Filled, with a hairline white edge — the same trick the home marker
uses. A ring is an outline drawn over whatever is beneath it, and eight
pixels of one over the satellite imagery was barely visible, which is
the whole reason the layer exists to be looked at.
Its own colour, in its own key, beside the switch that turns the layer
on. Separate from the decode colour deliberately: that one exists
because the band palette disappears over some basemaps, and cyan over a
pale sea has exactly the same problem for exactly the same reason. One
control for both would have forced the two layers into a single colour,
which is the distinction it took a shape to draw in the first place.
The picker only appears while the layer is on, since it configures that
layer and nothing else, and it is a portable UI pref like the other map
colours, so it travels with data/.
Two things the first version got wrong.
The arcs. One line per reporting station made a fan out of a single
square, and with a few dozen receivers it buried the decode arcs it was
drawn beside. They earned nothing either: an arc's job on the decode
layer is to say WHICH station a path belongs to, and here every path
starts in the same place. So marks only, and the mark is a hollow
diamond faded by age over the window. Shape rather than colour, because
the decode dots are filled circles and the arcs already use fourteen
colours — a fifteenth reads as another band. A divIcon, canvas being
able to draw circles and nothing else; affordable here at a few dozen
nodes where it would not be against three hundred arcs.
The controls. The basemap row was at left-12, which is 48 px, against a
zoom control that is 30 px of buttons plus its 10 px margin and a
border — so it overlapped, and the minus button took clicks meant for
Street. Then the colour input and the reverse-layer switch were added to
the same row, which grew until it reached the middle of a world map:
covering the Atlantic to spare a top-right corner that was empty
throughout, Leaflet putting nothing there but an attribution that lives
at the bottom. Basemaps stay left at left-16, the other two move right.
The map drew one direction of every path on it: what this receiver
decoded. The reverse — which stations are reporting our own
transmissions — is the half an operator cannot see from their own
radio, and on FT8 it is the half that decides whether calling is worth
the cycle.
It is the narrowest slice of the PSK Reporter feed there is. The v2
topic is
pskr/filter/v2/<band>/<mode>/<tx call>/<rx call>/…
so putting the operator's callsign in the TRANSMIT level makes the
broker send nothing else. For scale, from internal/pskr's own measured
numbers: four bands unfiltered is 83 messages a second, filtered on the
receiver's square 0.2 to 1.2 a second — one callsign in the transmit
level is a handful per FT8 cycle however open the band is. Both grids
are in the payload, so the arc is arithmetic and there is no lookup.
internal/pskrme, with its own connection, for the same reason
internal/pskrtgt has its own: the three want slices of the feed that
cannot be filtered out of one another. It also means this keeps working
with the band-opening watch off — hanging it off that feed's lifecycle
would have made it fail silently for anyone not chasing openings.
Nothing is persisted. One entry per STATION inside a fifteen-minute
window, carrying its freshest report: PSK Reporter's uploaders batch,
many every five minutes, so a tighter window would show a fraction of
who actually heard the last few calls. Stop clears the window, or
switching the layer back on would redraw who heard us before it was on.
On the map the layer is dashed and single-coloured. Solid is what we
decoded, dashed is somebody decoding us; colour alone could not carry
that distinction next to fourteen band colours. The receivers are rings
rather than filled dots for the same reason. Per profile, since the
callsign IS the subscription — a switch resubscribes rather than going
on reporting who hears the previous station.
Every FM bird's uplink, downlink and tone compared against N9KT's Live
FM Satellites table, and the image list beside it. Two corrections and
ten additions:
- AO-123 (ASRTU-1) had no CTCSS. Its 145.850 uplink wants 67.0 Hz, and
without the tone the transponder simply does not answer.
- SO-50 keeps 67.0 Hz, which is the tone that keys it, but its label now
says 74.4 Hz arms the ten-minute timer first. That is a one-off
two-second carrier, not a tone to key every transmission with, so it
belongs where it will be read rather than in a field the radio would
apply throughout.
- Eight SSTV satellites added as receive-only entries, no uplink field
at all, the way the ISS SSTV row was already written: HC1PX
(GALAPAGOS-UTE-SWSU), RS18S (SakhaCube-Cholbon), RS27S (UTMN-2),
RS38S (VIZARD-meteo), RS40S (UmKA-1), RS57S (Monitor-4), RS58S
(Monitor-3), RS83S (Lobachevsky). SSTV downlinks also added to
SONATE-2 (145.880, Martin M1) and QMR-KWT-2 (436.950, Robot 36), and
the ISS row now names its mode.
The NORAD numbers were read out of the element feed by name, not
guessed: five of the eight are in it and carry an exact catalogue
number, and the three that are not (HC1PX, RS18S, RS83S) are joined by
name and alias instead, so they read as having no elements until a feed
carries them rather than pointing an antenna somewhere arbitrary.
AO-91 was left alone. The table shows no tone in its uplink cell but
Fox-1B does require 67.0 Hz, so ours stands and the table is the
incomplete one.
The direction test that kept the wind out was applied to staying in as
well, and there it was wrong. Its unit is a four-degree step, which is
four seconds of travel at a degree a second — far longer than the 1600 ms
settle window — so between two accepted steps nothing said the mast was
still turning, the timer expired, and Stop went dark and lit again the
whole way round.
Entry stays strict: two consecutive steps of four degrees the same way,
which is what weather cannot produce. Staying in is now sensitive
instead — any continued progress in the direction already established
re-arms the timer, down to a single degree, because one degree the same
way is not a gust when the rotor is demonstrably under power. That
demonstration is exactly what the strict entry provides, so the loose
exit costs nothing: wind still cannot start it, and once stopped the
sensitive path is disarmed with the direction.
The settle window goes to 3000 ms so it outlasts one degree of progress
on a slow mast rather than four degrees on a fast one. Measuring
progress needs the PREVIOUS reading, which movementReferenceRef is not:
it deliberately holds still through sub-threshold steps so they can
accumulate into one.
"Later" set updateInfo to null and nothing else, while the check behind
it runs on a five-minute interval — so the notice came back four times
an hour, all evening, for a version the operator had already declined.
The button was doing exactly what it said and was useless anyway.
It now offers 1, 4, 12 or 24 hours, and the deferral is recorded with
the VERSION it applies to. That scoping is the part that matters: a
release newer than the one put off is different news and appears at
once, so a snooze can defer an update but never bury one.
The cross hides it for an hour — the shortest of the four — rather than
for the five minutes until the next poll, which is what made the notice
feel broken. The background check honours the deferral; checkUpdateNow,
behind About, deliberately does not, because asking is asking.
Kept out of the portable UI prefs: a machine told to wait four hours has
said nothing about the operator's other machines.
The tab read the followed list, the horizon and the locator once, when it
was mounted. It is normally open BEHIND the settings window while those
are edited, so un-following QO-100 left it in the pass table and in the
dropdown until the tab was reopened — indistinguishable, from the
outside, from a setting that had not saved at all.
SaveSatSettings now emits sat:settings, and the panel reloads the birds
and the passes on it, the same way it already does on sat:tle. One event
for every key: they all change what the tab should be showing. The
selection needs nothing new — a satellite that has stopped being
followed drops out of the dropdown, and the effect that keeps the
selection inside that list moves to the first one remaining.
The FTx button is gone. It was the wrong grain in both directions: the
Digital class already sat a contest RTTY square next to an FT8 one, and
FTx then sat FT8 next to FT4, when what this map answers is where ONE
mode has been heard. The four classes stay as buttons and a dropdown
beside them offers a single mode.
Its contents come from the log, not from a list in the code. That is
what settles FT2: it is not a registered ADIF mode yet, so a hardcoded
list meant either leaving out the operators already using it or shipping
a mode that does not officially exist. A query does neither, and needs
no change here the day it is registered. The mode offered is the SUBMODE
where there is one, because ADIF files PSK63 under PSK and "PSK" is not
the name anybody is looking for.
The band list is the station's own, unioned with anything worked outside
it so nothing in the log is unreachable, ordered by frequency through
the band plan's index — sorting the names puts 10m between 1.25m and
12m. The satellite list is drawn from SAT_NAME on the squares
themselves and the control is absent altogether on a terrestrial log: a
VHF square worked through AO-91 and one worked line-of-sight are not the
same achievement, and until now nothing separated them.
The mode dropdown shares the scope with the class buttons rather than
narrowing on top of them — mode is one question, and two controls both
answering it is how a map ends up showing PHONE ∩ FT8 and nothing else.
A stored FTX preference is read as Digital, so it cannot leave the map
filtered by something no control shows as selected.
Settings ▸ FlexRadio stores the band→antenna map keyed by the band name
in capitals — that is what the panel writes and what the entry form reads
back. applySatRadio looked its two bands up through bandForHz, which
returns the band plan's own spelling ("70cm"), so both lookups missed,
both antennas came back empty, and the early return left the pass on
whichever antenna the radio was last used on. An operator with XVTA on
2 m and XVTB on 70 cm had configured exactly the thing being ignored, and
nothing said so: the downlink ran through the wrong transverter in
silence.
The key is now computed by flexBandAntKey, which exists so the next
caller cannot make the same mistake, and a test pins the case in the
direction that broke. Both outcomes are logged — the resolved antennas,
or the bands nothing was configured for — because from the outside a
setting never made and a lookup that missed look identical.
Alongside, three things the same pass made obvious:
- Tracking shows the satellite's own azimuth, elevation, distance and
altitude beside the frequencies. The strip held where the ANTENNA was
pointing but not where the bird was, which is what says whether a pass
is worth calling on. Elevation dims below the horizon so a satellite
followed before its lever cannot be read as workable.
- Both satellite lists in the settings are sorted by name with the
numbers taken as numbers. The available column followed the order
birds.json happens to be written in and the followed column the order
of the clicks, so finding one bird among sixteen meant reading all
sixteen.
- A followed satellite that has since been renamed is resolved through
the plan's aliases. LILACSAT-2 became LO-90, and the followed list is
stored as plain text, so the bird the operator had chosen appeared as
having no elements while the same satellite sat in the available
column under its new name. Resolved in GetSatSettings rather than
satSettings, which is read at startup before the plan is loaded.
"la lilacsat je le vois en DATA ???!" — and the LO-90 fix could not have reached
him. The satellite file is copied out on the first run and was the operator's
from then on, so the merge could only ADD birds, never repair one. LilacSat-2
went out with an APRS digipeater and no FM transponder, and that mistake had
become his data, permanently.
The merge now distinguishes three cases, and the middle one is the whole point:
- a satellite he does not have is added;
- one he has, UNCHANGED from the plan he was given, is replaced — he never
edited it, so it is not his to keep: it is our data and ours was wrong;
- one he EDITED is left exactly alone, and named in the log. A frequency
somebody corrected by hand outranks anything shipped; they were on the air
and we were not.
"Unchanged" is decided against a baseline — satellites.shipped.json, the plan
this station was last handed — so the comparison is with what THEY were given
rather than with whatever ships today. Their edits survive every future release,
not just the next one.
The first run after this has no baseline, and there an edit of theirs and a
mistake of ours are indistinguishable. The shipped plan wins, once, with the
whole file copied to satellites.json.bak first and every replacement named. The
safe-looking alternative was the wrong one: standing down would have written a
baseline recording their entry as "edited" and frozen a known-wrong frequency
for the life of the install.
Verified against his own file: LILACSAT-2 becomes LO-90 with the FM transponder
first, and the twelve curated entries that were missing their catalog numbers
get them — which also closes the NORAD gap left open when the exact join went in.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Shipped with nothing but an APRS digipeater on 144.390, which is not what
anybody works that satellite on. The FM transponder — 144.350 up, 437.200 down —
was dropped by the generator, correctly by its own rule: SatNOGS marks that
transmitter inactive.
And SatNOGS is not wrong. LilacSat-2's transponder is switched on to an
announced schedule rather than left running, and from a database a scheduled
transponder looks exactly like a dead one. The judgement belongs to whoever
reads the diff, not to the filter.
So the transponder goes back in, labelled "(scheduled)" the way PO-101 already
is, and the satellite takes its OSCAR name, LO-90, with LILACSAT-2 and CAS-3H as
aliases so the element feeds still meet it.
The generator now reports what it refused on that ground and kept nothing else
for — one line per satellite, phrased as a question. Reviving every inactive
transponder would fill the list with birds that answer nothing; saying nothing
is how this one shipped wrong. There is no third option that a filter can decide
on its own.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
"I thought we had fixed this" — and we had not, we had only raised the number.
The antenna moves a few degrees in the wind and the Stop button lights up and
goes out, all evening, on a mast that has not turned.
A threshold cannot tell the two apart at any setting. A gust pushes a beam past
four degrees and brings it back — 100°, 105°, 100°, 106° — and every one of those
excursions cleared the threshold. Raising it only raises the wind speed it takes,
and blunts the detection of a rotor genuinely creeping.
What separates a rotation from the weather is not amplitude, it is SIGN. A rotor
under power advances; a gust reverses. So a step is only movement when the
previous step went the same way: wind gives +5, −5, +5 and never two in a row,
while a rotor gives +4, +4, +4 and is announced on the second — one poll, about
a second, on a mast that takes half a minute to cross a pass.
The threshold stays, doing the job it is actually good at: ignoring the degree
or two of potentiometer noise that is not a step at all.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
v0.27.22 was cut at ce7b368, and the SteppIR Calibrate button and the rotor-dial
fix landed after it. Their entries were appended to a block that had already
shipped, so an operator reading the What's new for the version they are running
would have been told about two things it does not contain.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A G-2800 sitting at 020° turned anticlockwise reports 020, 010, 000, 359, 358 …
340. The dial fed those straight into a CSS rotate, so the transition travelled
from 20deg to 359deg the long way: a complete revolution on screen while the
mast moved forty degrees in the opposite direction. On a rotator with an overlap
that happens on every pass through north, which is where the antenna spends much
of its time.
unwrapRotation was already there and already did the right thing — it was only
wired to the HOVER beam, where the mouse crossing north had made it obvious.
The antenna's own beam used the raw azimuth.
So the antenna angle is accumulated across renders too: 020 → 000 → −001 →
−020, which is the way the mast is moving. Both lobes of a bidirectional antenna
get their own accumulator, since they cross north at different moments, and the
ref only advances when the input changes so a re-render for any other reason
cannot make the beam creep.
The classic dial is unaffected: it draws its needle from SVG coordinates, with
no transition to send anywhere.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Calibrate was in the protocol notes at the top of the package ('V') and nowhere
else. It drives every element to its end stop so the controller re-learns where
zero is, and it is the cure for the one SteppIR fault an operator cannot reason
about: an antenna that tunes to the wrong length. The controller counts steps
from a remembered position, so once that memory is wrong — a power cut in
mid-move, elements pushed by hand, a motor that slipped — every frequency after
it is wrong by the same amount, silently.
It asks before it runs. Minutes, every element travelling its full length, and
the antenna unusable until it finishes: right answer to a mistuned beam, wrong
answer to a stray click during a contest.
It sits beside Retract as a two-column row rather than a second full-width
button. The docked widget is narrow and already tall, and the pair reads as what
it is — the two commands that move every element at once.
Retract (Home) was already there and worked on a SteppIR; it now carries a
tooltip saying what it is for, because "retract elements" does not tell an
operator that this is the storage position and that the next tune brings them
back out unaided.
An Ultrabeam controller has no calibration command, and the adapter says so
instead of pretending — the antenna interface is shared, and a button that
silently does nothing is worse than one that explains itself.
Untested on hardware: the byte is from SteppIR's own Transceiver Interface note,
which is the same source the rest of this package was cross-checked against.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
An operator spent three hours setting up, accepted the update, and reopened a
program that had forgotten everything. It is the second such report.
The updater is not the culprit — it touches its own exe and nothing else. The
pointer is. config.json is the only record of WHERE the database is, and it was
written with os.WriteFile: truncate, then fill. A process that stops between
those two steps — a crash, a power cut, an update's watchdog force-exiting the
old instance — leaves the file empty. readBootstrap then swallowed the parse
error, returned an empty pointer, and startup read that as "no database chosen":
it created a NEW one at the default path and opened it. Three hours of work
still on disk, and an application presenting itself as freshly installed.
Three changes, in the order they defend:
- The write is atomic. A temporary file, fsync'd, renamed into place — and a
rename within a volume cannot publish half a file. The previous contents are
kept as config.json.bak, because a pointer is a few dozen bytes and an
evening of configuration is not.
- A pointer that EXISTS and cannot be read is no longer treated as no pointer.
It is restored from the backup, and when there is nothing to restore from
the broken file is KEPT as config.json.broken — it is evidence, and it may
still be readable by hand.
- Creating a new, empty settings database in a folder that already holds a
full one is now said out loud, in the startup log and on screen. Nothing is
deleted and nothing is guessed — guessing which file is theirs is how the
wrong one gets opened — but the message names the other file, which is where
their settings still are.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Reported with two screenshots: both slices in USB on an inverting transponder,
slice B sitting at exactly 435.100000, and the red TX badge on the 2 m DOWNLINK.
The log named the cause in one line:
flex: satellite armed (rx slice 0, tx slice -1)
Creating a slice is asynchronous — "slice create" is answered later with the
index — and SetSatellite returned without waiting. Everything downstream then
ran against an uplink of -1 and silently did nothing: no antenna, no CTCSS tone,
no sideband, never tuned, and never sent "tx=1". So the radio went on
transmitting on the downlink, which is the one failure here that puts a signal
where it must not go, and the frequency and mode on screen were simply the ones
the slice had been created with.
Three fixes, because the ordering can fail in more than one way:
- Arming waits for both indices before reporting the pair armed, and says so
plainly when the radio does not produce them.
- The uplink is adopted from the SLICE STATUS as well as from the create
reply. The status needs no sequence-number correlation: if satellite mode is
armed, the uplink is unknown, and a slice is in use that is not the
downlink, that is it — the radio is saying so.
- What the uplink is owed is remembered — its mode, its antenna, its tone —
and given to it when it appears. Those three are sent ONCE; only the
frequency is re-sent every tick, so a slice that arrived late kept nothing.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Correcting the previous commit, which offered a 450° rotator range for the
Rotator Genius. It cannot do it.
The evidence is the operator's own box and 4O3A's manual together. His Rotator
Configuration reads "Limits: 5 to 4" — that is where the mechanical stop sits
within ONE turn, a dead zone at four and a half degrees, not a range of travel.
The manual is unambiguous about what happens past it: "you will not be able to
give it a target beyond the limits". A 450° mast on a Rotator Genius is a 450°
mast used as a 360° one, and that limit belongs to the controller.
So the setting goes. Offering an operator a 450° option that the box can only
ever refuse is worse than not offering one — it spends their evening proving
that the software was wrong about their station.
What stays is the half that was genuinely ours: GoTo no longer clamps to 360
before sending, and the limits the Genius reports on every heading query are now
read instead of skipped over. The overlap branch is driven entirely by what the
device answers — no setting, no assumption — so a controller that one day
reports a range past 360 is driven through it without a line changing here.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
An operator with a 450° mast watched the Genius stop at 359 and had to press
"clockwise" by hand to get through north. Half of that was ours: GoTo clamped
every target to 360 before sending it, although the wire command carries three
digits and always could have said 370.
So the clamp goes to 450, the rotator-range setting is offered for the Rotator
Genius like the other backends it applies to, and when a bearing can be reached
two ways the nearer one is taken — 010° sent as 370° when the antenna is already
at 350°, which is the entire point of having an overlap.
THE GENIUS DECIDES WHAT IS REACHABLE. It reports the limits it is configured
with, and they are the truth about what is bolted to the tower. This particular
station's box says "5 to 4" — the factory 360° range — and would refuse 370,
turning a working command into a rejected one. So the overlap is used only when
the Genius itself says it has one, and when OpsLog is set to 450 while the Genius
is not, the log says so once a minute and names the dialog to change: the setting
lives in the Genius's own Rotator Configuration, and nothing here can reach past
its limits.
The |h reply always carried those limits and they were being skipped over.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
THE FLICKER was a real bug and not a rotator problem. The rotator is asked where
it is at most every three seconds — a controller query binds a socket and waits —
but satTrackStep built a fresh status every second and carried over only Radio
and Error. So the antenna readout was filled in on one tick in three and blank
on the other two, which on screen is a rotator that keeps disconnecting. The
status is now rebuilt FROM the previous one, so a field nobody wrote this tick
keeps the value somebody wrote last tick.
THE HEADER now carries the two frequencies and the antenna bearing, beside the
button that started tracking. During a pass an operator watches the radio and
the antenna, not a column on the far side of the window — and that column is the
first thing they hide to get the map full width, which until now took the
numbers with it. The compass spins while the antenna is still on its way: a mast
takes tens of seconds to cross a pass, and "moving" against "stuck" is the whole
reason to look at it, which a number alone cannot show.
THE PRECISION drops from one hertz to a hundred. The Doppler moves about sixty
hertz a second on 70 cm, so the last two digits changed on every tick and the
display was a blur that could not be read and did not need to be. The radio
still gets the whole figure — the correction is computed and sent to the hertz —
this is only how much of it is worth putting in front of somebody. The shift
beside it says "+9.7 kHz" rather than "+9741 Hz", which is how it is read aloud.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
An operator with a WinKey2 USB had to run K1EL's WKdemo and close it again
before OpsLog could talk to the keyer at all. That workaround is the diagnosis:
closing another program does something to the keyer that opening the port does
not, and whatever state it was stuck in survives a failed connect.
So the second handshake attempt now does what closing WKdemo does, in an order
that survives each step failing:
- Host Close, in case the keyer is still in host mode from a session that
ended without one — a crash, a cable pulled, a machine switched off. It has
been waiting ever since for a host that went away.
- Admin Reset, which returns it to its power-up state. A parser stuck
part-way through a command whose parameters will never arrive cannot be
talked out of it any other way.
- A DTR pulse, which is what closing a program actually does to the line. On
a WKUSB and on every Arduino-based clone, DTR runs to the processor's reset:
it is a power-on reset in all but name.
RTS is left alone throughout — on a serial WinKeyer it is the negative rail the
RS-232 swing comes from, and driving it starves the chip.
A keyer that answers the echo and then refuses to open is the same leftover
session seen from the other side, so that case sends Host Close and asks again
rather than reporting a keyer that demonstrably just spoke to us as absent.
And a port already known to need the slow path gets the wake-up on the FIRST
attempt from then on: making the operator sit through a failure to earn it again
doubles the connect time for no new information.
The handshake bytes are already logged on every connect, so the next report of
this shape says where it stopped.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Two birds are often up at once, and switching between them left the frequencies
on the first. The selection on the page is the DISPLAY's; the tracker held its
own name and went on following whatever it was started with. The way through was
to stop tracking and start it again, which is how it was found — and which is
also how a Flex throws away and rebuilds both its slices for no reason.
RetargetSatelliteTracking changes what is being followed without letting go of
the radio or the rotator. Everything derived from the old satellite is cleared so
the next step sets it afresh: the frequencies, the mode on both slices (set once
per satellite, not per tick), the antennas and the CTCSS tone — the new bird may
be U/V where the old one was V/U, which swaps which slice sits on which band.
And it happens at once rather than at the next tick. The loop gained a wake
channel: a second of the previous satellite's frequencies is a second of the
wrong pass, and the antenna would otherwise wait for the new bird to drift a
step away from where the old one happened to be.
Selecting a satellite with the radio idle is unchanged — a look, not a command.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Three things the tracker was leaving to chance on a FlexRadio, all reported from
a real pass.
ANTENNAS. Settings ▸ FlexRadio holds a per-band RX/TX antenna map, and it was
applied in exactly one place: the entry form, on a band change, to the active
slice. A pass never goes through that path — the tracker arms two slices itself.
So both were left on whatever the radio last used, and a station with
transverters (XVTA on 2 m, XVTB on 70 cm) heard nothing at all, having
configured precisely the thing being ignored. The two slices are on two
different bands, so they cannot share one setting: the downlink takes the
receive antenna for ITS band, the uplink the transmit antenna for its. Per
slice, not through sendSlice, which addresses whichever slice is active — during
a pass that is the downlink, so the uplink would never have been set.
SIDEBAND. satMode forced USB above 30 MHz on both sides. An inverting
transponder turns the passband over, so lower sideband up comes back as upper
sideband down: FO-29, RS-44 and AO-73 were being worked with the operator's own
audio going through upside down. The tracker now decides both sidebands from the
transponder's inverting flag and passes them separately; a bare "SSB" still
means USB, so nothing else changes.
CTCSS. Nothing set it, on any bird. The frequency plan has carried the tone all
along — 67.0 on SO-50 and AO-91, 141.3 on PO-101 — and an FM repeater does not
answer without it, which is indistinguishable from a satellite that is not
there. It goes on the uplink slice, value before mode so the radio cannot
transmit the previous tone in the gap between two commands.
Written against the SmartSDR slice API and UNTESTED on hardware.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Reported as "the Doppler moves the frequency enormously", and it did: a 2 m
downlink was being shifted two megahertz across a pass instead of three
kilohertz, in the wrong direction.
The propagator library reports a range rate that is not one. Measured against
the range it is meant to be the derivative of, on this station's own cached
elements:
PO-101 library -1457.3 km/s measured +6.227 km/s
ISS library +2036.8 km/s measured -5.522 km/s
Wrong by a factor of some 250 and of the wrong sign, so the correction both
overshot and pushed the operator away from the station they could hear. Nothing
else was affected — the elevation and the passes come from the look angle, which
is right — which is why this survived: the satellite was in the correct place on
the map while the radio was told to go megahertz away from it.
So OpsLog computes it itself, as the difference between two ranges a second
apart. That cannot be wrong in either magnitude or sign: it differentiates the
very number the panel displays. Two extra propagations per call, which is
microseconds.
Two tests pin it, and both fail against the old behaviour: a range rate faster
than orbital velocity is a units mistake, and the Doppler on the two bands
satellites are worked on has a textbook size — about ±3.5 kHz on 2 m, ±10 kHz on
70 cm.
cmd/satdiag is the throwaway that found it, kept because the next report of this
shape ("the frequency moves oddly", "that pass is not real") is answered by the
same three numbers: which elements the satellite resolved to, the range rate
reported against the range rate measured, and the Doppler each transponder gets.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Asked why the satellite dropdown exists when the passes are on the right, and
the honest answer was that the table was incomplete. Three kinds of satellite
never appeared in it: QO-100, which has no pass because it never sets; one whose
elements have not arrived, which cannot be predicted at all; and one whose next
pass falls beyond the prediction window. Following a satellite and not finding it
in the list reads as OpsLog having lost it.
They are listed at the end now, each saying which of the three it is, and
clicking one selects it exactly like a pass row. The table stops being "what is
coming" and becomes "what you follow, soonest first" — which is what an operator
was scanning it for anyway.
The dropdown stays, and now for one reason rather than by default: it is the
only selector left when the readout column is hidden, which is what an operator
does when they want the map full width during a pass.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A satellite tracker that insists on an elevation motor is a tracker switched off
for nearly everybody. A pass at the edge of the footprint — which is most of
them — never climbs above ten or fifteen degrees for its whole length, and a
yagi's beamwidth swallows that: the bearing alone is enough, and it is how most
stations that work satellites are actually built. The same switch rescues an
az/el station whose elevation motor has failed.
So it is an option, not a silent fallback, because it does cost something: a
bird straight overhead is a moving azimuth and a bearing that means nothing, and
whether to accept that is the operator's call. With it on, any rotor in the list
can be chosen — the PstRotator, the Rotator Genius, the ARCO, the tower already
turned for HF.
That works because the per-backend command dispatch moved out of the three
RotatorGoTo/Stop/Heading methods into linkGoTo/linkStop/linkHeading, so the
satellite tracker drives any of the seven backends through the same code the
compass uses instead of a second implementation of each. GetRotatorHeading loses
sixty lines of near-duplicate switch in the process, and a rotor with no
elevation axis now says so (HasElevation) rather than reporting a zero that
looks like a real bearing.
One trap, with a test on it: the step check compared both axes, so with the
elevation never commanded its difference stayed above the step for the whole
pass and every tick ordered the antenna to the bearing it was already on. A mast
has a finite number of turns in it.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Both were already on screen, in a grey eleven-pixel footnote under the
frequencies, which is not where either of them belongs.
FM and SSB are two different evenings. One is a channel, a tone and a handheld;
the other is a passband, a beam and a VFO walked as the Doppler moves. So the
mode is a coloured badge — in the readout, in the transponder list where a
satellite offering both makes the choice matter, and in the header so it
survives hiding the readout column, which is exactly what an operator does when
they want the map full width during a pass.
The tone gets the weight of a frequency because that is what it is worth: a
repeater called without it does not answer, and what the operator hears is an
empty channel and concludes the satellite is not up. And when a bird needs no
tone it now says so — a blank line could not tell "no tone" from "OpsLog does
not know", and the difference is a pass.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
25 satellites, typed by hand and never revisited. Eight of them were the
first-generation Tevel constellation, which re-entered in 2024; four more had
come down too; and the nine Tevel-2 satellites that replaced them, the Chinese
space station, AO-27, AO-123 and twenty others were simply absent. So: 44
satellites now, and a generator instead of a memory.
cmd/satgen joins three public sources on the NORAD catalog number — Celestrak's
amateur group and PE0SAT's mirror for which birds OpsLog can actually get
elements for, and SatNOGS DB for the transmitters. It is a one-shot tool, run by
hand, the same arrangement as cmd/cntygen, and it is deliberately conservative:
- It never destroys a curated entry. The hand-written plans hold things
SatNOGS does not reliably carry — a CTCSS tone, the QO-100 passband as
operators describe it — so an existing bird keeps its data and only gains
its catalog number.
- It prunes on the re-entry date, which is a fact SatNOGS publishes rather
than a judgement about which of the missing satellites are missing for good.
- It refuses a digital uplink that does not say what it is. A GMSK uplink is a
command channel far more often than a digipeater, and shipping the wrong one
invites somebody to transmit on a control frequency. An analog uplink with
both ends is a contact by construction, which is what catches the repeaters
that describe themselves only as "Mode V/U FM".
- Its output is deterministic. One satellite can hold two catalog entries —
GreenCube is 53106 in one feed and 53109 in the other — and iterating a map
picked a different one each run.
A bird now carries its NORAD number, and that is how its elements are found.
Names were the only join before, and they are written differently by every party
involved: "TIANYAN 01" and "TO-108" are one satellite that had never once met,
so TO-108 tracked nothing at all.
And the plan now reaches a station that has already run OpsLog. The editable
copy was written on the first launch and was the operator's list for ever after,
so a release adding nine satellites reached nobody who had opened the tab. It is
merged on each load instead: a satellite they already have is untouched, edits
and corrections included, and only the ones they have never seen are added.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A footprint is thousands of kilometres across. A dozen of them overlap into a
wash of translucent circles that hides the coastline, the ground track and the
satellites themselves — and the question a footprint answers, "can I hear it",
is only ever asked about the bird being worked.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
An unselected satellite was a four-pixel dot with a one-pixel white edge, and
the basemap decides whether that is visible at all: a pale marker vanishes into
pale terrain, a grey one into a dark ocean, and the operator can switch between
the two in one session. So: a dark halo under a white ring around a larger dot.
Two rings, because no single colour reads on both.
The ones above the horizon now carry their name. Not all of them — a dozen
labels is a map nobody can read — but the two or three an operator is choosing
between right now, which is what saves hovering each grey dot in turn to find
them.
Hovering said "name · elevation · altitude", none of which decides anything.
What decides whether to reach for the radio is how long is left, how high it
will get and where to point, so the tooltip now carries the pass: elevation and
azimuth with its compass point, distance with an arrow for closing or receding,
rise and set with a countdown and a direction, and the peak. A bird already in
view shows its SET countdown instead of its rise — that is the number that
matters at that moment. No pass in the window says so, because a blank reads as
a fault. It costs no extra prediction: the pass list on screen is indexed by
name, and the first entry for a name is its next pass.
The tooltip also had to stop closing itself. The layer is rebuilt every five
seconds as the birds move, and a rebuilt marker is a new marker, so the detail
being read disappeared mid-sentence. The map now tracks the pointer and reopens
the tooltip of the dot it is still on — that one only, so nothing hangs open
once the mouse has moved away.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
The satellite page configured its own EasyComm or PstRotator link while five
other backends were configured in the rotator list. An operator with one az/el
mast therefore described it twice, and could describe it differently the second
time — a station that works on HF and not on a pass, for no reason visible
anywhere on screen.
Now every interface lives in Settings ▸ Rotator, once, and the satellite page
stores only a KEY into that list plus the tracking policy that is genuinely its
own (minimum elevation, step, park). The key and not the index: deleting the
first rotor must not silently point the tracker at a different mast.
migrateSatRotator() turns an existing satellite link into a real entry in the
list, selects it, and clears the old keys so it cannot run twice.
Which rotors have an elevation axis is now a question with one answer, in Go:
rotatorTypes plus rotorHasElevation, exposed to the panel by GetRotatorTypes.
The dropdown, the labels, each backend's default port and default baud all come
from there, so TypeScript no longer keeps a second copy of the same knowledge to
drift out of step. Three cases do not follow from the type alone and are treated
as such: PstRotator forwards elevation to a mast that may not have any, so the
operator says; a SPID's dialect decides (Rot1Prog has no elevation in its reply
format); and an ARCO and an ERC-M speak the same GS-232 while only one of them
lifts.
Each interface carries an Az / Az+El badge beside it. The satellite rotor
dropdown LISTS the azimuth-only ones, disabled, rather than hiding them: an
operator who owns one rotator and does not see it concludes OpsLog cannot find
it, where a greyed row saying "azimuth only" teaches the actual thing.
ERC-M by DF9GR is new — the az/el interface for a Yaesu G-5500. It emulates
GS-232, so internal/rotator/gs232 grew the elevation half: W for a two-axis
move, C2 to read both, falling back to C+B for the firmware that answers C2 with
the azimuth alone. That fallback is the point of the parser tests: reading such
a reply as "elevation zero" would put the antenna on the horizon, which is the
one wrong answer that looks plausible.
EasyComm II is promoted to an ordinary rotator interface, so it can also turn
the antenna from the compass and from a spot click.
The ERC-M is UNTESTED on hardware. Its Test button reads BOTH axes rather than
just the azimuth, so a controller wired for azimuth alone says so there instead
of during a pass.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
The dial painted its background as a full-bleed square, so inside the
rotor widget — which is already a card — it read as a hole punched in the
panel rather than an instrument sitting on it. It is a circle now, at the
radius the map already used, and the corners are left to whatever it is
drawn on. The wrapper loses its own border and background for the same
reason: one card, not two.
And the continents were barely there. At #202832 on a #0B1015 ground the
land was some eight per cent brighter than the sea — technically a map,
practically a dark square with a suggestion in it. The new shades read as
coastlines while staying well under the beams, which are what the dial is
actually for.
The palette stays deliberately unthemed. That was never the problem: a
map that repaints itself in four colour schemes stops being readable, and
the beams' green, orange and yellow have to mean the same thing in every
one of them.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
The rigs and antennas are already declared once, in Settings ▸ Operating
conditions — a station per rig with its antennas hanging off it — and
then typed again into every contact. That is work, and it is a source of
spellings that do not match: "IC-7610", "IC 7610" and "ic7610" are three
different rigs to an award, to a filter, and to anyone reading the log
later.
Both fields now offer that list, in the entry form and in the QSO editor,
and the antenna field offers the antennas of the rig that was picked
because that is the structure the tree already has. It falls back to all
of them for a rig the tree does not know, so an operator typing a
borrowed rig is still offered their own antennas rather than nothing.
Free text stays allowed throughout — a contact made from somebody else's
station, or imported from another logger, carries a rig that was never in
this tree and must still be loggable. Same rule the satellite-name field
follows.
And the decodes list gains a ceiling of two thousand rows. The rolling
half hour was never a limit on a crowded evening — three decoders put
several thousand rows inside it — and the panel slows down long before
the age cut removes any of them, because each row is a layout, a status
lookup and a distance. Past the ceiling the oldest go: what has already
been scrolled past, rather than the period being read.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A Flex 6700 owner's log: no connection attempt of any kind, and six
saves each answered with
cat: settings saved, link unchanged - staying connected
His settings were right — FlexRadio selected, 192.168.1.154, port 4992,
and the detector had found and named his radio back at him. What was off
was the master CAT switch, which sits above the radio dropdown.
Three things, because each of them failed him on its own.
The log line was false on both counts: a disabled CAT has the signature
"off", so every save took the unchanged branch and announced a link that
was staying connected when nothing was connected at all. It now says CAT
is switched off, which is the fact he needed and the only place it could
have reached him.
The panel says it too, where he was actually looking: a notice under the
switch while it is off, because everything below it can be perfectly
right and still connect to nothing.
And choosing a radio now ticks the switch. Picking a brand, typing an
address, running a detector and clicking the radio it found are all one
gesture — "connect to this" — and making the operator find a separate
master switch afterwards is a trap. Nothing here ever turns CAT off on
its own; they can still untick it.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Reported on an IC-7610 shared through OmniRig: clicking a CW spot put
the radio in CW-R every time, and the only way out was to edit the
OmniRig rig file by hand.
OmniRig's model has two CW modes, PM_CW_U and PM_CW_L, and nothing in it
says which one an .ini calls plain CW. Icom rig files disagree with each
other — on some PM_CW_U is CI-V mode 0x03 (CW), on others 0x07 (CW-R) —
so asking for "CW" is a question with two right answers and OpsLog was
only ever giving one of them.
Settings ▸ CAT ▸ OmniRig now carries the answer for THIS rig, and it is
pushed to a radio that is already connected rather than waiting for a
reconnect: which bit a mode maps to is not worth dropping the link, and
with it WSJT-X's rigctl session. Same shape as the Yaesu RTTY sideband,
for the same reason.
Nothing changes for anyone whose rig file already agrees with us.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
HAMLOG.online was added after most profiles were configured, so it had no
entry in the shipped defaults and no stored value either: it came back
blank, and blank is not a status anybody chose. Every service now starts
the same way — the sent side at R, the received side at N — and a blank
left by a service that did not exist when the operator last saved is
filled in from that. A status they chose themselves is untouched.
Two tests hold the line: no sent side may default to Y, and no field may
be left without a default. Y means "already sent", so it makes the
uploader skip the contact for ever — an operator with eQSL Sent at Y had
a logbook that never reached eQSL, and the only trace was one line in the
application log.
Wiki, both from operator reports:
QSL Management opens with Confirmations — what the page actually is (the
status stamped on every new QSO, not an action), what each status does,
and the warning about Y in the plainest words available, because it fails
silently and by design.
Digital Modes and GridTracker is new. Unicast and multicast explained
from the operating problem rather than the networking: one letterbox that
two programs watch, against a broadcast everyone can tune to. It carries
the real evidence — two starts of one station an hour apart, decodes in
the second and none in the first, the only difference being whether
GridTracker or OpsLog reached port 2237 first — then the settings for
WSJT-X, JTDX, MSHV, GridTracker and OpsLog, the 127.0.0.1-in-the-group-box
mistake, what to do if unicast is unavoidable, and how to check it from
the log.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Reported from a station whose two WSJT-X rows had been dead for weeks:
both were ticked multicast with 127.0.0.1 in the group box, and both
failed the join on every interface with
setsockopt: l'adresse demandée n'est pas valide dans son contexte
which names nothing the operator typed and does not say what is wrong
with it. A multicast group runs 224.0.0.0 to 239.255.255.255; 127.0.0.1
is loopback unicast, and it is an understandable thing to type — it is
the address every other field in every other program wants.
The address is now checked before the join. When it is not a multicast
one the row listens on unicast instead, which is what such an address
means, and the log says why it is not multicast. The row works, and the
reason is in a sentence rather than in a kernel error code.
0.27.18 went out before it was committed, so it does not belong in that
block — and it is the one entry an operator on 0.27.18 needs, since it is
their update that will not come back.
The relaunch after an update stopped working, and the regression is mine:
removing the PowerShell helper — which is what Defender was reading as a
dropper — also removed the wait it was doing. Nothing took over the job.
The numbers made it certain rather than unlucky. The instance being
replaced is allowed THIRTY seconds to shut down (armExitWatchdog forces
it out at that point) because it closes a remote logbook, a CAT session
and sometimes a backup. The new instance was patient with the
single-instance mutex for TWENTY. On any station where shutting down ran
past that, the new process gave up and exited in silence: no window after
an update, and the previous OpsLog still in the task manager. Exactly the
report.
Both relaunch paths now pass --wait-pid, and the new process waits on
that process's handle — a plain kernel wait, which ends the instant the
old one ends, however long or short that is, and looks nothing like a
script starting another program. The mutex retry stays as a backstop and
goes to forty-five seconds, so it is longer than the wait it exists for
rather than shorter.
And when the old process really has not gone, the message says that
instead of "OpsLog is already running" — after an update the operator did
not start a second copy, and what they need to know is which one to
close.
A test keeps the two spawn sites honest: a relaunch added without
--wait-pid is this bug again.
Once the callsign stopped giving up its width, the RST boxes were the
next widest things in the row and became the ones that shifted when the
date field appeared. Both are now a notch narrower and shrink-0, like the
callsign; "59+20" is the longest report either ever holds and still fits.
Its row gains a date field for a manual entry, and a flex row makes room
by shrinking its children — so the widest box, which is the one the eye
is on while typing, was the one that visibly narrowed.
A notch narrower and shrink-0: the size it will always be, with or
without the date, and the slack now comes from the boxes beside it.
0.27.17 went out, so everything committed after its release commit needs
a block of its own — the padlock, the duplicated section name, the
digital row of the matrix and the MQTT chip.
The backdrop-blur fix goes with them rather than staying in 0.27.17. It
had been folded into that version's typing-lag entry, but it is not in
the binary 0.27.17 shipped: what that release actually carries is the
cluster-macro half, which is what its entry now says again.
Three things reported together.
Preferences said the section name twice: a small line above every panel
repeated the heading right underneath it — "GENERAL" over "General" —
while the sidebar next to them already shows which section is open,
highlighted. Two lines and a highlight for one fact; the small line goes.
The band matrix opens on the digital row the operator chooses. Its
digital row has always rotated — DIGI, then each digital mode in their
own list — but it always STARTED on DIGI, so somebody who only works FT8
clicked through to their own mode on every callsign. Settings ▸ General
now says where the rotation starts, and the dropdown offers exactly the
modes the matrix rotates through (the phone-mode rule is now shared
rather than copied, so the two cannot drift). DIGI stays the default: it
is the right answer for anyone working several digital modes.
And the MQTT chip is gone from the status bar. That is the name of a
message protocol, not of anything an operator has — a chip in the status
bar has to say what it is about, and this one told nobody anything. What
it carried is shown in the Chase New panel, which is the place that uses
it; its poller went with it.
Logging a contact from a piece of paper — a contest sheet, a friend's
report, a QSO worked on another radio — means the frequency, the band,
the mode, the date and both times all have to stop following the rig and
the clock at once. That is a single decision, and it was five clicks in
five different places, each of which had to be found first.
The padlock beside Start UTC now holds all five and releases all five,
with the same behaviour as before: locking pre-fills the times so the
fields are not empty, unlocking hands them back to the clock and snaps a
QSO already in progress to now.
The five per-field locks stay underneath. Everything downstream reads
them, and individually they say the right thing — "this value is
decoupled from the rig". Only the control is one.
A backdrop-filter covers the whole window and is recomputed every time
anything above it repaints. Behind these dialogs is an application that
never stops moving — CAT polls four times a second, spots arrive, meters
sweep, maps redraw — so the filter was being recomputed continuously, and
each keystroke's repaint dragged a full-window blur with it. That is why
the lag was felt in Preferences and nowhere else, and why fixing the
cluster macros did not end it.
Worse in one place: the cluster server editor opens FROM Preferences, so
its overlay was the SECOND full-window filter stacked over the first.
That is exactly where the delay was first reported.
The dialogs an operator types in for minutes — Preferences, the cluster
editor, the QSO editor, bulk edit, alert rules, award definitions — now
dim the background harder instead of blurring it. Everything else keeps
the blur: a confirmation you click through in a second costs nothing.
This tab began as the relay and rotator dashboard and stopped there: the
three things an operator touches most — the radio, the CW keyer, the
voice keyer — were the ones missing from the page that claims to show the
station.
The radio card carries the frequency and the mode large, because that is
what gets glanced at, and the split pair only when there IS a split: a
second frequency shown at all times is one more number to read past. When
CAT is down it says which kind of down — switched off, or on and not
answering.
The CW keyer card carries the speed, which is the control an operator
reaches for mid-QSO when a station comes back faster than expected, and
Stop beside it because a message going to the wrong callsign has to end
now. The voice keyer card carries the recorded messages themselves: a
card that only said "idle" would be a light, not a control.
Each polls its own binding and holds its own state, like the supply card
above them, so they drop into the grid and reorder with everything else.
The two keyers appear only when there is something behind them — a port
configured, a message actually recorded — because an operator who works
neither should not be handed two dead cards. That question is asked once
on opening the tab: a keyer is bought and wired, not something that
appears mid-session.
Typing a cluster macro cost two things per character. The state lived on
SettingsModal, so every keystroke in one of those twenty-four boxes
re-rendered the WHOLE preferences dialog — every list, every form, every
panel. And the save wrote through to Go and into the database on each
one: a round trip per character.
The editor is now its own module-scoped component with its own state, so
a keystroke re-renders twelve rows. And writeUiPrefDebounced holds the
database write until the typing stops, while the local cache — which is
what everything reads back — is still written at once. Pending values are
flushed when the page goes away, so typing and immediately closing does
not lose the last word.
The behaviour is unchanged: still saved as you type, still no Save
button, because a text box whose contents only take effect on some other
button is how work gets lost.
Also folds the satellite changelog into one [NEW] entry. Satellites are
new in this version — nobody reading the notes has seen any of it — so a
running account of how it was built, tab then tracking then rotator then
where the settings moved to, is the wrong shape. One entry saying what it
does.
Setting up a telnet cluster is the step operators get stuck on. The
address and the port are two pieces of information nobody has to hand,
and a typo in either looks exactly like a node that is down — there is
nothing to read, only a connection that never comes.
So the editor offers a list, and choosing one fills the fields. Six to
begin with: F4BPO, DXFun, SOTA, POTA, and the two Reverse Beacon feeds —
listed separately and named for what they carry, because they are one
network on two ports where 7000 is CW and RTTY and 7001 is FT8 and FT4,
and no amount of trying will tell you which is which.
The preset then gets out of the way. Everything stays editable, the name
is only filled when the operator has not chosen one of their own, and a
node typed in by hand behaves exactly the same. Reopening a node created
from a preset shows it selected, so the list also answers "which one is
this".
rouggy
