, labels []string, insecure bool) Device {
 	if count <= 0 {
 		count = len(onURLs)
 	}
@@ -74,7 +77,7 @@ func NewHTTPGeneric(onURLs, offURLs []string, onPat, offPat, user, pass string,
 	return &httpGen{
 		onURLs: onURLs, offURLs: offURLs,
 		onPat: onPat, offPat: offPat, labels: labels,
-		user: user, pass: pass, count: count,
+		user: user, pass: pass, count: count, insecure: insecure,
 		state: make([]bool, count),
 	}
 }
@@ -196,7 +199,7 @@ func (h *httpGen) Set(ctx context.Context, relay int, on bool) error {
 	}
 	u := h.urlFor(relay, on)
 	u = withScheme(u)
-	if _, err := get(ctx, u, h.user, h.pass); err != nil {
+	if _, err := get(ctx, u, h.user, h.pass, h.insecure); err != nil {
 		return err
 	}
 	h.mu.Lock()

internal/relaydev/httpgen_test.go

@@ -22,7 +22,7 @@ func TestHTTPGenericPattern(t *testing.T) {
 
 	d := NewHTTPGeneric(nil, nil,
 		srv.URL+"/relay?n={relay}&state=on",
-		srv.URL+"/relay?n={relay}&state=off", "", "", 4, nil)
+		srv.URL+"/relay?n={relay}&state=off", "", "", 4, nil, false)
 	if err := d.Set(context.Background(), 2, true); err != nil {
 		t.Fatalf("Set on: %v", err)
 	}
@@ -53,7 +53,7 @@ func TestHTTPGenericPerRelayURLsWinOverThePattern(t *testing.T) {
 	d := NewHTTPGeneric(
 		[]string{srv.URL + "/FF0101", "", srv.URL + "/weird/on"},
 		[]string{srv.URL + "/FF0100", "", ""},
-		srv.URL+"/pattern/on/{relay}", srv.URL+"/pattern/off/{relay}", "", "", 3, nil)
+		srv.URL+"/pattern/on/{relay}", srv.URL+"/pattern/off/{relay}", "", "", 3, nil, false)
 
 	_ = d.Set(context.Background(), 1, true)  // its own URL
 	_ = d.Set(context.Background(), 2, true)  // empty → falls back to the pattern
@@ -82,7 +82,7 @@ func TestHTTPGenericValueIsTheRelayLabel(t *testing.T) {
 		[]string{srv.URL + "/relay?on={value}"}, // per-relay URL
 		nil,
 		"", srv.URL+"/relay?off={value}", // and the pattern, for the other direction
-		"", "", 3, []string{"Ant1", "Beam 20m", ""})
+		"", "", 3, []string{"Ant1", "Beam 20m", ""}, false)
 	_ = d.Set(context.Background(), 1, true)
 	_ = d.Set(context.Background(), 2, false)
 	mu.Lock()
@@ -107,7 +107,7 @@ func TestHTTPGenericRelayOffset(t *testing.T) {
 	defer srv.Close()
 
 	d := NewHTTPGeneric(nil, nil,
-		srv.URL+"/set0/{relay-1}/1", srv.URL+"/set0/{relay-1}/0", "", "", 4, nil)
+		srv.URL+"/set0/{relay-1}/1", srv.URL+"/set0/{relay-1}/0", "", "", 4, nil, false)
 	_ = d.Set(context.Background(), 1, true)
 	_ = d.Set(context.Background(), 4, false)
 	mu.Lock()
@@ -123,7 +123,7 @@ func TestHTTPGenericRelayOffset(t *testing.T) {
 // movement. It must be refused, and the message must say the label is what is
 // missing.
 func TestHTTPGenericRefusesValueWithoutALabel(t *testing.T) {
-	d := NewHTTPGeneric(nil, nil, "http://x/relay?on={value}", "", "", "", 2, []string{"", ""})
+	d := NewHTTPGeneric(nil, nil, "http://x/relay?on={value}", "", "", "", 2, []string{"", ""}, false)
 	err := d.Set(context.Background(), 1, true)
 	if err == nil || !strings.Contains(err.Error(), "label") {
 		t.Errorf("err = %v, want it to name the missing label", err)
@@ -149,7 +149,7 @@ func TestHTTPGenericSuppliesTheScheme(t *testing.T) {
 // A switch with the ON URLs filled and OFF left empty latches. The error has to
 // name the direction, or the operator cannot tell which half is missing.
 func TestHTTPGenericNamesTheMissingDirection(t *testing.T) {
-	d := NewHTTPGeneric([]string{"http://x/on"}, nil, "", "", "", "", 1, nil)
+	d := NewHTTPGeneric([]string{"http://x/on"}, nil, "", "", "", "", 1, nil, false)
 	err := d.Set(context.Background(), 1, false)
 	if err == nil || !strings.Contains(err.Error(), "OFF") {
 		t.Errorf("err = %v, want it to name the OFF direction", err)
@@ -160,7 +160,7 @@ func TestHTTPGenericNamesTheMissingDirection(t *testing.T) {
 func TestHTTPGenericRemembersWhatItCommanded(t *testing.T) {
 	srv := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {}))
 	defer srv.Close()
-	d := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", srv.URL+"/off/{relay}", "", "", 3, nil)
+	d := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", srv.URL+"/off/{relay}", "", "", 3, nil, false)
 	_ = d.Set(context.Background(), 2, true)
 	st, err := d.Status(context.Background())
 	if err != nil {

internal/relaydev/httpstls_test.go

@@ -0,0 +1,80 @@
+package relaydev
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+)
+
+// A relay board on the LAN signs its own certificate — there is no authority
+// anywhere that could have signed it. httptest.NewTLSServer presents exactly
+// that: a certificate from an unknown issuer, which is what the hardware does.
+func selfSignedRelay(t *testing.T) (*httptest.Server, func() []string) {
+	t.Helper()
+	var mu sync.Mutex
+	var got []string
+	srv := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		mu.Lock()
+		got = append(got, r.URL.Path)
+		mu.Unlock()
+		w.WriteHeader(http.StatusOK)
+	}))
+	t.Cleanup(srv.Close)
+	return srv, func() []string {
+		mu.Lock()
+		defer mu.Unlock()
+		return append([]string(nil), got...)
+	}
+}
+
+// With the box ticked, the board answers.
+func TestHTTPSRelayWithASelfSignedCertificate(t *testing.T) {
+	srv, seen := selfSignedRelay(t)
+	d := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", srv.URL+"/off/{relay}", "", "", 2, nil, true)
+	if err := d.Set(context.Background(), 1, true); err != nil {
+		t.Fatalf("Set over HTTPS: %v", err)
+	}
+	if paths := seen(); len(paths) != 1 || paths[0] != "/on/1" {
+		t.Errorf("the board was asked for %v, want /on/1", paths)
+	}
+}
+
+// Without it, the request is refused — and the refusal has to name the box.
+//
+// Go's own message, "x509: certificate signed by unknown authority", is
+// accurate and tells an operator nothing about what to do next. This is the
+// difference between a dead end and an instruction, and it is the whole reason
+// the default can safely stay OFF.
+func TestARefusedCertificateNamesTheSetting(t *testing.T) {
+	srv, seen := selfSignedRelay(t)
+	d := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", srv.URL+"/off/{relay}", "", "", 2, nil, false)
+	err := d.Set(context.Background(), 1, true)
+	if err == nil {
+		t.Fatal("an unverifiable certificate was accepted with the box unticked")
+	}
+	if !strings.Contains(err.Error(), "self-signed") {
+		t.Errorf("the refusal reads %q — it does not say which setting to change", err)
+	}
+	if len(seen()) != 0 {
+		t.Error("the request reached the board despite the certificate being refused")
+	}
+}
+
+// The box belongs to ONE board. An operator with a self-signed switch on the
+// LAN and a second board reached through a proper HTTPS proxy must keep real
+// verification on the second — that link crosses the internet, and it commands
+// an antenna.
+func TestAcceptingOneBoardsCertificateDoesNotAffectAnother(t *testing.T) {
+	srv, _ := selfSignedRelay(t)
+	lan := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", "", "", "", 1, nil, true)
+	if err := lan.Set(context.Background(), 1, true); err != nil {
+		t.Fatalf("the LAN board: %v", err)
+	}
+	strict := NewHTTPGeneric(nil, nil, srv.URL+"/on/{relay}", "", "", "", 1, nil, false)
+	if err := strict.Set(context.Background(), 1, true); err == nil {
+		t.Error("the second board accepted the certificate too — the setting is not per board")
+	}
+}

internal/relaydev/relaydev.go

@@ -17,7 +17,10 @@ package relaydev
 
 import (
 	"context"
+	"crypto/tls"
+	"crypto/x509"
 	"encoding/xml"
+	"errors"
 	"fmt"
 	"io"
 	"net/http"
@@ -28,8 +31,8 @@ import (
 
 // Device is one relay board.
 type Device interface {
-	Count() int                                       // number of user-controllable relays
-	Status(ctx context.Context) ([]bool, error)       // state of each relay (index 0 = relay 1)
+	Count() int                                        // number of user-controllable relays
+	Status(ctx context.Context) ([]bool, error)        // state of each relay (index 0 = relay 1)
 	Set(ctx context.Context, relay int, on bool) error // relay is 1-based
 	// Close releases any OS handle the driver holds (serial port, FTDI handle).
 	// Network boards hold nothing and no-op. MUST be called when a cached driver is
@@ -40,8 +43,47 @@ type Device interface {
 
 func httpClient() *http.Client { return &http.Client{Timeout: 5 * time.Second} }
 
+// insecureClient talks to a board presenting a certificate nothing can verify.
+//
+// Which is nearly every board that offers HTTPS at all: a relay box on the LAN
+// signs its own certificate, and there is no authority anywhere that could have
+// signed it. Refusing that means refusing HTTPS on the hardware, which is not a
+// security decision, only an outcome.
+//
+// So it is offered, per board, and OFF by default — because the other HTTPS
+// case is real and opposite: a board reached from outside through a proxy with
+// a genuine certificate, where verification is the only thing standing between
+// an antenna switch and the internet. One box, on the board that needs it.
+//
+// Built once. A Transport per request would open a fresh TLS connection every
+// time and never reuse one.
+var insecureClient = &http.Client{
+	Timeout: 5 * time.Second,
+	Transport: &http.Transport{
+		TLSClientConfig: &tls.Config{InsecureSkipVerify: true}, //nolint:gosec // the operator ticked the box for this board
+	},
+}
+
+// certError says which box to tick when TLS is what failed.
+//
+// Go's own message — "x509: certificate signed by unknown authority" — is
+// accurate and tells an operator nothing about what to do next. Naming the
+// setting turns a dead end into an instruction.
+func certError(err error) error {
+	var unknown x509.UnknownAuthorityError
+	var host x509.HostnameError
+	var verify *tls.CertificateVerificationError
+	if errors.As(err, &unknown) || errors.As(err, &host) || errors.As(err, &verify) {
+		return fmt.Errorf("%w — the board's HTTPS certificate cannot be verified; "+
+			"tick \"Accept a self-signed certificate\" for this board if it is on your own network", err)
+	}
+	return err
+}
+
 // get issues a GET with optional basic auth and returns the body on 2xx.
-func get(ctx context.Context, url, user, pass string) ([]byte, error) {
+//
+// insecure skips certificate verification, for a board that signs its own.
+func get(ctx context.Context, url, user, pass string, insecure bool) ([]byte, error) {
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
 	if err != nil {
 		return nil, err
@@ -49,9 +91,13 @@ func get(ctx context.Context, url, user, pass string) ([]byte, error) {
 	if user != "" || pass != "" {
 		req.SetBasicAuth(user, pass)
 	}
-	resp, err := httpClient().Do(req)
+	client := httpClient()
+	if insecure {
+		client = insecureClient
+	}
+	resp, err := client.Do(req)
 	if err != nil {
-		return nil, err
+		return nil, certError(err)
 	}
 	defer resp.Body.Close()
 	body, _ := io.ReadAll(resp.Body)
@@ -87,8 +133,8 @@ type webswitch struct {
 // NewWebswitch builds a WebSwitch 1216H client (5 relays).
 func NewWebswitch(host string) Device { return &webswitch{host: host, count: 5} }
 
-func (w *webswitch) Count() int    { return w.count }
-func (w *webswitch) Close() error  { return nil } // stateless HTTP, nothing to release
+func (w *webswitch) Count() int   { return w.count }
+func (w *webswitch) Close() error { return nil } // stateless HTTP, nothing to release
 
 func (w *webswitch) Set(ctx context.Context, relay int, on bool) error {
 	if relay < 1 || relay > w.count {
@@ -98,7 +144,7 @@ func (w *webswitch) Set(ctx context.Context, relay int, on bool) error {
 	if on {
 		action = "on"
 	}
-	_, err := get(ctx, fmt.Sprintf("%s/relaycontrol/%s/%d", relayBase(w.host), action, relay), "", "")
+	_, err := get(ctx, fmt.Sprintf("%s/relaycontrol/%s/%d", relayBase(w.host), action, relay), "", "", false)
 	return err
 }
 
@@ -109,7 +155,7 @@ func (w *webswitch) Status(ctx context.Context) ([]bool, error) {
 		sel.WriteString(strconv.Itoa(i))
 		sel.WriteByte('$')
 	}
-	body, err := get(ctx, fmt.Sprintf("%s/relaystate/get2/%s", relayBase(w.host), sel.String()), "", "")
+	body, err := get(ctx, fmt.Sprintf("%s/relaystate/get2/%s", relayBase(w.host), sel.String()), "", "", false)
 	if err != nil {
 		return nil, err
 	}
@@ -156,7 +202,7 @@ func (k *kmtronic) Set(ctx context.Context, relay int, on bool) error {
 		state = "01"
 	}
 	// FF<rr><ss>: e.g. FF0101 = relay 1 on, FF0800 = relay 8 off.
-	_, err := get(ctx, fmt.Sprintf("%s/FF%02d%s", relayBase(k.host), relay, state), k.user, k.pass)
+	_, err := get(ctx, fmt.Sprintf("%s/FF%02d%s", relayBase(k.host), relay, state), k.user, k.pass, false)
 	return err
 }
 
@@ -170,7 +216,7 @@ type kmStatus struct {
 }
 
 func (k *kmtronic) Status(ctx context.Context) ([]bool, error) {
-	body, err := get(ctx, fmt.Sprintf("%s/status.xml", relayBase(k.host)), k.user, k.pass)
+	body, err := get(ctx, fmt.Sprintf("%s/status.xml", relayBase(k.host)), k.user, k.pass, false)
 	if err != nil {
 		return nil, err
 	}

internal/syncfolder/syncfolder.go

@@ -0,0 +1,308 @@
+// Package syncfolder keeps one operator's logbook in step across several PCs
+// through a folder they already have — Seafile, OneDrive, Dropbox, a NAS share,
+// a USB stick.
+//
+// WHY NOT THE DATABASE FILE ITSELF. Because it corrupts. SQLite relies on
+// advisory file locks that SMB and NFS implement partially or cache, so two
+// machines can both believe they hold the lock; and in WAL mode — which OpsLog
+// uses — the shared-memory index (-shm) has no meaning across machines at all.
+// A cloud folder is worse again: it replicates the file whole while it is open,
+// and .db / .db-wal / .db-shm drift apart, giving a database that opens
+// perfectly and is silently missing the last few hours.
+//
+// THE RULE THAT MAKES A SHARED FOLDER SAFE: one writer per file, append only.
+// Each machine writes ONLY its own <machine>.ndjson and never touches another's.
+// A sync tool that replicates whole files can therefore never merge two writers
+// into one file, because there are never two writers. This is the exact opposite
+// of putting the database there, and it is why it works.
+//
+// WHAT THIS IS NOT. Not live. Two operators logging the same contest second by
+// second want the shared MySQL logbook, which OpsLog already does; that is a
+// different need and it stays. This is for ONE operator with a shack PC, a
+// laptop and a portable rig — the case where a server running day and night to
+// serve forty QSOs a month is the wrong shape.
+//
+// It is the other half of internal/offlineq, whose own doc says it is
+// "deliberately NOT a sync engine: no mirror, no pull, no merge, no tombstones".
+// Those four are precisely what is here.
+package syncfolder
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+)
+
+// DirName is the sub-folder created inside whatever the operator picked. A
+// dedicated folder, so pointing OpsLog at a documents directory by mistake does
+// not scatter files through it.
+const DirName = "opslog-sync"
+
+// FormatVersion is stamped on every record. A future OpsLog that changes the
+// shape can then recognise — and skip — what it does not understand, instead of
+// misreading it. Records from the future are skipped, never guessed at.
+const FormatVersion = 1
+
+// Op is what happened to a contact.
+type Op string
+
+const (
+	OpAdd    Op = "add"
+	OpUpdate Op = "update"
+	// OpDelete is a TOMBSTONE, and it is the reason this is a change log rather
+	// than a pile of ADIF. Without one, a QSO deleted on the laptop comes
+	// straight back on the next sync from the shack PC, for ever.
+	OpDelete Op = "delete"
+)
+
+// Record is one line of a machine's file. NDJSON: one object per line, appended,
+// never rewritten — so a half-written line at the tail costs one record, not the
+// file, and a reader can resume from a byte offset.
+type Record struct {
+	V    int             `json:"v"`
+	Op   Op              `json:"op"`
+	UID  string          `json:"uid"` // the contact's stable identity
+	At   time.Time       `json:"at"`  // when this CHANGE was made, UTC
+	Seq  uint64          `json:"seq"` // this machine's own counter, monotonic
+	By   string          `json:"by"`  // machine id that wrote it
+	Data json.RawMessage `json:"data,omitempty"`
+}
+
+// NewUID mints a contact's identity.
+//
+// A stable id per contact is what lets an edit or a deletion be addressed at
+// all: "the QSO with M0ABC at 14:32" is a guess, and two machines can disagree
+// about which row that is. OpsLog already mints one for the offline outbox
+// (APP_OPSLOG_QUEUEID); this is the same idea, kept for the life of the record.
+func NewUID() string {
+	var b [16]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		return fmt.Sprintf("t%d", time.Now().UnixNano())
+	}
+	return hex.EncodeToString(b[:])
+}
+
+// NewMachineID mints this installation's id, once. The name is the operator's
+// (they may call it "shack" or "portable"); the suffix keeps two machines named
+// the same from writing to one file.
+func NewMachineID(name string) string {
+	var b [4]byte
+	_, _ = rand.Read(b[:])
+	n := sanitiseName(name)
+	if n == "" {
+		n = "opslog"
+	}
+	return n + "-" + hex.EncodeToString(b[:])
+}
+
+// sanitiseName keeps a machine id usable as a FILENAME on every platform the
+// folder may be synced across — a Windows name written to a Linux NAS and back.
+func sanitiseName(s string) string {
+	s = strings.ToLower(strings.TrimSpace(s))
+	var b strings.Builder
+	for _, r := range s {
+		switch {
+		case r >= 'a' && r <= 'z', r >= '0' && r <= '9':
+			b.WriteRune(r)
+		case r == '-' || r == '_' || r == ' ':
+			b.WriteByte('-')
+		}
+	}
+	out := strings.Trim(b.String(), "-")
+	for strings.Contains(out, "--") {
+		out = strings.ReplaceAll(out, "--", "-")
+	}
+	if len(out) > 24 {
+		out = strings.Trim(out[:24], "-")
+	}
+	return out
+}
+
+// Wins decides between two changes to the same contact.
+//
+// Last writer wins, but ordered on (At, Seq, By) rather than on time alone.
+// Two PCs' clocks are never exactly equal and one may be minutes out, so a bare
+// timestamp comparison is not even deterministic: two machines merging the same
+// pair in different orders could reach different answers and then disagree for
+// ever. Adding the writer's own counter and finally its id makes the order
+// total — every machine reaches the same conclusion from the same records,
+// whatever sequence they arrive in.
+//
+// Clock skew still decides WHICH edit wins, and nothing here can fix that. What
+// it guarantees is that all machines agree on the winner.
+func Wins(a, b Record) bool {
+	if !a.At.Equal(b.At) {
+		return a.At.After(b.At)
+	}
+	if a.Seq != b.Seq {
+		return a.Seq > b.Seq
+	}
+	return a.By > b.By
+}
+
+// Store is one machine's view of the shared folder.
+type Store struct {
+	root      string // the folder the operator picked
+	machineID string
+}
+
+// New returns a store. root is the operator's chosen folder; the package
+// creates and uses its own sub-folder inside it.
+func New(root, machineID string) *Store {
+	return &Store{root: strings.TrimSpace(root), machineID: machineID}
+}
+
+// Dir is where the files live.
+func (s *Store) Dir() string { return filepath.Join(s.root, DirName) }
+
+// MyFile is the only file this machine ever writes.
+func (s *Store) MyFile() string { return filepath.Join(s.Dir(), s.machineID+".ndjson") }
+
+// Append adds one record to this machine's file.
+//
+// Opened, written and closed per call, with O_APPEND: a sync client that
+// uploads the file between two contacts sees a complete file every time, and a
+// process killed mid-write loses at most the line it was writing.
+func (s *Store) Append(rec Record) error {
+	if s.root == "" {
+		return fmt.Errorf("syncfolder: no folder configured")
+	}
+	if err := os.MkdirAll(s.Dir(), 0o755); err != nil {
+		return fmt.Errorf("syncfolder: create %s: %w", s.Dir(), err)
+	}
+	rec.V = FormatVersion
+	rec.By = s.machineID
+	if rec.At.IsZero() {
+		rec.At = time.Now().UTC()
+	}
+	rec.At = rec.At.UTC()
+	line, err := json.Marshal(rec)
+	if err != nil {
+		return fmt.Errorf("syncfolder: encode: %w", err)
+	}
+	f, err := os.OpenFile(s.MyFile(), os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0o644)
+	if err != nil {
+		return fmt.Errorf("syncfolder: open %s: %w", s.MyFile(), err)
+	}
+	defer f.Close()
+	if _, err := f.Write(append(line, '\n')); err != nil {
+		return fmt.Errorf("syncfolder: write: %w", err)
+	}
+	return nil
+}
+
+// Peer is another machine's file and how far this one has read it.
+type Peer struct {
+	MachineID string
+	Path      string
+	Size      int64
+}
+
+// Peers lists the other machines' files, skipping this machine's own.
+func (s *Store) Peers() ([]Peer, error) {
+	if s.root == "" {
+		return nil, fmt.Errorf("syncfolder: no folder configured")
+	}
+	entries, err := os.ReadDir(s.Dir())
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, nil // nobody has written anything yet
+		}
+		return nil, err
+	}
+	var out []Peer
+	for _, e := range entries {
+		if e.IsDir() || !strings.HasSuffix(e.Name(), ".ndjson") {
+			continue
+		}
+		id := strings.TrimSuffix(e.Name(), ".ndjson")
+		if id == s.machineID {
+			continue // never read our own back — that is how a loop starts
+		}
+		info, err := e.Info()
+		if err != nil {
+			continue
+		}
+		out = append(out, Peer{MachineID: id, Path: filepath.Join(s.Dir(), e.Name()), Size: info.Size()})
+	}
+	return out, nil
+}
+
+// ReadFrom returns the records in a peer's file after byte offset `from`, and
+// the offset to resume at next time.
+//
+// Resuming by byte offset is what keeps a sync cheap: a file with 40 000
+// contacts is read once, and thereafter only its tail. The returned offset
+// advances ONLY past complete lines — a file caught mid-upload ends in a
+// partial line, and stopping short of it means the next pass reads that record
+// whole instead of discarding it.
+func ReadFrom(path string, from int64) ([]Record, int64, error) {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return nil, from, err
+	}
+	// A file that SHRANK was replaced, not appended to — a sync conflict copy,
+	// a restore, a machine id reused. Start again rather than read from an
+	// offset that now points into the middle of a different record.
+	if int64(len(data)) < from {
+		from = 0
+	}
+	tail := data[from:]
+	var recs []Record
+	consumed := int64(0)
+	for {
+		i := indexByte(tail, '\n')
+		if i < 0 {
+			break // an incomplete final line: leave it for next time
+		}
+		line := tail[:i]
+		tail = tail[i+1:]
+		consumed += int64(i) + 1
+		if len(strings.TrimSpace(string(line))) == 0 {
+			continue
+		}
+		var rec Record
+		if err := json.Unmarshal(line, &rec); err != nil {
+			continue // one unreadable line must not stop the file
+		}
+		if rec.V > FormatVersion {
+			continue // written by a newer OpsLog: skip, never guess
+		}
+		if rec.UID == "" || rec.Op == "" {
+			continue
+		}
+		recs = append(recs, rec)
+	}
+	return recs, from + consumed, nil
+}
+
+func indexByte(b []byte, c byte) int {
+	for i := range b {
+		if b[i] == c {
+			return i
+		}
+	}
+	return -1
+}
+
+// Merge reduces a batch of records to ONE decision per contact — the winner.
+//
+// Applying every record in turn would work but would write the same row several
+// times over, and on a first sync of a large log that is thousands of pointless
+// updates. It also makes the result independent of the order the peers'
+// files happened to be read in.
+func Merge(recs []Record) map[string]Record {
+	out := make(map[string]Record, len(recs))
+	for _, r := range recs {
+		cur, seen := out[r.UID]
+		if !seen || Wins(r, cur) {
+			out[r.UID] = r
+		}
+	}
+	return out
+}

internal/syncfolder/syncfolder_test.go

@@ -0,0 +1,172 @@
+package syncfolder
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func rec(uid string, op Op, at string, seq uint64, by string) Record {
+	t, _ := time.Parse(time.RFC3339, at)
+	return Record{V: FormatVersion, Op: op, UID: uid, At: t, Seq: seq, By: by}
+}
+
+// A machine must never read its own file back. That is how a change loops
+// round the folder for ever, each pass re-applying what this machine wrote.
+func TestPeersExcludesOurselves(t *testing.T) {
+	root := t.TempDir()
+	s := New(root, "shack-aabbccdd")
+	if err := s.Append(rec("u1", OpAdd, "2026-08-16T10:00:00Z", 1, "")); err != nil {
+		t.Fatalf("append: %v", err)
+	}
+	must(t, os.WriteFile(filepath.Join(s.Dir(), "laptop-11223344.ndjson"), []byte("{}\n"), 0o644))
+
+	peers, err := s.Peers()
+	if err != nil {
+		t.Fatalf("peers: %v", err)
+	}
+	if len(peers) != 1 || peers[0].MachineID != "laptop-11223344" {
+		t.Fatalf("peers = %+v — our own file must not be among them", peers)
+	}
+}
+
+// Reading resumes from a byte offset, and only ever advances past COMPLETE
+// lines. A folder sync catches a file mid-upload sooner or later, and the
+// partial last line must be read whole on the next pass, not thrown away.
+func TestReadFromStopsAtAnIncompleteLine(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "peer.ndjson")
+
+	full, _ := json.Marshal(rec("u1", OpAdd, "2026-08-16T10:00:00Z", 1, "peer"))
+	partial, _ := json.Marshal(rec("u2", OpAdd, "2026-08-16T10:01:00Z", 2, "peer"))
+	must(t, os.WriteFile(path, append(append(full, '\n'), partial[:len(partial)/2]...), 0o644))
+
+	recs, off, err := ReadFrom(path, 0)
+	if err != nil {
+		t.Fatalf("ReadFrom: %v", err)
+	}
+	if len(recs) != 1 || recs[0].UID != "u1" {
+		t.Fatalf("got %d record(s) %+v, want just u1", len(recs), recs)
+	}
+	if off != int64(len(full))+1 {
+		t.Fatalf("offset %d, want %d — it must stop before the partial line", off, len(full)+1)
+	}
+
+	// The upload completes; the second record is now read whole.
+	must(t, os.WriteFile(path, append(append(append(full, '\n'), partial...), '\n'), 0o644))
+	recs, off2, err := ReadFrom(path, off)
+	if err != nil {
+		t.Fatalf("ReadFrom 2: %v", err)
+	}
+	if len(recs) != 1 || recs[0].UID != "u2" {
+		t.Fatalf("resumed with %+v, want just u2", recs)
+	}
+	if off2 <= off {
+		t.Errorf("offset did not advance: %d → %d", off, off2)
+	}
+}
+
+// A file that shrank was replaced, not appended to — a sync conflict copy, a
+// restore. Resuming at the old offset would read from the middle of a record.
+func TestReadFromRestartsIfTheFileShrank(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "peer.ndjson")
+	line, _ := json.Marshal(rec("u1", OpAdd, "2026-08-16T10:00:00Z", 1, "peer"))
+	must(t, os.WriteFile(path, append(line, '\n'), 0o644))
+
+	recs, _, err := ReadFrom(path, 999999)
+	if err != nil {
+		t.Fatalf("ReadFrom: %v", err)
+	}
+	if len(recs) != 1 {
+		t.Errorf("got %d record(s) — a shrunken file must be re-read from the start", len(recs))
+	}
+}
+
+// One unreadable line must not cost the rest of the file, and a record from a
+// FUTURE format must be skipped rather than guessed at.
+func TestReadFromSurvivesRubbish(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "peer.ndjson")
+	good, _ := json.Marshal(rec("u1", OpAdd, "2026-08-16T10:00:00Z", 1, "peer"))
+	future, _ := json.Marshal(Record{V: FormatVersion + 1, Op: OpAdd, UID: "u9", At: time.Now().UTC()})
+	body := append([]byte("this is not json\n"), append(good, '\n')...)
+	body = append(body, append(future, '\n')...)
+	body = append(body, []byte("\n")...) // a blank line
+	must(t, os.WriteFile(path, body, 0o644))
+
+	recs, _, err := ReadFrom(path, 0)
+	if err != nil {
+		t.Fatalf("ReadFrom: %v", err)
+	}
+	if len(recs) != 1 || recs[0].UID != "u1" {
+		t.Fatalf("got %+v, want only the one readable current-format record", recs)
+	}
+}
+
+// Every machine must reach the SAME winner from the same records, whatever
+// order they arrive in. Ordering on time alone is not even deterministic: two
+// PCs' clocks are never equal, and a tie would be resolved differently on each
+// machine, which is how two logs disagree for ever.
+func TestMergeIsOrderIndependent(t *testing.T) {
+	a := rec("u1", OpUpdate, "2026-08-16T10:00:00Z", 5, "shack")
+	b := rec("u1", OpDelete, "2026-08-16T10:00:00Z", 5, "laptop") // same time AND seq
+	c := rec("u1", OpUpdate, "2026-08-16T09:00:00Z", 9, "shack")  // older, higher seq
+
+	forward := Merge([]Record{a, b, c})
+	reverse := Merge([]Record{c, b, a})
+	if forward["u1"].By != reverse["u1"].By || forward["u1"].Op != reverse["u1"].Op {
+		t.Fatalf("order changed the winner: %+v vs %+v", forward["u1"], reverse["u1"])
+	}
+	// Time beats sequence: c is an hour older whatever its counter says.
+	if forward["u1"].At.Equal(c.At) {
+		t.Error("an older change won on its counter — time is the first key")
+	}
+}
+
+// A delete is a record like any other, and it must be able to WIN. Without
+// tombstones a QSO removed on one machine returns on the next sync from
+// another, for ever.
+func TestATombstoneCanWin(t *testing.T) {
+	add := rec("u1", OpAdd, "2026-08-16T10:00:00Z", 1, "shack")
+	del := rec("u1", OpDelete, "2026-08-16T11:00:00Z", 1, "laptop")
+	if got := Merge([]Record{add, del})["u1"]; got.Op != OpDelete {
+		t.Errorf("winner is %q — a later deletion must beat an earlier add", got.Op)
+	}
+	// …and an add made AFTER a deletion wins, so re-logging a contact works.
+	readd := rec("u1", OpAdd, "2026-08-16T12:00:00Z", 2, "laptop")
+	if got := Merge([]Record{add, del, readd})["u1"]; got.Op != OpAdd {
+		t.Errorf("winner is %q — a contact logged again after a deletion must come back", got.Op)
+	}
+}
+
+// The machine id becomes a FILENAME, on a folder that may be synced between
+// Windows, Linux and macOS. Anything that cannot be a filename everywhere has
+// to go.
+func TestMachineIDIsAUsableFilename(t *testing.T) {
+	for _, name := range []string{"Shack PC", "portable/rig", "Café ☕", "", "   ", "a::b*c?", strings.Repeat("x", 60)} {
+		id := NewMachineID(name)
+		for _, bad := range []string{"/", "\\", ":", "*", "?", "\"", "<", ">", "|", " "} {
+			if strings.Contains(id, bad) {
+				t.Errorf("NewMachineID(%q) = %q contains %q", name, id, bad)
+			}
+		}
+		if id == "" || strings.HasPrefix(id, "-") || strings.HasSuffix(id, "-") {
+			t.Errorf("NewMachineID(%q) = %q", name, id)
+		}
+	}
+	// Two machines the operator called the same must not share a file.
+	if NewMachineID("shack") == NewMachineID("shack") {
+		t.Error("two installations named alike produced the same id — they would write to one file")
+	}
+}
+
+func must(t *testing.T, err error) {
+	t.Helper()
+	if err != nil {
+		t.Fatal(err)
+	}
+}

internal/tciserver/tciserver.go

@@ -0,0 +1,620 @@
+// Package tciserver shares OpsLog's CAT link with programs that speak TCI.
+//
+// It is the second half of internal/rigctld, and exists for the same reason:
+// Windows gives a COM port to ONE process, so the moment OpsLog talks to the
+// radio directly nothing else can. rigctld answers the programs that speak
+// Hamlib NET rigctl (WSJT-X, JTDX, MSHV, Log4OM); this answers the ones built
+// around Expert Electronics' TCI instead — and it answers them whatever radio
+// is actually connected, because it sits on the same backend-agnostic
+// interface. An operator with an Icom or a Yaesu can hand a TCI-only program a
+// working rig.
+//
+// ── The protocol ──────────────────────────────────────────────────────────
+// Text commands over a WebSocket, "name:arg,arg;", the same syntax in both
+// directions. On connection the server sends a block of initialisation
+// commands describing the device, ending with ready; and start;. Thereafter
+// either side may send a control command, and the server echoes every change
+// to all connected clients so they stay in step with each other.
+//
+//	vfo:0,0,14074000;      receiver 0, channel A (RX), Hz
+//	vfo:0,1,14080000;      channel B — the TX frequency when split is on
+//	modulation:0,usb;      mode
+//	trx:0,true;            PTT
+//	split_enable:0,true;   split
+//	vfo:0,0;               a READ: the reply is the three-argument form
+//
+// Written against the official TCI Protocol document (ExpertSDR3/TCI, 12
+// January 2024, MIT) — the initialisation set and the argument order of every
+// command below are from §4.1 and §4.2, not from guesswork about what a client
+// might accept.
+package tciserver
+
+import (
+	"fmt"
+	"net"
+	"net/http"
+	"strconv"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/gorilla/websocket"
+)
+
+// Rig is what the server needs from OpsLog's CAT manager. An interface, so this
+// package stays testable without a radio and without importing internal/cat —
+// which also keeps it building on every platform.
+type Rig interface {
+	Freq() int64          // TX frequency in Hz (ADIF sense), 0 if unknown
+	RxFreq() int64        // RX frequency in Hz; equals Freq when not split
+	Mode() string         // ADIF mode (SSB, CW, FT8…)
+	Split() (bool, int64) // split on?, and the TX frequency
+	SetFreq(hz int64) error
+	SetMode(mode string) error
+	SetPTT(on bool) error
+	SetSplit(on bool, txHz int64) error
+}
+
+// DefaultPort is TCI's own default, which is what a client offers first.
+const DefaultPort = 40001
+
+// pollInterval is how often the rig is compared with what the clients were last
+// told. TCI is an event protocol — a client is entitled to sit silent and be
+// told when something moves — so this is the rate at which a knob turned on the
+// radio reaches it.
+const pollInterval = 250 * time.Millisecond
+
+type Server struct {
+	port int
+	rig  Rig
+	log  func(string, ...any)
+
+	mu     sync.Mutex
+	ln     net.Listener
+	http   *http.Server
+	conns  map[*client]struct{}
+	closed bool
+
+	// pendingTxHz is a transmit frequency a client set on channel B while the rig
+	// was still simplex.
+	//
+	// It must be REMEMBERED, not discarded. A client working split sends two
+	// commands and is free to send them in either order; when the frequency comes
+	// first, throwing it away means the split is then armed on whatever the
+	// transmit VFO happened to hold — the receive frequency — and the operator
+	// transmits straight onto the DX while their software shows exactly what they
+	// asked for. rigctld learned this the same way, and pairs set_split_vfo with
+	// set_split_freq for the same reason.
+	pendingTxHz int64
+
+	// ptt mirrors the last PTT state a client commanded, so a repeat can be
+	// recognised. A client is free to restate PTT as often as it likes, and one
+	// does: through the rigctl server Nexus sent set_ptt 0 about sixteen times a
+	// second, and the Flex's own "xmit 1" landed between two of them and was
+	// overwritten inside a millisecond — a transmit request that simply did
+	// nothing. The same radio sits behind this server.
+	ptt      bool
+	pttKnown bool
+
+	// last is what the clients have been told, so only changes are sent. TCI
+	// clients redraw on every command they receive; re-sending an unchanged
+	// frequency four times a second makes a VFO readout flicker and, in some
+	// clients, fights the operator's own tuning.
+	last state
+}
+
+// state is the part of the rig the clients are kept in step with.
+type state struct {
+	rxHz  int64
+	txHz  int64
+	mode  string
+	split bool
+	valid bool
+}
+
+// clientLogCap bounds how many of one client's commands reach the log.
+const clientLogCap = 200
+
+// client is one connected program.
+type client struct {
+	conn *websocket.Conn
+	mu   sync.Mutex // one writer at a time: gorilla panics on concurrent writes
+	// logged counts what has been written to the log for this connection. Only
+	// the reader goroutine touches it, so it needs no lock of its own.
+	logged int
+}
+
+func (c *client) send(s string) error {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.conn == nil {
+		return nil // a client with no socket: the tests exercise the protocol, not the transport
+	}
+	_ = c.conn.SetWriteDeadline(time.Now().Add(3 * time.Second))
+	return c.conn.WriteMessage(websocket.TextMessage, []byte(s))
+}
+
+func New(port int, rig Rig, logf func(string, ...any)) *Server {
+	if port <= 0 || port > 65535 {
+		port = DefaultPort
+	}
+	if logf == nil {
+		logf = func(string, ...any) {}
+	}
+	return &Server{port: port, rig: rig, log: logf, conns: map[*client]struct{}{}}
+}
+
+// Start binds the port and serves until Stop.
+func (s *Server) Start() error {
+	ln, err := net.Listen("tcp", fmt.Sprintf(":%d", s.port))
+	if err != nil {
+		return fmt.Errorf("tci server: port %d: %w", s.port, err)
+	}
+	up := websocket.Upgrader{
+		// Any origin: the clients are desktop programs on the same machine or
+		// LAN, and they send whatever Origin their toolkit happens to set. This
+		// is the same trust boundary as the rigctl server on 4532 — a plain TCP
+		// port with no authentication, which is what every logger expects.
+		CheckOrigin: func(*http.Request) bool { return true },
+	}
+	mux := http.NewServeMux()
+	// Any path: clients connect to ws://host:port/ but some append a name.
+	mux.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
+		conn, err := up.Upgrade(w, r, nil)
+		if err != nil {
+			s.log("tci server: upgrade from %s failed: %v", r.RemoteAddr, err)
+			return
+		}
+		s.serve(&client{conn: conn}, r.RemoteAddr)
+	})
+	srv := &http.Server{Handler: mux}
+	s.mu.Lock()
+	s.ln, s.http, s.closed = ln, srv, false
+	s.mu.Unlock()
+	go func() { _ = srv.Serve(ln) }()
+	go s.pushLoop()
+	s.log("tci server: listening on :%d", s.port)
+	return nil
+}
+
+// Stop closes the listener and every client.
+func (s *Server) Stop() {
+	s.mu.Lock()
+	if s.closed {
+		s.mu.Unlock()
+		return
+	}
+	s.closed = true
+	ln, srv := s.ln, s.http
+	conns := make([]*client, 0, len(s.conns))
+	for c := range s.conns {
+		conns = append(conns, c)
+	}
+	s.conns = map[*client]struct{}{}
+	s.last = state{}
+	s.mu.Unlock()
+	for _, c := range conns {
+		_ = c.conn.Close()
+	}
+	if srv != nil {
+		_ = srv.Close()
+	}
+	if ln != nil {
+		_ = ln.Close()
+	}
+	s.log("tci server: stopped")
+}
+
+// Clients reports how many programs are connected — the one thing an operator
+// wants to know when a client says it cannot find the rig.
+func (s *Server) Clients() int {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return len(s.conns)
+}
+
+// serve runs one connection: the initialisation block, then commands until it
+// closes.
+func (s *Server) serve(c *client, remote string) {
+	s.mu.Lock()
+	if s.closed {
+		s.mu.Unlock()
+		_ = c.conn.Close()
+		return
+	}
+	s.conns[c] = struct{}{}
+	s.mu.Unlock()
+	s.log("tci server: %s connected", remote)
+
+	for _, line := range s.initBlock() {
+		if err := c.send(line); err != nil {
+			break
+		}
+	}
+	for {
+		_, data, err := c.conn.ReadMessage()
+		if err != nil {
+			break
+		}
+		// One frame may carry several ";"-terminated commands.
+		for _, cmd := range strings.Split(string(data), ";") {
+			if cmd = strings.TrimSpace(cmd); cmd == "" {
+				continue
+			}
+			// Every command the client sends, in the log.
+			//
+			// This is the only evidence there will ever be about a program on
+			// someone else's machine: "MSHV's PTT test does nothing" is
+			// unanswerable without knowing whether MSHV sent trx at all, and if
+			// so in what form. Cheap, because TCI is event-driven — a client
+			// speaks when the operator does something, not on a timer.
+			//
+			// Capped so a client that DOES poll cannot quietly fill the
+			// operator's log; the cap says so once and then stays quiet.
+			if c.logged < clientLogCap {
+				c.logged++
+				s.log("tci server: ← %s;", cmd)
+			} else if c.logged == clientLogCap {
+				c.logged++
+				s.log("tci server: (further commands from this client are not logged)")
+			}
+			s.handle(c, cmd)
+		}
+	}
+	s.mu.Lock()
+	delete(s.conns, c)
+	s.mu.Unlock()
+	_ = c.conn.Close()
+	s.log("tci server: %s disconnected", remote)
+}
+
+// initBlock is the initialisation set from §4.1 of the protocol document, in
+// the documented order, followed by the current state so a client that has just
+// connected shows the right frequency instead of waiting for the first change.
+//
+// A client will not proceed without these: they are how it learns the device
+// exists, what it can do, and that the server has finished setting up.
+func (s *Server) initBlock() []string {
+	rx, tx, mode, split := s.read()
+	return []string{
+		"protocol:ExpertSDR3,1.9;",
+		"device:OpsLog;",
+		"receive_only:false;",
+		"trx_count:1;",
+		"channel_count:2;",
+		// The whole HF/VHF/UHF span OpsLog itself works over. A client uses this
+		// to bound its own tuning; too narrow a range and it refuses to follow the
+		// rig onto 2 m.
+		"vfo_limits:10000,470000000;",
+		"if_limits:-48000,48000;",
+		"modulations_list:am,sam,dsb,lsb,usb,cw,nfm,digl,digu;",
+		"ready;",
+		"start;",
+		fmt.Sprintf("vfo:0,0,%d;", rx),
+		fmt.Sprintf("vfo:0,1,%d;", tx),
+		fmt.Sprintf("modulation:0,%s;", mode),
+		fmt.Sprintf("split_enable:0,%t;", split),
+		"trx:0,false;",
+		// TRANSMIT PERMISSION, and it is not optional in practice.
+		//
+		// The document files TX_ENABLE under unidirectional control rather than
+		// initialisation, but its own note says it is "sent to the client when
+		// connected". A client that models permission — and one written for
+		// ExpertSDR users has every reason to — starts out assuming it may NOT
+		// transmit, and without this it never even tries: PTT does nothing and
+		// the server never sees a trx command to refuse.
+		//
+		// Always true. OpsLog is not the thing that decides: the radio behind
+		// whichever backend is connected does, and its refusal comes back through
+		// SetPTT and into the log.
+		"tx_enable:0,true;",
+		fmt.Sprintf("tx_frequency:%d;", tx),
+	}
+}
+
+// read takes one consistent snapshot of the rig in TCI's terms: channel A is
+// where we LISTEN and channel B where we transmit, which is the opposite way
+// round from ADIF's RigState and the one mistake here that would make a client
+// transmit on the DX's frequency.
+func (s *Server) read() (rxHz, txHz int64, mode string, split bool) {
+	split, txHz = s.rig.Split()
+	rxHz = s.rig.RxFreq()
+	if !split {
+		txHz = s.rig.Freq()
+		if rxHz == 0 {
+			rxHz = txHz
+		}
+	}
+	if rxHz == 0 {
+		rxHz = s.rig.Freq()
+	}
+	if txHz == 0 {
+		txHz = rxHz
+	}
+	mode = adifToTCIMode(s.rig.Mode(), rxHz)
+	return rxHz, txHz, mode, split
+}
+
+// pushLoop tells the clients what has changed on the radio.
+func (s *Server) pushLoop() {
+	t := time.NewTicker(pollInterval)
+	defer t.Stop()
+	for range t.C {
+		s.mu.Lock()
+		done := s.closed
+		s.mu.Unlock()
+		if done {
+			return
+		}
+		s.publish()
+	}
+}
+
+// publish sends only what moved. Returns the lines sent, for the tests.
+func (s *Server) publish() []string {
+	rx, tx, mode, split := s.read()
+	cur := state{rxHz: rx, txHz: tx, mode: mode, split: split, valid: true}
+
+	s.mu.Lock()
+	prev := s.last
+	s.last = cur
+	s.mu.Unlock()
+
+	var lines []string
+	if !prev.valid || prev.rxHz != cur.rxHz {
+		lines = append(lines, fmt.Sprintf("vfo:0,0,%d;", cur.rxHz))
+	}
+	if !prev.valid || prev.txHz != cur.txHz {
+		lines = append(lines, fmt.Sprintf("vfo:0,1,%d;", cur.txHz))
+		// The transmit frequency has its own command, which is what a client
+		// showing "TX 14.080" reads. Channel B alone leaves that stale.
+		lines = append(lines, fmt.Sprintf("tx_frequency:%d;", cur.txHz))
+	}
+	if (!prev.valid || prev.mode != cur.mode) && cur.mode != "" {
+		lines = append(lines, fmt.Sprintf("modulation:0,%s;", cur.mode))
+	}
+	if !prev.valid || prev.split != cur.split {
+		lines = append(lines, fmt.Sprintf("split_enable:0,%t;", cur.split))
+	}
+	for _, l := range lines {
+		s.broadcast(l)
+	}
+	return lines
+}
+
+func (s *Server) broadcast(line string) {
+	s.mu.Lock()
+	conns := make([]*client, 0, len(s.conns))
+	for c := range s.conns {
+		conns = append(conns, c)
+	}
+	s.mu.Unlock()
+	for _, c := range conns {
+		_ = c.send(line)
+	}
+}
+
+// handle answers one command from a client. Returns what was sent back, which
+// is "" for a command that only acts on the radio.
+//
+// A command that SETS something is echoed to every client, not just answered to
+// the one that sent it: the protocol document is explicit that the server
+// synchronises all connected clients, and two loggers that disagree about the
+// frequency are worse than one that is merely slow.
+func (s *Server) handle(c *client, cmd string) string {
+	name, args := cmd, ""
+	if i := strings.IndexByte(cmd, ':'); i >= 0 {
+		name, args = cmd[:i], cmd[i+1:]
+	}
+	f := strings.Split(args, ",")
+	arg := func(i int) string {
+		if i < len(f) {
+			return strings.TrimSpace(f[i])
+		}
+		return ""
+	}
+	num := func(i int) int64 {
+		v, _ := strconv.ParseInt(arg(i), 10, 64)
+		return v
+	}
+	reply := func(line string) string {
+		_ = c.send(line)
+		return line
+	}
+	rx, tx, mode, split := s.read()
+
+	switch strings.ToLower(strings.TrimSpace(name)) {
+	case "vfo":
+		// Read form: two arguments. Set form: three.
+		if len(f) < 3 || arg(2) == "" {
+			if arg(1) == "1" {
+				return reply(fmt.Sprintf("vfo:0,1,%d;", tx))
+			}
+			return reply(fmt.Sprintf("vfo:0,0,%d;", rx))
+		}
+		hz := num(2)
+		if hz <= 0 {
+			return ""
+		}
+		if arg(1) == "1" {
+			// Channel B is the transmit frequency. Setting it while simplex must
+			// not move the rig's only VFO — the client asked to prepare a split
+			// transmit frequency, not to QSY — but it must not be thrown away
+			// either: it is where the split will be armed a moment from now.
+			s.mu.Lock()
+			s.pendingTxHz = hz
+			s.mu.Unlock()
+			if !split {
+				s.broadcast(fmt.Sprintf("vfo:0,1,%d;", hz))
+				return ""
+			}
+			if err := s.rig.SetSplit(true, hz); err != nil {
+				s.log("tci server: split TX %d Hz refused: %v", hz, err)
+				return ""
+			}
+		} else if err := s.rig.SetFreq(hz); err != nil {
+			s.log("tci server: tune to %d Hz refused: %v", hz, err)
+			return ""
+		}
+		s.broadcast(fmt.Sprintf("vfo:0,%s,%d;", orZero(arg(1)), hz))
+		return ""
+
+	case "modulation":
+		if len(f) < 2 || arg(1) == "" {
+			return reply(fmt.Sprintf("modulation:0,%s;", mode))
+		}
+		m := tciModeToADIF(arg(1))
+		if m == "" {
+			return ""
+		}
+		if err := s.rig.SetMode(m); err != nil {
+			s.log("tci server: mode %s refused: %v", m, err)
+			return ""
+		}
+		s.broadcast(fmt.Sprintf("modulation:0,%s;", strings.ToLower(arg(1))))
+		return ""
+
+	case "trx":
+		if len(f) < 2 || arg(1) == "" {
+			return reply("trx:0,false;")
+		}
+		on := strings.EqualFold(arg(1), "true")
+		// Only touch the radio on a CHANGE — restating a state is not a request
+		// to change it. The first command always goes through, since there is no
+		// knowing how the radio was left.
+		s.mu.Lock()
+		known, prev := s.pttKnown, s.ptt
+		s.ptt, s.pttKnown = on, true
+		s.mu.Unlock()
+		if known && prev == on {
+			s.broadcast(fmt.Sprintf("trx:0,%t;", on))
+			return ""
+		}
+		if err := s.rig.SetPTT(on); err != nil {
+			s.log("tci server: PTT %v refused: %v", on, err)
+			return ""
+		}
+		s.log("tci server: PTT %s", map[bool]string{true: "ON", false: "off"}[on])
+		s.broadcast(fmt.Sprintf("trx:0,%t;", on))
+		return ""
+
+	case "split_enable":
+		if len(f) < 2 || arg(1) == "" {
+			return reply(fmt.Sprintf("split_enable:0,%t;", split))
+		}
+		on := strings.EqualFold(arg(1), "true")
+		// Already in the state asked for? Then it is done, and nothing goes to
+		// the radio. This is the lesson the rigctl server paid for: JTDX in "Fake
+		// It" uses no split but still says so to be sure, and a backend that
+		// cannot set split answered an error to a request that was already true.
+		// JTDX read that as rig control failing and abandoned the transmission a
+		// second into the frame. A refusal is only honest when something actually
+		// needed doing.
+		if on == split {
+			s.broadcast(fmt.Sprintf("split_enable:0,%t;", on))
+			return ""
+		}
+		// Arm on the frequency the client gave for channel B, which it is free to
+		// have sent before this command rather than after.
+		s.mu.Lock()
+		pending := s.pendingTxHz
+		s.mu.Unlock()
+		txHz := tx
+		if on && pending > 0 {
+			txHz = pending
+		}
+		if err := s.rig.SetSplit(on, txHz); err != nil {
+			// The refusal is the useful part: a backend that cannot split says
+			// so, and the client can tell the operator instead of transmitting
+			// on the wrong frequency believing all is well.
+			s.log("tci server: split %v refused: %v", on, err)
+			return ""
+		}
+		s.log("tci server: split %s, TX %d Hz", map[bool]string{true: "ON", false: "off"}[on], txHz)
+		s.broadcast(fmt.Sprintf("split_enable:0,%t;", on))
+		return ""
+
+	case "dds":
+		// The panorama's centre frequency. OpsLog has no panorama, so it answers
+		// with the receive frequency — which is where a client draws its own.
+		return reply(fmt.Sprintf("dds:0,%d;", rx))
+
+	case "if":
+		// Offset of the tuning filter inside the panorama: zero, since our "dds"
+		// is the receive frequency itself.
+		return reply("if:0,0,0;")
+
+	case "start", "stop", "ready":
+		return ""
+
+	default:
+		// Everything else — audio streams, CW macros, the E-Coder, the
+		// panorama's own settings — belongs to a radio, not to a CAT link.
+		// Silence rather than an error: a client sends these hopefully at
+		// connect, and a refusal it did not ask for reads as a fault.
+		return ""
+	}
+}
+
+func orZero(s string) string {
+	if s == "" {
+		return "0"
+	}
+	return s
+}
+
+// adifToTCIMode maps an ADIF mode to a TCI modulation.
+//
+// SSB carries no sideband, so it is resolved from the frequency the way every
+// operator does: below 10 MHz lower, above it upper. A client told "ssb" would
+// not recognise it — the modulation list is the vocabulary.
+func adifToTCIMode(mode string, hz int64) string {
+	switch strings.ToUpper(strings.TrimSpace(mode)) {
+	case "":
+		return ""
+	case "CW", "CWR":
+		return "cw"
+	case "USB":
+		return "usb"
+	case "LSB":
+		return "lsb"
+	case "SSB":
+		if hz > 0 && hz < 10_000_000 {
+			return "lsb"
+		}
+		return "usb"
+	case "AM":
+		return "am"
+	case "FM", "NFM":
+		return "nfm"
+	case "RTTY":
+		return "digl"
+	}
+	// Everything else is a data mode: FT8, FT4, JT65, PSK31, MSK144, VARA…
+	// TCI has one pair for the whole family, and the sideband follows the same
+	// rule the data modes themselves use — upper, but for the few HF corners
+	// where LSB is conventional the radio is already there.
+	return "digu"
+}
+
+// tciModeToADIF maps a TCI modulation back to an ADIF mode.
+func tciModeToADIF(m string) string {
+	switch strings.ToLower(strings.TrimSpace(m)) {
+	case "cw":
+		return "CW"
+	case "usb":
+		return "USB"
+	case "lsb":
+		return "LSB"
+	case "am", "sam":
+		return "AM"
+	case "nfm", "fm", "wfm":
+		return "FM"
+	case "digl", "digu", "dsb", "drm":
+		// The data family: the mode the operator is actually running (FT8, RTTY)
+		// is chosen in OpsLog, and a client switching to "digital" must not
+		// overwrite it with a guess. DATA is the honest ADIF answer.
+		return "DATA"
+	}
+	return ""
+}

internal/tciserver/tciserver_test.go

@@ -0,0 +1,330 @@
+package tciserver
+
+import (
+	"fmt"
+	"strings"
+	"testing"
+)
+
+// fakeRig is a radio that remembers what it was told. Everything here is about
+// what OpsLog does with a client's command, so the rig only has to answer and
+// record.
+type fakeRig struct {
+	freq, rxFreq int64
+	mode         string
+	split        bool
+	txHz         int64
+	ptt          bool
+	splitErr     error
+	calls        []string
+}
+
+func (r *fakeRig) Freq() int64          { return r.freq }
+func (r *fakeRig) RxFreq() int64        { return r.rxFreq }
+func (r *fakeRig) Mode() string         { return r.mode }
+func (r *fakeRig) Split() (bool, int64) { return r.split, r.txHz }
+func (r *fakeRig) SetFreq(hz int64) error {
+	r.calls = append(r.calls, fmt.Sprintf("freq=%d", hz))
+	r.freq, r.rxFreq = hz, hz
+	return nil
+}
+func (r *fakeRig) SetMode(m string) error {
+	r.calls = append(r.calls, "mode="+m)
+	r.mode = m
+	return nil
+}
+func (r *fakeRig) SetPTT(on bool) error {
+	r.calls = append(r.calls, fmt.Sprintf("ptt=%v", on))
+	r.ptt = on
+	return nil
+}
+func (r *fakeRig) SetSplit(on bool, txHz int64) error {
+	if r.splitErr != nil {
+		return r.splitErr
+	}
+	r.calls = append(r.calls, fmt.Sprintf("split=%v,%d", on, txHz))
+	r.split, r.txHz = on, txHz
+	return nil
+}
+
+// srv builds a server with no listener — handle() and publish() are the whole
+// protocol, and neither needs a socket.
+func srv(r *fakeRig) *Server { return New(0, r, nil) }
+
+// A client with no connection: send() would need one, so reads are checked
+// through the returned line instead. This is why handle returns what it sent.
+func ask(t *testing.T, s *Server, cmd string) string {
+	t.Helper()
+	return s.handle(&client{}, cmd)
+}
+
+// The initialisation block is what a client needs before it will believe there
+// is a radio at all. Its contents come from §4.1 of the protocol document, and
+// a client that does not see ready; simply waits for ever.
+func TestInitBlockCarriesTheDocumentedInitialisationSet(t *testing.T) {
+	s := srv(&fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"})
+	block := strings.Join(s.initBlock(), "")
+	for _, want := range []string{
+		"protocol:ExpertSDR3,", "device:", "receive_only:false;", "trx_count:1;",
+		"channel_count:2;", "vfo_limits:", "if_limits:", "modulations_list:",
+		"ready;", "start;",
+		// Transmit permission. A client that models it starts out assuming it
+		// may NOT transmit, and without this never even tries — PTT does
+		// nothing and the server never sees a trx command at all.
+		"tx_enable:0,true;",
+	} {
+		if !strings.Contains(block, want) {
+			t.Errorf("the initialisation block is missing %q — a client would not proceed past connect", want)
+		}
+	}
+	// And the current state, so a client that connects mid-session shows the
+	// right frequency instead of waiting for the operator to touch something.
+	if !strings.Contains(block, "vfo:0,0,14074000;") {
+		t.Errorf("no current frequency in the block:\n%s", block)
+	}
+	if !strings.Contains(block, "modulation:0,usb;") {
+		t.Errorf("no current mode in the block:\n%s", block)
+	}
+}
+
+// Channel A is where we LISTEN, channel B where we transmit. Handing these to a
+// client the wrong way round is the one mistake here that puts a station on the
+// DX's own frequency, so it is pinned in both directions.
+func TestSplitPutsTheListeningFrequencyOnChannelA(t *testing.T) {
+	// OpsLog's RigState is ADIF: Freq is the TRANSMIT frequency, RxFreq where we
+	// listen. A DX transmitting on 14025 and listening up 2.
+	r := &fakeRig{freq: 14027000, rxFreq: 14025000, mode: "CW", split: true, txHz: 14027000}
+	s := srv(r)
+	block := strings.Join(s.initBlock(), "")
+	if !strings.Contains(block, "vfo:0,0,14025000;") {
+		t.Errorf("channel A is not the receive frequency:\n%s", block)
+	}
+	if !strings.Contains(block, "vfo:0,1,14027000;") {
+		t.Errorf("channel B is not the transmit frequency:\n%s", block)
+	}
+	if !strings.Contains(block, "split_enable:0,true;") {
+		t.Errorf("split was not announced:\n%s", block)
+	}
+}
+
+// Simplex: both channels report the one frequency, so a client reading either
+// gets the right answer.
+func TestSimplexReportsTheSameFrequencyOnBothChannels(t *testing.T) {
+	s := srv(&fakeRig{freq: 7100000, rxFreq: 7100000, mode: "SSB"})
+	if got := ask(t, s, "vfo:0,0"); got != "vfo:0,0,7100000;" {
+		t.Errorf("read of channel A = %q", got)
+	}
+	if got := ask(t, s, "vfo:0,1"); got != "vfo:0,1,7100000;" {
+		t.Errorf("read of channel B = %q", got)
+	}
+	// 7 MHz is below 10, so SSB is lower sideband — a client told "ssb" would
+	// not recognise it at all, the modulation list is the vocabulary.
+	if got := ask(t, s, "modulation:0"); got != "modulation:0,lsb;" {
+		t.Errorf("read of the mode = %q, want lsb below 10 MHz", got)
+	}
+}
+
+// The client tunes the radio.
+func TestAClientCanTuneAndSetModeAndKey(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"}
+	s := srv(r)
+	ask(t, s, "vfo:0,0,14200000")
+	ask(t, s, "modulation:0,cw")
+	ask(t, s, "trx:0,true")
+	ask(t, s, "trx:0,false")
+	want := []string{"freq=14200000", "mode=CW", "ptt=true", "ptt=false"}
+	if strings.Join(r.calls, " ") != strings.Join(want, " ") {
+		t.Errorf("the radio was told %v, want %v", r.calls, want)
+	}
+}
+
+// Channel B is the SPLIT transmit frequency. Writing it while the rig is
+// simplex must not move the only VFO there is: the client asked to prepare a
+// transmit frequency, not to QSY — and a logger that did this on every spot
+// click would drag the operator off the station they were listening to.
+func TestWritingChannelBWhileSimplexLeavesTheRigAlone(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"}
+	s := srv(r)
+	ask(t, s, "vfo:0,1,14080000")
+	if len(r.calls) != 0 {
+		t.Errorf("the radio was told %v — a split TX frequency moved a simplex rig", r.calls)
+	}
+	// With split armed it means what it says.
+	r.split, r.txHz = true, 14074000
+	ask(t, s, "vfo:0,1,14080000")
+	if len(r.calls) != 1 || r.calls[0] != "split=true,14080000" {
+		t.Errorf("with split on the radio was told %v, want the new transmit frequency", r.calls)
+	}
+}
+
+// A backend that cannot split says so, and the refusal must not be dressed up
+// as success: the client can then tell the operator to use Fake It, where
+// before it would transmit on the receive frequency believing all was well.
+func TestARefusedSplitIsNotAnnouncedAsDone(t *testing.T) {
+	r := &fakeRig{freq: 14025000, rxFreq: 14025000, mode: "CW", splitErr: fmt.Errorf("this backend cannot split")}
+	s := srv(r)
+	if got := ask(t, s, "split_enable:0,true"); got != "" {
+		t.Errorf("a refused split answered %q", got)
+	}
+	if r.split {
+		t.Error("the rig was recorded as split after the backend refused")
+	}
+}
+
+// Only what moved is sent. TCI clients redraw on every command they receive, so
+// re-sending an unchanged frequency four times a second makes a VFO readout
+// flicker and, in some clients, fights the operator's own tuning.
+func TestOnlyChangesAreSent(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"}
+	s := srv(r)
+
+	first := s.publish()
+	if len(first) == 0 {
+		t.Fatal("the first pass sent nothing — a client would never learn the state")
+	}
+	if got := s.publish(); len(got) != 0 {
+		t.Errorf("an unchanged radio produced %v", got)
+	}
+
+	r.freq, r.rxFreq = 14200000, 14200000
+	got := strings.Join(s.publish(), "")
+	// Both channels move together on a simplex rig, and the transmit frequency
+	// has its own command besides — a client showing "TX 14.200" reads that one,
+	// and channel B alone leaves it stale.
+	for _, want := range []string{"vfo:0,0,14200000;", "vfo:0,1,14200000;", "tx_frequency:14200000;"} {
+		if !strings.Contains(got, want) {
+			t.Errorf("after a QSY the clients were not told %q — got %q", want, got)
+		}
+	}
+	if got := s.publish(); len(got) != 0 {
+		t.Errorf("the QSY was re-sent: %v", got)
+	}
+}
+
+// Modes travel both ways, and the data family is the interesting half: a client
+// switching to "digital" must not overwrite the mode the operator chose in
+// OpsLog with a guess at which data mode it was.
+func TestModeMapping(t *testing.T) {
+	up := []struct {
+		adif string
+		hz   int64
+		want string
+	}{
+		{"CW", 14025000, "cw"},
+		{"SSB", 14200000, "usb"},
+		{"SSB", 7100000, "lsb"},
+		{"USB", 7100000, "usb"}, // an explicit sideband is never second-guessed
+		{"FT8", 14074000, "digu"},
+		{"RTTY", 14080000, "digl"},
+		{"AM", 3700000, "am"},
+		{"FM", 145500000, "nfm"},
+		{"", 14074000, ""},
+	}
+	for _, c := range up {
+		if got := adifToTCIMode(c.adif, c.hz); got != c.want {
+			t.Errorf("adifToTCIMode(%q, %d) = %q, want %q", c.adif, c.hz, got, c.want)
+		}
+	}
+	down := map[string]string{
+		"cw": "CW", "usb": "USB", "lsb": "LSB", "am": "AM", "sam": "AM",
+		"nfm": "FM", "digu": "DATA", "digl": "DATA", "": "",
+	}
+	for in, want := range down {
+		if got := tciModeToADIF(in); got != want {
+			t.Errorf("tciModeToADIF(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+// A command for something OpsLog is not — audio streams, CW macros, the
+// panorama's settings — is met with silence rather than an error. A client
+// sends these hopefully at connect, and a refusal it did not ask for reads as a
+// fault with the rig.
+func TestUnknownCommandsAreQuiet(t *testing.T) {
+	s := srv(&fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"})
+	for _, cmd := range []string{"audio_start:0", "cw_macros_speed:25", "rx_filter_band:0,-2700,-100", "iq_start:0"} {
+		if got := ask(t, s, cmd); got != "" {
+			t.Errorf("%q answered %q", cmd, got)
+		}
+	}
+}
+
+// Split, with the client sending the two commands in the order it prefers.
+//
+// A client working split has to say two things: where to transmit, and that
+// split is on. Nothing obliges it to say them in that order, and the frequency
+// arriving first is the dangerous case: discarding it and then arming split
+// leaves the transmit VFO on whatever it held — the RECEIVE frequency — so the
+// operator transmits straight onto the DX while their software shows exactly
+// what they asked for.
+func TestSplitIsArmedOnTheFrequencyTheClientGaveWhicheverOrderItCame(t *testing.T) {
+	// Frequency first, then split — the order that used to lose the frequency.
+	r := &fakeRig{freq: 14025000, rxFreq: 14025000, mode: "CW"}
+	s := srv(r)
+	ask(t, s, "vfo:0,1,14027000")
+	ask(t, s, "split_enable:0,true")
+	if len(r.calls) != 1 || r.calls[0] != "split=true,14027000" {
+		t.Errorf("frequency first: the radio was told %v, want split armed on 14027000", r.calls)
+	}
+
+	// Split first, then the frequency — the order that always worked.
+	r2 := &fakeRig{freq: 14025000, rxFreq: 14025000, mode: "CW"}
+	s2 := srv(r2)
+	ask(t, s2, "split_enable:0,true")
+	ask(t, s2, "vfo:0,1,14027000")
+	if len(r2.calls) == 0 || r2.calls[len(r2.calls)-1] != "split=true,14027000" {
+		t.Errorf("split first: the radio was told %v, want it to end on 14027000", r2.calls)
+	}
+}
+
+// "Fake It" uses no split at all: the client shifts the DIAL at the start of
+// transmit and shifts it back at the end. All it needs is channel A, and it
+// must reach the radio both ways.
+func TestFakeItIsJustTheDialMoving(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"}
+	s := srv(r)
+	ask(t, s, "vfo:0,0,14075300") // up for the over
+	ask(t, s, "vfo:0,0,14074000") // and back
+	want := []string{"freq=14075300", "freq=14074000"}
+	if strings.Join(r.calls, " ") != strings.Join(want, " ") {
+		t.Errorf("the radio was told %v, want %v", r.calls, want)
+	}
+}
+
+// A client in Fake It still says "split off" to be sure. The rig is already
+// simplex, so there is nothing to do — and saying so beats asking a backend
+// that may not be able to set split at all.
+//
+// This is what broke JTDX through the rigctl server: an error answered to a
+// request that was already true, read as rig control failing, and the
+// transmission abandoned a second into the frame.
+func TestSayingSplitOffWhenAlreadySimplexTouchesNothing(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB",
+		splitErr: fmt.Errorf("this backend cannot split")}
+	s := srv(r)
+	ask(t, s, "split_enable:0,false")
+	if len(r.calls) != 0 {
+		t.Errorf("the radio was told %v for a state it was already in", r.calls)
+	}
+}
+
+// A client restating PTT must not re-command the radio. Through the rigctl
+// server, one sent set_ptt 0 sixteen times a second and the Flex's own transmit
+// request was overwritten between two of them inside a millisecond.
+func TestRepeatedPTTIsNotResentToTheRadio(t *testing.T) {
+	r := &fakeRig{freq: 14074000, rxFreq: 14074000, mode: "USB"}
+	s := srv(r)
+	for i := 0; i < 5; i++ {
+		ask(t, s, "trx:0,false")
+	}
+	if len(r.calls) != 1 || r.calls[0] != "ptt=false" {
+		// The FIRST one always goes through: there is no knowing how the radio
+		// was left.
+		t.Errorf("the radio was told %v, want one unkey and no repeats", r.calls)
+	}
+	ask(t, s, "trx:0,true")
+	ask(t, s, "trx:0,true")
+	if len(r.calls) != 2 || r.calls[1] != "ptt=true" {
+		t.Errorf("the radio was told %v, want the change through and the repeat dropped", r.calls)
+	}
+}

lookupwhen_test.go

@@ -0,0 +1,45 @@
+package main
+
+import (
+	"testing"
+	"time"
+)
+
+// A ClubLog exception has a validity window, and a DXpedition's window closes.
+// So a backdated entry must be enriched as of WHEN IT HAPPENED — 3Y0K typed
+// months after the activation, with the activation's date in the form, resolved
+// at today's date, found no live exception, and fell back to cty.dat:
+// Antarctica, where the log says Bouvet Island.
+//
+// The date is trusted to move the resolution BACKWARDS only. A half-typed date
+// must not send the lookup to the year 20, and a mistyped future one must not
+// resolve against a window that has not opened.
+func TestLookupWhen(t *testing.T) {
+	now := time.Now().UTC()
+
+	if got := lookupWhen("2026-03-08"); got.Format("2006-01-02") != "2026-03-08" {
+		t.Errorf("a past date gave %v — the activation's own date is the whole point", got)
+	}
+	// Midday, not midnight: a window given in whole days is inclusive of its end
+	// date, and 00:00 sits exactly on the boundary.
+	if h := lookupWhen("2026-03-08").Hour(); h != 12 {
+		t.Errorf("resolved at %02d:00, want 12:00 — midnight sits on the window boundary", h)
+	}
+
+	for name, in := range map[string]string{
+		"empty":        "",
+		"spaces":       "   ",
+		"half-typed":   "2026-0",
+		"not a date":   "hier",
+		"wrong format": "08/03/2026",
+	} {
+		if got := lookupWhen(in); got.Before(now.Add(-time.Minute)) {
+			t.Errorf("%s (%q) resolved to %v — anything unusable must mean now", name, in, got)
+		}
+	}
+
+	future := now.AddDate(1, 0, 0).Format("2006-01-02")
+	if got := lookupWhen(future); got.After(now.Add(time.Minute)) {
+		t.Errorf("a future date (%s) resolved to %v — a mistyped year must not open a window early", future, got)
+	}
+}

syncapply_test.go

@@ -0,0 +1,176 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"hamlog/internal/db"
+	"hamlog/internal/qso"
+	"hamlog/internal/syncfolder"
+)
+
+// syncTestApp is an App with nothing but a logbook: applySyncRecord touches the
+// repository and the log file, and no more. Settings are nil, which is exactly
+// the state it must survive anyway — the loop runs before the active profile is
+// known.
+func syncTestApp(t *testing.T) *App {
+	t.Helper()
+	conn, err := db.Open(filepath.Join(t.TempDir(), "log.db"))
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { conn.Close() })
+	return &App{ctx: context.Background(), qso: qso.NewRepo(conn)}
+}
+
+func syncRecord(t *testing.T, op syncfolder.Op, uid string, q qso.QSO) syncfolder.Record {
+	t.Helper()
+	rec := syncfolder.Record{V: syncfolder.FormatVersion, Op: op, UID: uid, At: time.Now().UTC(), By: "other"}
+	if op != syncfolder.OpDelete {
+		b, err := json.Marshal(q)
+		if err != nil {
+			t.Fatalf("marshal: %v", err)
+		}
+		rec.Data = b
+	}
+	return rec
+}
+
+func countQSOs(t *testing.T, a *App) int64 {
+	t.Helper()
+	n, err := a.qso.Count(a.ctx)
+	if err != nil {
+		t.Fatalf("count: %v", err)
+	}
+	return n
+}
+
+// The ordinary life of a contact made on the other PC: it arrives, it is
+// corrected, it is deleted. One row throughout — a sync that inserted a second
+// copy on the edit would be worse than no sync at all.
+func TestSyncRecordAddThenUpdateThenDelete(t *testing.T) {
+	a := syncTestApp(t)
+	when := time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC)
+	uid := syncfolder.NewUID()
+
+	if !a.applySyncRecord(syncRecord(t, syncfolder.OpAdd, uid, qso.QSO{
+		Callsign: "M0ABC", QSODate: when, Band: "20m", Mode: "CW", Name: "Ann",
+	})) {
+		t.Fatal("the added contact was not applied")
+	}
+	if n := countQSOs(t, a); n != 1 {
+		t.Fatalf("log holds %d QSOs after an add, want 1", n)
+	}
+	// Stamped with the identity from the record — without this every later
+	// change naming it would look like a contact never seen before.
+	id, found, err := a.qso.IDBySyncUID(a.ctx, uid)
+	if err != nil || !found {
+		t.Fatalf("IDBySyncUID = (%d,%v,%v), want the new row", id, found, err)
+	}
+
+	if !a.applySyncRecord(syncRecord(t, syncfolder.OpUpdate, uid, qso.QSO{
+		Callsign: "M0ABC", QSODate: when, Band: "20m", Mode: "CW", Name: "Annette",
+	})) {
+		t.Fatal("the correction was not applied")
+	}
+	if n := countQSOs(t, a); n != 1 {
+		t.Fatalf("log holds %d QSOs after an edit, want 1 — the edit was logged as a second contact", n)
+	}
+	got, err := a.qso.GetByID(a.ctx, id)
+	if err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	if got.Name != "Annette" {
+		t.Errorf("name = %q after the correction, want %q", got.Name, "Annette")
+	}
+
+	if !a.applySyncRecord(syncRecord(t, syncfolder.OpDelete, uid, qso.QSO{})) {
+		t.Fatal("the tombstone was not applied")
+	}
+	if n := countQSOs(t, a); n != 0 {
+		t.Fatalf("log holds %d QSOs after the deletion, want 0", n)
+	}
+	// A tombstone that arrives twice — both peers relayed it, or the file was
+	// re-read after a restore — must be quiet, not an error and not a change.
+	if a.applySyncRecord(syncRecord(t, syncfolder.OpDelete, uid, qso.QSO{})) {
+		t.Error("a repeated tombstone reported a change; the grid would refresh for nothing, for ever")
+	}
+}
+
+// The case the whole no-backfill decision rests on.
+//
+// Both PCs already hold the operator's 123 000 contacts — seeded from one
+// database or one ADIF — and neither row carries an identity, because nothing
+// has touched them since. The day the shack PC corrects a 2019 QSO it stamps an
+// identity and sends an update naming it; the laptop has never seen that
+// identity. Inserting would give the operator two copies of a contact they
+// merely corrected, and would do it for every edit for ever.
+func TestSyncRecordAdoptsTheContactAlreadyInTheLog(t *testing.T) {
+	a := syncTestApp(t)
+	when := time.Date(2019, 3, 2, 9, 15, 0, 0, time.UTC)
+
+	// The copy that was already here, with no identity.
+	localID, err := a.qso.Add(a.ctx, qso.QSO{
+		Callsign: "M0ABC", QSODate: when, Band: "40m", Mode: "SSB", Name: "Ann",
+	})
+	if err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+
+	uid := syncfolder.NewUID() // minted on the OTHER machine
+	if !a.applySyncRecord(syncRecord(t, syncfolder.OpUpdate, uid, qso.QSO{
+		Callsign: "M0ABC", QSODate: when, Band: "40m", Mode: "SSB", Name: "Annette", QTH: "Bristol",
+	})) {
+		t.Fatal("the correction was not applied")
+	}
+	if n := countQSOs(t, a); n != 1 {
+		t.Fatalf("log holds %d QSOs, want 1 — the contact was duplicated instead of recognised", n)
+	}
+	got, err := a.qso.GetByID(a.ctx, localID)
+	if err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	if got.Name != "Annette" || got.QTH != "Bristol" {
+		t.Errorf("the row already here was not corrected: name=%q qth=%q", got.Name, got.QTH)
+	}
+	// And it now carries the identity, so the NEXT change goes straight to it
+	// without needing the contact-matching fallback again.
+	if id, found, _ := a.qso.IDBySyncUID(a.ctx, uid); !found || id != localID {
+		t.Errorf("IDBySyncUID = (%d,%v), want the row already here (%d)", id, found, localID)
+	}
+}
+
+// A different contact must NOT be adopted. The matching is deliberately narrow
+// — same callsign, same minute, same band, same mode — and this pins that a
+// second contact with the same station on another band stays a second contact.
+func TestSyncRecordDoesNotAdoptADifferentContact(t *testing.T) {
+	a := syncTestApp(t)
+	when := time.Date(2026, 8, 16, 14, 32, 0, 0, time.UTC)
+	if _, err := a.qso.Add(a.ctx, qso.QSO{Callsign: "M0ABC", QSODate: when, Band: "40m", Mode: "CW"}); err != nil {
+		t.Fatalf("seed: %v", err)
+	}
+	if !a.applySyncRecord(syncRecord(t, syncfolder.OpAdd, syncfolder.NewUID(), qso.QSO{
+		Callsign: "M0ABC", QSODate: when, Band: "20m", Mode: "CW",
+	})) {
+		t.Fatal("the contact was not applied")
+	}
+	if n := countQSOs(t, a); n != 2 {
+		t.Fatalf("log holds %d QSOs, want 2 — a contact on another band was swallowed as a duplicate", n)
+	}
+}
+
+// A record with no callsign is not a contact. It reaches here from a file
+// truncated by a sync client mid-upload, or from a future format read
+// optimistically, and inserting it would put a blank row in the log.
+func TestSyncRecordIgnoresAContactWithNoCallsign(t *testing.T) {
+	a := syncTestApp(t)
+	if a.applySyncRecord(syncRecord(t, syncfolder.OpAdd, syncfolder.NewUID(), qso.QSO{Band: "20m", Mode: "CW"})) {
+		t.Error("a record with no callsign was applied")
+	}
+	if n := countQSOs(t, a); n != 0 {
+		t.Fatalf("log holds %d QSOs, want 0", n)
+	}
+}

syncfolder.go

@@ -0,0 +1,554 @@
+package main
+
+// Folder synchronisation — one operator, several PCs, one logbook.
+//
+// The operator points every OpsLog at the SAME folder (Seafile, OneDrive,
+// Dropbox, a NAS share). Each machine appends what it logs, edits and deletes
+// to its own file in there, and reads the others'. internal/syncfolder holds
+// the format and the merge rules, and its package doc explains why the change
+// log is a set of append-only files rather than the database itself.
+//
+// This file is the wiring: settings, the loop, and the three hooks on the
+// logging path.
+//
+// WHAT SYNCHRONISES. Only what happens from the moment it is switched on.
+// There is deliberately no mass backfill of the log already on disk: the two
+// PCs of an operator who has been logging for years hold the same history
+// already (one was seeded from the other, or from the same ADIF), and pushing
+// 123 000 contacts through a synced folder to tell the other machine what it
+// already knows would cost hours and gain nothing. A contact is stamped with an
+// identity when it is touched — logged, edited, deleted — and that is what the
+// other machines are told about.
+//
+// WHY IT STILL RECOGNISES OLD CONTACTS. Because an edit to a 2019 QSO does
+// travel, and the receiving machine has that QSO under a different row id and
+// no identity. It matches on the contact itself (callsign, minute, band, mode)
+// before inserting, so an edit lands on the row already there instead of
+// creating a second copy. That is IDByDedupeKey, and it is the whole reason the
+// no-backfill decision is safe.
+//
+// NOT LIVE, AND NOT MEANT TO BE. Two operators working a contest together want
+// the shared MySQL logbook, which OpsLog already does. This is for one operator
+// whose contacts are spread across a shack PC, a laptop and a portable rig.
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+
+	wruntime "github.com/wailsapp/wails/v2/pkg/runtime"
+
+	"hamlog/internal/applog"
+	"hamlog/internal/qso"
+	"hamlog/internal/syncfolder"
+)
+
+// Settings keys. All PROFILE-SCOPED, and that is load-bearing: each profile can
+// point at its own logbook, so each needs its own folder, its own machine id
+// (hence its own file — two profiles sharing a folder would otherwise write
+// two logbooks into one) and its own read positions.
+const (
+	keySyncFolder        = "syncfolder.config"
+	keySyncFolderMachine = "syncfolder.machine" // this installation's id, minted once
+	keySyncFolderOffsets = "syncfolder.offsets" // peer machine id → bytes already read
+	keySyncFolderSeq     = "syncfolder.seq"     // this machine's own counter
+)
+
+// syncPollInterval is how often the folder is examined. A synced folder is not
+// instant anyway — Seafile and OneDrive take seconds to notice a change and
+// seconds more to push it — so polling faster would only burn a directory
+// listing to learn nothing.
+const syncPollInterval = 20 * time.Second
+
+// FolderSyncConfig is what the operator sets.
+type FolderSyncConfig struct {
+	Enabled bool   `json:"enabled"`
+	Folder  string `json:"folder"`
+	// Machine is the operator's own name for this PC — "shack", "portable".
+	// It only labels the file and the status; the identity that matters is the
+	// id minted from it, which carries a random suffix so two PCs both called
+	// "shack" still never write to one file.
+	Machine string `json:"machine"`
+}
+
+// FolderSyncPeer is another machine seen in the folder.
+type FolderSyncPeer struct {
+	Machine string `json:"machine"`
+	// LastChange is the file's modification time — "when did that PC last log
+	// anything", which is the question an operator actually asks of this list.
+	LastChange string `json:"last_change"`
+	Behind     int64  `json:"behind"` // bytes written but not yet read here
+}
+
+// FolderSyncStatus is what the settings panel shows.
+type FolderSyncStatus struct {
+	Enabled   bool             `json:"enabled"`
+	Folder    string           `json:"folder"`
+	MachineID string           `json:"machine_id"`
+	Peers     []FolderSyncPeer `json:"peers"`
+	LastSync  string           `json:"last_sync"`
+	Sent      int64            `json:"sent"`
+	Received  int64            `json:"received"`
+	Error     string           `json:"error"`
+}
+
+func (a *App) loadFolderSync() FolderSyncConfig {
+	var cfg FolderSyncConfig
+	if a.settings == nil || !a.settingsScoped.Load() {
+		return cfg
+	}
+	s, _ := a.settings.Get(a.ctx, keySyncFolder)
+	if strings.TrimSpace(s) != "" {
+		_ = json.Unmarshal([]byte(s), &cfg)
+	}
+	return cfg
+}
+
+// GetFolderSync returns the configuration for the settings panel.
+func (a *App) GetFolderSync() FolderSyncConfig {
+	a.syncMu.Lock()
+	defer a.syncMu.Unlock()
+	return a.loadFolderSync()
+}
+
+// SaveFolderSync persists the configuration.
+//
+// The folder is checked by WRITING to it, not by asking whether it exists: a
+// cloud folder that is read-only, or a NAS share whose credentials have
+// expired, exists perfectly well and would swallow every contact in silence.
+// Better to refuse in the settings panel, where the operator is looking.
+func (a *App) SaveFolderSync(cfg FolderSyncConfig) error {
+	a.syncMu.Lock()
+	defer a.syncMu.Unlock()
+	cfg.Folder = strings.TrimSpace(cfg.Folder)
+	cfg.Machine = strings.TrimSpace(cfg.Machine)
+	if cfg.Enabled {
+		if cfg.Folder == "" {
+			return fmt.Errorf("choose the synchronised folder first")
+		}
+		if err := checkWritableDir(cfg.Folder); err != nil {
+			return err
+		}
+		if cfg.Machine == "" {
+			cfg.Machine = "PC"
+		}
+	}
+	// The id is minted from the name ONCE and then kept, even if the operator
+	// renames the PC afterwards. Re-minting would orphan the file already in
+	// the folder: the other machines would go on reading the old one for ever
+	// and never see another contact from here.
+	if cfg.Enabled && a.settings != nil {
+		if cur, _ := a.settings.Get(a.ctx, keySyncFolderMachine); strings.TrimSpace(cur) == "" {
+			a.setSetting(keySyncFolderMachine, syncfolder.NewMachineID(cfg.Machine))
+		}
+	}
+	b, _ := json.Marshal(cfg)
+	a.setSetting(keySyncFolder, string(b))
+	applog.Printf("foldersync: enabled=%v folder=%q machine=%q", cfg.Enabled, cfg.Folder, cfg.Machine)
+	return nil
+}
+
+// checkWritableDir proves the folder can be written to, and cleans up after
+// itself.
+func checkWritableDir(dir string) error {
+	info, err := os.Stat(dir)
+	if err != nil {
+		return fmt.Errorf("cannot reach %s: %w", dir, err)
+	}
+	if !info.IsDir() {
+		return fmt.Errorf("%s is not a folder", dir)
+	}
+	probe := filepath.Join(dir, ".opslog-write-test")
+	if err := os.WriteFile(probe, []byte("opslog"), 0o644); err != nil {
+		return fmt.Errorf("cannot write to %s: %w", dir, err)
+	}
+	_ = os.Remove(probe)
+	return nil
+}
+
+// PickFolderSyncFolder opens the folder chooser.
+func (a *App) PickFolderSyncFolder() (string, error) {
+	if a.ctx == nil {
+		return "", fmt.Errorf("no app context")
+	}
+	return wruntime.OpenDirectoryDialog(a.ctx, wruntime.OpenDialogOptions{
+		Title: "Choose the folder your PCs already synchronise",
+	})
+}
+
+// syncStore returns this machine's view of the folder, or nil when folder
+// synchronisation is off or not configured. Every caller treats nil as "not
+// our business" — the hooks on the logging path especially, where this must
+// cost nothing at all for the operators who never turn it on.
+func (a *App) syncStore() (*syncfolder.Store, FolderSyncConfig) {
+	cfg := a.loadFolderSync()
+	if !cfg.Enabled || cfg.Folder == "" || a.settings == nil {
+		return nil, cfg
+	}
+	id, _ := a.settings.Get(a.ctx, keySyncFolderMachine)
+	if strings.TrimSpace(id) == "" {
+		return nil, cfg
+	}
+	return syncfolder.New(cfg.Folder, id), cfg
+}
+
+// nextSyncSeq hands out this machine's next counter value.
+//
+// Persisted on every use rather than at shutdown: the counter breaks ties
+// between two changes made in the same second, and one that restarted at zero
+// after a crash would make an older change beat a newer one for ever.
+func (a *App) nextSyncSeq() uint64 {
+	n := uint64(0)
+	if a.settings != nil {
+		s, _ := a.settings.Get(a.ctx, keySyncFolderSeq)
+		fmt.Sscanf(strings.TrimSpace(s), "%d", &n)
+	}
+	n++
+	a.setSetting(keySyncFolderSeq, fmt.Sprintf("%d", n))
+	return n
+}
+
+// syncUIDFor returns a contact's identity, minting and stamping one if it has
+// none. This is where an old QSO joins the sync: not in bulk, but the first
+// time it is touched.
+func (a *App) syncUIDFor(id int64, known string) string {
+	if strings.TrimSpace(known) != "" {
+		return known
+	}
+	if a.qso == nil || id <= 0 {
+		return ""
+	}
+	if q, err := a.qso.GetByID(a.ctx, id); err == nil && strings.TrimSpace(q.SyncUID) != "" {
+		return q.SyncUID
+	}
+	uid := syncfolder.NewUID()
+	if err := a.qso.SetSyncUID(a.ctx, id, uid); err != nil {
+		applog.Printf("foldersync: stamping QSO %d failed: %v", id, err)
+		return ""
+	}
+	return uid
+}
+
+// syncPublish records one local change for the other machines.
+//
+// Never on the critical path of logging: a folder on a network share can block
+// for seconds, and a contact must be in the database and on screen long before
+// anyone cares that another PC knows about it. Callers run it in a goroutine.
+func (a *App) syncPublish(op syncfolder.Op, id int64, q *qso.QSO) {
+	a.syncMu.Lock()
+	defer a.syncMu.Unlock()
+	store, _ := a.syncStore()
+	if store == nil {
+		return
+	}
+	known := ""
+	if q != nil {
+		known = q.SyncUID
+	}
+	uid := a.syncUIDFor(id, known)
+	if uid == "" {
+		return
+	}
+	rec := syncfolder.Record{Op: op, UID: uid, Seq: a.nextSyncSeq()}
+	// A deletion carries no contact — the tombstone is the whole message, and
+	// the receiving machine finds the row by the identity.
+	if op != syncfolder.OpDelete {
+		full := q
+		if full == nil || full.ID != id {
+			got, err := a.qso.GetByID(a.ctx, id)
+			if err != nil {
+				applog.Printf("foldersync: reading QSO %d back failed: %v", id, err)
+				return
+			}
+			full = &got
+		}
+		// The row id is this machine's and means nothing anywhere else. Left in,
+		// it would be read back as "update local row 4711" on a PC where 4711 is
+		// somebody else entirely.
+		cp := *full
+		cp.ID = 0
+		cp.SyncUID = uid
+		b, err := json.Marshal(cp)
+		if err != nil {
+			applog.Printf("foldersync: encoding QSO %d failed: %v", id, err)
+			return
+		}
+		rec.Data = b
+	}
+	if err := store.Append(rec); err != nil {
+		a.syncErr = err.Error()
+		applog.Printf("foldersync: append failed: %v", err)
+		return
+	}
+	a.syncErr = ""
+	a.syncSent++
+}
+
+// syncPublishAsync is what the logging path calls.
+func (a *App) syncPublishAsync(op syncfolder.Op, id int64, q *qso.QSO) {
+	if a.qso == nil {
+		return
+	}
+	var cp *qso.QSO
+	if q != nil {
+		c := *q
+		cp = &c
+	}
+	go a.syncPublish(op, id, cp)
+}
+
+// syncPublishDeletes records tombstones for rows about to be deleted.
+//
+// Called BEFORE the delete and synchronously, for the same reason
+// deleteRemoteCopies is: once the rows are gone their identities are gone with
+// them, and a tombstone naming nothing tells the other machines nothing.
+func (a *App) syncPublishDeletes(ids []int64) {
+	if a.qso == nil || len(ids) == 0 {
+		return
+	}
+	a.syncMu.Lock()
+	store, _ := a.syncStore()
+	a.syncMu.Unlock()
+	if store == nil {
+		return
+	}
+	for _, id := range ids {
+		q, err := a.qso.GetByID(a.ctx, id)
+		if err != nil {
+			continue
+		}
+		// A contact never touched since the sync was switched on has no identity,
+		// and giving it one now is what makes the deletion addressable at all.
+		a.syncMu.Lock()
+		uid := a.syncUIDFor(id, q.SyncUID)
+		if uid != "" {
+			if err := store.Append(syncfolder.Record{Op: syncfolder.OpDelete, UID: uid, Seq: a.nextSyncSeq()}); err != nil {
+				applog.Printf("foldersync: tombstone for QSO %d failed: %v", id, err)
+			} else {
+				a.syncSent++
+			}
+		}
+		a.syncMu.Unlock()
+	}
+}
+
+func (a *App) loadSyncOffsets() map[string]int64 {
+	out := map[string]int64{}
+	if a.settings == nil {
+		return out
+	}
+	s, _ := a.settings.Get(a.ctx, keySyncFolderOffsets)
+	if strings.TrimSpace(s) != "" {
+		_ = json.Unmarshal([]byte(s), &out)
+	}
+	return out
+}
+
+func (a *App) saveSyncOffsets(m map[string]int64) {
+	b, _ := json.Marshal(m)
+	a.setSetting(keySyncFolderOffsets, string(b))
+}
+
+// folderSyncLoop reads the other machines' files on an interval, for the life
+// of the app. Cheap when switched off: one settings read.
+func (a *App) folderSyncLoop() {
+	tick := time.NewTicker(syncPollInterval)
+	defer tick.Stop()
+	for range tick.C {
+		if a.ctx == nil || a.qso == nil {
+			continue
+		}
+		if n, err := a.folderSyncPass(); err != nil {
+			applog.Printf("foldersync: %v", err)
+		} else if n > 0 {
+			applog.Printf("foldersync: applied %d change(s) from the folder", n)
+		}
+	}
+}
+
+// SyncFolderNow runs one pass immediately — the "Synchronise now" button, and
+// what makes a first setup verifiable without waiting for the timer.
+func (a *App) SyncFolderNow() (int, error) {
+	return a.folderSyncPass()
+}
+
+// folderSyncPass reads every peer's new records once and applies the winners.
+func (a *App) folderSyncPass() (int, error) {
+	a.syncMu.Lock()
+	store, _ := a.syncStore()
+	a.syncMu.Unlock()
+	if store == nil || a.qso == nil {
+		return 0, nil
+	}
+	peers, err := store.Peers()
+	if err != nil {
+		a.syncMu.Lock()
+		a.syncErr = err.Error()
+		a.syncMu.Unlock()
+		return 0, err
+	}
+	offsets := a.loadSyncOffsets()
+	var batch []syncfolder.Record
+	advanced := map[string]int64{}
+	for _, p := range peers {
+		recs, next, err := syncfolder.ReadFrom(p.Path, offsets[p.MachineID])
+		if err != nil {
+			// One unreadable peer — a file mid-upload, a share that dropped —
+			// must not stop the others. Its offset is left where it was, so
+			// nothing is skipped when it comes back.
+			applog.Printf("foldersync: reading %s: %v", p.MachineID, err)
+			continue
+		}
+		batch = append(batch, recs...)
+		advanced[p.MachineID] = next
+	}
+	if len(batch) == 0 {
+		a.syncMu.Lock()
+		a.syncLast = time.Now()
+		a.syncErr = ""
+		a.syncMu.Unlock()
+		for id, off := range advanced {
+			offsets[id] = off
+		}
+		a.saveSyncOffsets(offsets)
+		return 0, nil
+	}
+	applied := 0
+	for _, rec := range syncfolder.Merge(batch) {
+		if a.applySyncRecord(rec) {
+			applied++
+		}
+	}
+	// Offsets advance only after the batch has been applied. Saved first, a
+	// crash in between would lose those changes permanently — the records would
+	// never be read again.
+	for id, off := range advanced {
+		offsets[id] = off
+	}
+	a.saveSyncOffsets(offsets)
+
+	a.syncMu.Lock()
+	a.syncLast = time.Now()
+	a.syncReceived += int64(applied)
+	a.syncErr = ""
+	a.syncMu.Unlock()
+
+	if applied > 0 {
+		a.invalidateAwardStats()
+		a.clusterStatusMu.Lock()
+		a.clusterStatusIdx = nil
+		a.clusterStatusMu.Unlock()
+		if a.ctx != nil {
+			wruntime.EventsEmit(a.ctx, "logbook:changed")
+		}
+	}
+	return applied, nil
+}
+
+// applySyncRecord writes one incoming change to the logbook. Reports whether
+// anything actually changed.
+//
+// Deliberately uses the repository directly and NOT AddQSO/UpdateQSO/DeleteQSO:
+// those publish to the folder, and a change applied here would be written
+// straight back out — two machines echoing each other for ever.
+func (a *App) applySyncRecord(rec syncfolder.Record) bool {
+	id, found, err := a.qso.IDBySyncUID(a.ctx, rec.UID)
+	if err != nil {
+		applog.Printf("foldersync: looking up %s: %v", rec.UID, err)
+		return false
+	}
+	if rec.Op == syncfolder.OpDelete {
+		if !found {
+			return false // never had it, or already deleted here
+		}
+		if err := a.qso.Delete(a.ctx, id); err != nil {
+			applog.Printf("foldersync: deleting QSO %d: %v", id, err)
+			return false
+		}
+		return true
+	}
+	var q qso.QSO
+	if err := json.Unmarshal(rec.Data, &q); err != nil {
+		applog.Printf("foldersync: unreadable record for %s: %v", rec.UID, err)
+		return false
+	}
+	if strings.TrimSpace(q.Callsign) == "" {
+		return false
+	}
+	q.SyncUID = rec.UID
+	// Not under this identity — but very possibly the same contact under
+	// another one, or under none: both PCs were seeded from the same ADIF long
+	// before any of this existed. Recognise it rather than log it twice.
+	if !found {
+		if lid, _, ok, err := a.qso.IDByDedupeKey(a.ctx, q.Callsign, q.QSODate.UTC().Format("2006-01-02T15:04"), q.Band, q.Mode); err == nil && ok {
+			id, found = lid, true
+			_ = a.qso.SetSyncUID(a.ctx, id, rec.UID)
+		}
+	}
+	if found {
+		q.ID = id
+		if err := a.qso.Update(a.ctx, q); err != nil {
+			applog.Printf("foldersync: updating QSO %d: %v", id, err)
+			return false
+		}
+		return true
+	}
+	q.ID = 0
+	newID, err := a.qso.Add(a.ctx, q)
+	if err != nil {
+		applog.Printf("foldersync: inserting %s: %v", q.Callsign, err)
+		return false
+	}
+	// sync_uid is not in the insert column list — on purpose, so no ordinary
+	// write can clobber an identity — so it is stamped straight after.
+	if err := a.qso.SetSyncUID(a.ctx, newID, rec.UID); err != nil {
+		applog.Printf("foldersync: stamping the new QSO %d: %v", newID, err)
+	}
+	return true
+}
+
+// GetFolderSyncStatus reports what the operator needs to see: which other PCs
+// are in the folder, when each last logged something, and whether anything is
+// waiting to be read.
+func (a *App) GetFolderSyncStatus() FolderSyncStatus {
+	a.syncMu.Lock()
+	cfg := a.loadFolderSync()
+	store, _ := a.syncStore()
+	st := FolderSyncStatus{
+		Enabled:  cfg.Enabled,
+		Folder:   cfg.Folder,
+		Sent:     a.syncSent,
+		Received: a.syncReceived,
+		Error:    a.syncErr,
+	}
+	if !a.syncLast.IsZero() {
+		st.LastSync = a.syncLast.UTC().Format(time.RFC3339)
+	}
+	if a.settings != nil {
+		st.MachineID, _ = a.settings.Get(a.ctx, keySyncFolderMachine)
+	}
+	a.syncMu.Unlock()
+	if store == nil {
+		return st
+	}
+	peers, err := store.Peers()
+	if err != nil {
+		st.Error = err.Error()
+		return st
+	}
+	offsets := a.loadSyncOffsets()
+	for _, p := range peers {
+		fp := FolderSyncPeer{Machine: p.MachineID}
+		if behind := p.Size - offsets[p.MachineID]; behind > 0 {
+			fp.Behind = behind
+		}
+		if info, err := os.Stat(p.Path); err == nil {
+			fp.LastChange = info.ModTime().UTC().Format(time.RFC3339)
+		}
+		st.Peers = append(st.Peers, fp)
+	}
+	return st
+}

telemetry.go

@@ -21,7 +21,7 @@ import (
 
 const (
 	// appVersion is stamped on every heartbeat (and could feed the About box).
-	appVersion = "0.25.6"
+	appVersion = "0.25.8"
 
 	// posthogHost is the PostHog ingestion endpoint. EU cloud by default; change
 	// to https://us.i.posthog.com for a US project.