hamradiotools

Nexus 0.24.0

Released 31 July 2026. The current release is 1.24.0. Download it · Every version

Added: satellite operating — full Doppler, and a rotator that behaves

Nexus now works a satellite pass end to end. Your rotator, a standard computer interface, and Nexus — no separate tracking appliance.

Doppler on both legs, continuously. Your downlink is corrected so you hear the bird where it actually is, and your uplink is corrected so the bird hears you where it is listening. Both, always: correcting only the downlink sounds right while your signal slides off the far end of the passband and nobody comes back to you.

Inverting transponders are handled properly. On an RS-44 or AO-7 the passband is mirrored — tune up the band and your uplink goes down, and the sidebands swap. That comes from the satellite database per transponder, not from a checkbox you have to remember, so the common way to land on somebody else's QSO is closed by default.

Tune the downlink and the uplink follows. Chase a station drifting through the passband the way you would on HF — turn the VFO knob and your transmit frequency tracks them, mirrored if the transponder inverts, while Doppler moves the whole band underneath. Tuning outside the passband is treated as what it is, leaving the transponder, so a QSY away from the pass never drags your uplink to a passband edge.

You can see where you are in the passband. A strip shows the transponder as a band with your position marked on both legs at once. On an inverting transponder the two markers sit on opposite sides of centre and move in opposite directions as you tune — which is the clearest way to learn the one rule that catches everybody out, and no other display draws it. Birds with no passband to tune inside, like the FM repeater satellites, say so instead of drawing an axis that does not exist.

Digital modes are handled honestly. For slot-timed modes like FT8 the dial is held still for the length of a transmission and re-corrected between overs, because stepping the frequency underneath a transmission in progress smears it. The physical Doppler on a fast low-orbit pass still applies and no software can remove it, so FT modes remain best suited to the high-orbit birds. SSB, CW and FM steer continuously, which is what you want.

Rotator control worth using. Park and ready positions with a post-pass action, an optional flip above 90° elevation so a high pass goes over the top instead of spinning the mast at zenith, an az/el deadband so the rotator stops hunting for ten minutes straight, and calibration trim for the difference between where the controller thinks it points and where the boom actually points.

Nothing moves until you say so: Doppler is off until enabled, the VFO mapping starts at "off" until you say which VFO carries which leg, flip is off, and the post-pass action leaves the antenna where the pass ended.

Satellite contacts count. A QSO logged during a pass carries SAT_NAME and PROP_MODE, which is exactly what LoTW needs to credit it.

You can see the pass, not just read it. The sky dome is now the main view while a pass is running — a proper polar plot with labelled elevation rings and the track weighted so the high, workable part of the pass reads first. The satellite moves along it as the pass runs.

Beside the satellite, Nexus draws where your antenna was actually pointed, and states the gap between the two as a single angle. Every other display prints satellite position and antenna position as two rows of numbers and leaves you to do the subtraction, which matters because a deadband means the two legitimately differ by a degree or two — knowing that is normal is the difference between trusting the rotator and chasing a fault that is not there. If your rotator is azimuth-only, Nexus draws an azimuth line and says in words that no elevation was sent, rather than drawing an antenna lying on the horizon. Before the pass, when auto-track is deliberately keeping its hands off your rotator, it draws nothing at all and tells you the azimuth the bird will rise at.

The 3-D globe becomes the pass view. When a pass is tracked the globe frames your station and the satellite together, draws the orbit behind and ahead, the footprint, and a line of sight from your antenna to the bird — the range figure, drawn. The flat map stays the "where is everything" view. Neither shows anything before the satellite is above your horizon.

The Doppler readout shows both legs with their live frequency and correction, and both views carry a written description for screen readers.

Fixed: Test CAT now root-causes the Icom "answers nothing" failure

An IC-7610 (or IC-9700/7300/705) that never answers CAT — "isn't answering (got \"\")" in both normal and native CI-V mode — is almost never a broken radio. It is one of two setup traps, and Test CAT now tells you which one you are in:

  • Wrong baud.

    When the configured rate gets silence on a serial Icom, Test CAT now re-probes the same COM port directly at every rate the rig's CI-V menu offers (19200 / 9600 / 4800 / 38400 / 57600 / 115200, read-only — it only ever asks for the dial frequency). If the rig answers at another rate, the result says so and gives you both fixes: change Baud in Settings, or set the rig itself — MENU » SET » Connectors » CI-V » "CI-V USB Baud Rate", with "CI-V USB Port" = "Unlink from [REMOTE]". That last menu is the usual culprit: from the factory the USB CI-V port is linked to the [REMOTE] jack and tops out at 19200, so the 115200 the native scope needs gets you nothing at all, not even garbage.

  • Wrong COM port.

    The IC-7610 and IC-9700 present TWO COM ports and only one speaks CI-V. If no rate answers, Test CAT now walks you through telling them apart (Device Manager: the CP210x port marked "Enhanced" — Icom's driver calls it "Serial Port A (CI-V)"), and Detect / the setup wizard / port Auto-test now label and prefer the CI-V side of the pair instead of showing two identical "Icom IC-7610" rows. On the single-port models (IC-7300/705/905) the silent-at-every-rate verdict instead checks that the chosen COM port is really the rig. Test CAT also now says which backend it exercised — native CI-V or Hamlib rigctld — in green and red results alike, including when the native daemon failed to start and CAT quietly fell back to Hamlib (previously that fallback was invisible, so you could spend an evening debugging the wrong one). And the result you read is the result of the probe you clicked: the button used to report a stale status whenever the rig rebuild took longer than its fixed wait.

Changed: the SSTV screen — a big crisp picture, and panes that take only what they need

A decoding SSTV picture now grows to fit the window in exact whole steps of its native resolution (up to 6×), so a big monitor shows a big, still-crisp image instead of a postage stamp surrounded by dead space — and it steps down cleanly on small windows instead of spilling out of its area. Below the picture, the Transmit box now takes only the height it needs and the Gallery gets everything left over; the old layout gave each exactly half the space whatever they held, which left a fresh install staring at a large empty Gallery band. Both panes carry the same title-bar frame as the other screens, and the ⊞ Panels menu hides them exactly as before. On a window too short for all of it, the screen scrolls: the picture area and the Gallery keep a usable minimum instead of vanishing, and the Send/Stop bar sits below the panes and stays parked at the bottom edge — Stop is reachable at any scroll position. RTTY's decoded-text pane keeps its minimum height on short windows the same way.

Changed: the full caller card is back while operating — QRZ photo, bearing and all

Resolving a call in the Phone or CW log strip once again shows the full recall card: the QRZ profile photo, name, QTH, country, the distance and beam heading from your own grid, your most recent note on the station, and the real list of previous contacts (date, band, mode, reports). The card had been cut to a single line in 0.18.0 because it could push the operating controls off screen; the new pane layout scrolls the log column internally, so the card can take its full height without squeezing anything. A long history scrolls inside the card rather than growing it.

Fixed: satellite predictions were off by a second of orbital motion

Every satellite prediction — Doppler, pass times, look angles, the ground track — was computed from an element-set epoch rounded down to the nearest whole second. A TLE states its epoch to sub-millisecond precision, and the fraction that was being discarded is effectively random from one element set to the next. A low-orbit satellite travels about 7.6 km in the second that was being thrown away.

In practice that put the bird up to several kilometres from where it actually was, and moved the Doppler correction by up to about 75 Hz on 70 cm — enough to sit noticeably off a narrow CW or SSB signal on a fast pass, and over 1.5 kHz on a 10 GHz downlink. Pass rise and set times were off by up to a second for the same reason.

The epoch is now kept at full precision. Nothing about how you operate changes; the numbers are simply right.

This was found by a new test that checks predicted Doppler against carriers actually recorded off the air by volunteer ground stations, rather than against another copy of the same theory. The existing cross-check against an independent implementation had missed it, because its reference data had been generated with the same rounding and the two errors cancelled — which is exactly the blind spot that testing against real recorded signals exists to close. That reference has been regenerated correctly, and agreement on the quantity that reaches the radio improved by 40%.

Fixed: the propagation advisor stops changing its mind on identical data

On a quiet band with one or two spots — exactly when you lean on it — the advisor's "best region", the beam heading, and the confidence word could flip between polls with nothing changed on the air, and when two openings tied, which one got the single "just opened — jump on it now" alert was re-rolled every poll. Ties are now broken deterministically everywhere one reaches the screen, so identical data always gives the same answer. The same fix covers the Journey "most QSOs in a day" date, which station leaves a full APRS list, and the link readout's frequency on equal-strength decodes.

Fixed: the torn-off band map disagreed with the docked one about who you need

On a pop-out band map or a second-monitor Connect map, a station that is an all-time-new entity on the band in front of you could be painted in the dim "needs a confirmation" colour because it also wants a QSL on another band — while the docked window showed it correctly. Both windows now colour from the same shared logic, and a test pins the rule.

Fixed: the 3-D globe's open-band glow was frozen

The globe's band heat updated its "breathing" only once a minute, so an open band could sit dimmer than a closed one and the opening wedges never pulsed — on the 2-D map, that motion is how you tell a live opening from stale spots at a glance. The globe now breathes exactly like the map, from one shared clock, and stays still only when nothing is open (or the tab is hidden).

Fixed: logging could freeze the app when HRD forwarding pointed at a slow hostname

Forwarding a logged QSO to Ham Radio Deluxe resolved the target hostname while holding the app's main lock — with DNS slow or down, every logged contact froze the interface for the timeout and could cost the transmit slot right after logging. The send now runs off-thread, like the N1MM forwarder always has.

Fixed: watch-list stations vanished from a filtered Needed board

Turning on any Needed-board filter hid the callsigns you explicitly asked to be told about — the watch-list tier had no filter bucket, even though it outranks everything. It has its own chip now, and the dupe cue ("already worked this station on this band") also lights for contacts that arrived from a LoTW or QRZ import, which a case-sensitive compare left dark.

Fixed: the shipped AI CW decoder was on a vulnerable model loader

Turning on the supply-chain scan for the desktop build's own dependency set (it was never scanned — only the test tree was) surfaced a known out-of-bounds read in the neural-net model loader the AI CW decoder ships with, plus two denial-of-service advisories in an XML parser. All three are gone: the decoder moved to the patched inference library (also ending a situation where the tested decoder version differed from the shipped one), and the XML parser was replaced by its patched line. CI now scans the shipped dependency set, runs the 38 propagation tests that previously ran nowhere — including the ones that prove connector credentials never travel unencrypted — and keeps the two dependency trees aligned. Building Nexus from source now needs Rust 1.91 (the patched inference library's own minimum).

Fixed: every transmitting mode now refuses to key without a real callsign and grid

The blank-identity guard that has always protected FT8 and FT4 now covers every mode that can transmit. Before this, selecting Q65, FST4, MSK144 or JT65 and pressing Call CQ with no callsign set would key the rig and send a standard message with the identity missing — an unidentified transmission. WSPR and FST4W beaconing now insist on a real callsign and a real grid too, not merely a non-empty box, since those reports are published to wsprnet under whatever you typed. The check is wired into each mode's own capability declaration, so a future mode gets the guard the day it learns to transmit.

Fixed: an over planned just before you changed something could still go out against the old settings

There is a short window while a transmission is being prepared in which the app used to re-check only the mode tier before keying. Stop TX pressed in that window, a QSY to a new frequency on the same band, a sideband or operating-mode change, or the T/R period rolling over could all slip through, and the prepared over went out anyway — against the frequency, mode or slot you had just left. The commit step now re-checks everything the go/no-go decision was based on and quietly drops the over if any of it moved; the next slot plans a fresh one. The PTT hold time is also now measured from when the audio actually starts rather than from the start of the preparation, so the tail is never cut short.

Fixed: one internal crash could silently kill receive — or all of TX/RX — for the whole session

A crash inside one decode used to be contained but leave the decoder's lock unusable, so every later period failed instantly: the waterfall kept painting and the app looked alive while it had gone completely deaf, and switching modes or opening the snapshot could freeze the UI with it. In the worst case the radio loop thread itself stopped — with nothing left to drop PTT if it died mid-over, and the "RADIO ENGINE STOPPED" banner unable to appear in exactly that case. All of the app's shared state now recovers from a crashed thread instead of seizing, the radio loop drops PTT before it ever exits on an error, and a contained decode crash now shows up as a visible notice instead of a line in a log nobody sees.

Fixed: an APRS beacon or auto-ack could transmit on top of your live microphone or a logger's over

With the mic keyed on FM — or WSJT-X keying the rig through Nexus's CAT broker — a queued APRS beacon or an armed auto-ack would key anyway and lay packet audio across the transmission in progress. Every transmit gate now asks one shared "who owns the transmitter" arbiter that knows about all of it: a slot over, the tune carrier, your mic, a broker client's key, the voice keyer, CW, RTTY and SSTV. APRS holds its queue until the air is actually free, and a broker client asking to key while an SSTV image or voice message is playing is now refused too.

Fixed: the tune carrier's auto-release could be configured into a minutes-long unattended carrier

The "Tune after t s" auto-release honoured any number typed into it, and settings.json is hand-editable — a mistyped 120 for 12 meant a two-minute continuous carrier into your finals or a dead load with nothing to catch it. The auto-release is now capped at 60 seconds however the setting got its value; deliberate longer settings up to that minute still work.

Fixed: one malformed APRS packet or log record could kill APRS receive or the whole radio session

A single APRS position report carrying an 8-bit character — a latin-1 degree sign from a real tracker or an APRS-IS feed is enough — crashed the APRS receive thread, and the station list simply stopped updating for the session with no sign anything was wrong. Roughly one in ten corrupted position packets could do it. Separately, an ADIF record whose end-of-contact time carried a multibyte character — from a hand-edited file, another logger's export, or a bad WSJT-X logging datagram — could crash the radio loop itself, ending TX/RX until restart. Both parsers now treat malformed input as the one bad packet or record it is and carry on.

Fixed: Tempo QSOs no longer lose their protocol identity when the app restarts

A TempoFast or TempoDeep contact rides in ADIF as MFSK plus a submode (that is what TQSL and the services accept), but the importer read only the MODE field — so the app's own log re-read every Tempo QSO as plain "MFSK" at launch, and the next save wrote that loss to disk permanently. The importer now reads the identity fields the exporter writes, and a WSJT-X log's FT4/Q65/FST4 rows (which ride the same MFSK-plus-submode shape) import as their real modes too. A Tempo row already collapsed by an earlier build reads as MFSK on disk; the original identity is recoverable from the one-time log.adi.bak made on first load.

Fixed: logging from a named channel mis-filed the QSO's band forever

Working a contact from a suffixed band-plan channel (the 2 m FM simplex data channel, the 6 m second channel, the DX/EU windows) stored the channel id — "2m-fm", "6m-2" — as the QSO's band, pushed that exact string to QRZ and eQSL, earned no DXCC/VUCC/WAS credit, and could never be confirmed. The channel id is now translated to the real band the moment it enters the app, so the log, the uploads, the awards engine and the Needed board all see the same "2m". A QSO that was already logged with a channel id keeps it as stored; the fix protects everything logged from now on.

Fixed: imported contacts claiming midnight could never confirm at LoTW or eQSL

Contacts imported without a time of day used to become 00:00:00 — and were then uploaded asserting that midnight as fact. LoTW and eQSL match on the two operators' times agreeing, so those contacts sat unmatched forever while the "Upload to LoTW" count never went down. The app now remembers that a time is unknown, never writes an invented one, accepts the 4-digit HHMM time form other loggers use, and leaves time-less contacts out of upload batches — the button's tooltip says how many and why. Contacts an earlier build already stamped with an invented 00:00:00 are recognized too: a bare midnight with no end-of-contact time reads as "time unknown" (a genuine 00:00 UTC contact carries one, and still counts as timed).

Fixed: correcting a busted callsign now actually reaches LoTW and the other services

Fixing a mis-copied call used to change it only inside Nexus: the record still counted as "already uploaded", so the correction never went out, the QSO could never confirm — and any confirmation that had matched the WRONG call stayed attached to the corrected contact. A callsign correction now clears the upload record (the QSO re-queues to every service under the right call) and removes confirmations and credit earned under the wrong one. Ordinary edits — band, grid, name — keep everything, as before. Note that LoTW itself still holds the record uploaded under the old call; nothing an upload can send retracts it.

Fixed: the NEW ONE badge cried wolf on entire countries, and the two DXCC counts disagreed

The awards engine identifies a DXCC entity from the callsign; the log screens compared the free-text country name, which QRZ spells its own way — "Germany" can never match the "Fed. Rep. of Germany" already in your log, so every German and Russian contact showed NEW ONE forever, and Statistics and Awards counted entities differently. Every comparison now uses the callsign-resolved entity; the country text is display only. Your log carries both spelling families today, so you should see both numbers agree for the first time.

Fixed: loading a saved config profile could silently remove your RF power ceiling

A profile saved before a given setting existed used to load with that setting reset to its default — and for the per-mode power caps the default is "no cap", so loading a three-week-old profile quietly re-armed full power at FT8's 100% duty cycle. Loading is now a merge: anything the profile doesn't carry keeps its current value, and your callsign, license class, radio roster and sync history never come from a profile at all. Saving a profile now snapshots the last-saved settings rather than a half-edited form.

Fixed: importing a master log from another logger silently threw fields away

The importer kept only the fields Nexus models and discarded the rest — contest exchanges, QSL dates, COUNTY, and the satellite fields LoTW requires for satellite credit were gone from the moment of import, while the manual claimed a full round-trip. Every field now survives import and export verbatim, and the award-relevant ones (numeric DXCC entity, PROP_MODE/SAT_NAME, OPERATOR, STATION_CALLSIGN) are first-class. Upload bookkeeping also stopped re-reading the whole log file before every stamp, which matters on a multi-megabyte log.

Fixed: the Phone and CW screens no longer cut off the log form — the scrollbar is back

At the default window size the Phone screen hid the entire LOG THIS QSO form below the bottom edge with no scrollbar and no way to reach it, while the Band Activity box sat mostly empty. The cause was a set of layout rules that promised "the view never scrolls" while a guaranteed-minimum region and the controls below it added up to more height than any window could hold — and the rule meant to let the page scroll had shipped in a form that could never take effect. The view scrolls again when it must, the empty box is gone (Band Activity takes the height its content needs), and spare room goes to the waterfall instead of to blank space. The same repair covers CW, RTTY and SSTV, and the CW decode text now fills its pane instead of six fixed lines floating in a tall empty panel.

Changed: Phone and CW arrange themselves to the window — wide screens get columns

On a wide monitor the log form now takes its own column beside Band Activity and the voice keyer instead of stretching metre-wide input fields below them; very wide screens get three columns. On a narrow window everything stacks in a single scrolling column. Push-to-talk, the CW send bar and Stop TX live in a fixed dock at the bottom of the screen that can never scroll away or shrink, and switching column layouts can never interrupt a log entry you are typing or a voice-keyer message in flight. Panes you remove in the ⊞ Panels menu behave exactly as before.

Fixed: window sizes, splitter drags and UI-scale settings are checked before they are applied

A pane width or splitter position saved on one monitor was replayed unchecked on another: a rail dragged wide on an ultrawide could reopen on a laptop with the centre workspace squeezed to nothing, a UI scale pinned for a big screen carried into small pop-out windows that have no control to undo it, and a band-map window left on a since-removed second monitor came back entirely off-screen. Every stored size, position and scale is now validated against the window and monitors actually present before it is used. The first paint also matches the final layout — no more one-frame flash of wrong sizes on launch.

Fixed: screens respond to the size they really are, and dialogs respect your UI scale

The rules that adapt layouts to narrow windows measured the raw window width, which the app's own UI scaling makes wrong in both directions: the whole app flipped to a phone-style stack at the minimum window size where there was actually plenty of room, and an operator who raises the UI scale to read the screen never got the narrow layouts at all. All of it now keys on the effective size. Satellites, DXpeditions and APRS gain the narrow layouts they always declared but that were keyed to names that never matched anything. The setup wizard, dialogs and tooltips — which ignored the UI scale entirely and could render at half size for exactly the operator who had raised it — now follow it. The Logbook and radio-programming screens use more of an ultrawide monitor instead of a fixed centre column.

Fixed: live text panes no longer yank you back down while you read

The CW transcript re-pinned itself to the newest text every half-tick, so scrolling back to re-read a callsign mid-copy was impossible; the SENT echo, RTTY stream, Field Day log and Tempo conversation did the same. All now follow only while you are at the bottom, the way Band Activity already did, and switching Tempo conversations opens at the newest message instead of wherever the previous conversation was scrolled. Logging a contact no longer makes the screen jump.

Fixed: opening Field Day bonuses no longer hides the Sections board

The bonus checklist grew without limit and pushed the Sections board and the newest log rows out of a screen that could not scroll. It now scrolls within its own bounded area.

Fixed: the call resolution details are reachable again

When a callsign resolves, the prior-contact history sits inside the log form — with the layout repairs above, it is on screen or a short scroll away instead of clipped below the window edge at every size, which is how it has behaved since the compact recall card arrived.

Added: Unassisted mode — one switch for a no-assistance contest entry

Settings ▸ Contesting has a new Unassisted entry switch. It turns off the AI CW decoder, DX cluster / RBN spots, and the PSK Reporter needs feed together, in one action, and takes effect the moment you press it rather than at the next restart. Every change is written to a dated record beside your settings, so if you are ever asked what was running during an event you have an answer with timestamps on it.

This matters because the shipped defaults put you in an assisted category without saying so. The AI CW decoder is on by default, and CQ WW rule VIII.2 counts "a CW decoder, DX cluster, DX spotting Web sites … local or remote call sign and frequency decoding technology (e.g., CW Skimmer or Reverse Beacon Network)" as QSO-finding assistance, which places an entry in Single Operator Assisted. ARRL calls it spotting assistance and names "PSKReporter, Telnet, DX spotting websites or bulletin board systems, automated multi-channel decoders"; Single Operator may not use it, Single Operator Unlimited may. A footer in the CW cockpit now states which of those sources are running, with the rule citations one click away.

Two details worth knowing. Your own settings are never rewritten: the switch overrides them while it is on, so ending unassisted mode brings your decoder and feeds back exactly as you had them. And your own radio's decodes keep feeding the Needed board, as do your outbound PSK Reporter uploads, because ARRL says plainly that "Generating spotting information for use by other stations is not considered to be spotting assistance."

Rules differ by contest and change between years. The note reports what CQ WW and ARRL currently publish and tells you to check the contest you are entering. It does not rule on your category.

Fixed: beacons and W1AW bulletins were offered as new countries

14.100 MHz carries the international beacon network: eighteen beacons in eighteen different countries, each transmitting every three minutes, all day and all night. Anything watching that frequency therefore offered 4U1UN as an all-time-new country, then the Canadian beacon, then the Californian one, and around again forever. W1AW's code practice and bulletin frequencies did the same thing with a station that is broadcasting one-way and will never come back to your call.

Beacons and bulletins are no longer scored as needs, no longer fire a Pounce alert, and no longer take a new-country colour. They are still shown, because hearing a beacon is real evidence that the band is open, and that is worth knowing. They now carry a B or W badge saying what they are.

The suppression is by frequency, not by callsign, and that is deliberate: 4U1UN is the United Nations headquarters station as well as a beacon site, so blanking the callsign would have hidden a genuine new country. Heard on 14.100 it is the beacon; heard anywhere else it is the station, and it still counts. W1AW is suppressed only on its own published bulletin frequencies, so an ordinary contact on those frequencies, or W1AW itself operating elsewhere, still scores normally.

Fixed: the Call Roster and Band Activity filters reset on every restart

"Needed only" and "Hide worked" on the Operate Call Roster, and the Band Activity filter chip (All / CQ / To me / On RX / B4 / New), now come back the way you left them. They were held in screen state only, so every launch put them back to showing everything and you re-ticked them at the start of each session.

Each pane remembers its own set, so a torn-off Operate window can sit on Needed-only while the docked one still shows the whole band. A window that has never been given its own filters opens on the ones you are already using rather than on defaults. If you have never touched these controls, nothing changes: both checkboxes start off and the chip starts on All, exactly as before.

A stored value that is damaged, or left over from a build whose filters were named differently, is ignored rather than applied. The roster can never come up quietly hiding rows with no ticked checkbox to explain why.

Fixed: "sort by need" on the Call Roster had no discernible order

Sorting the roster by Need now ranks by how much the station is worth working: a call you asked for by name, then a new entity, new zone, new state, new grid, new band, new mode, then one you have worked but not confirmed. That is the same ranking the Needed board uses, so the two agree row for row, and a rare grid or a live park activation keeps the extra pull it has on the board.

Two things were wrong. A station heard on more than one band was ranked by its WEAKEST need instead of its best, so a new country on 20 metres that also needed a confirmation on 40 sorted as the confirmation, well down the list. And among stations of equal need the roster listed the quietest first, which is backwards: of two equally-needed stations the louder one is the better bet. Both are fixed, and the row's colour now names the same need the sort ranked it by.

That weakest-need mix-up was not confined to the roster. The same per-station need was feeding the map and the band strip, so a station worth chasing could be painted in the colour of the least interesting thing about it anywhere it appeared. Every surface now takes a station's strongest need, from one shared piece of logic rather than three.

Fixed: the Needed board no longer claims a "new mode" you already worked

An operator with roughly 11,000 FT8 contacts was shown a new mode needed for Asiatic Russia on 30m, on a band and entity they have six confirmed FT8 contacts with. The need behind it was real — they have never worked that entity on CW — but nothing on the screen said so, so the board looked like it could not read the log.

Two separate things were wrong, and both are fixed.

  • The row said the wrong thing.

    A mode need is judged per DXCC entity and mode class across all bands, which is what the per-mode DXCC awards count. The headline appended whichever band the station happened to be spotted on, turning "you have never worked Asiatic Russia on CW" into "New mode — CW Asiatic Russia 30m". Mode rows now read "New mode — CW Asiatic Russia (any band)" and state exactly the claim being made. Band slots remain the separate new band row.

  • The roster showed chips you could not act on.

    Need chips on the Call Roster and Stations panel were keyed by callsign alone and pooled every alert for that call, so a CW mode need painted an unlabelled MODE chip onto a 30m FT8 roster. Chips, row colour, and the Needed filter are now scoped to the band and mode class actually in front of you, matching what Band Activity has always done. A genuinely cross-band need still shows on the Needed board, where the band is named. Three further mis-classifications surfaced in the same audit, each of which could invent or hide a need:

  • Phone contacts logged as PH

    — the token the N3FJP family exports, and present in real imported logs — were counted as digital. They credited a digital mode slot they never earned and left the phone slot reading unworked. PH and the digital-voice modes (DIGITALVOICE, DSTAR, FUSION, M17, FREEDV) are now classed as phone everywhere, including in LoTW confirmation matching, where the mismatch could leave a confirmed contact showing as a confirmation opportunity forever. This is read-side: existing logs reclassify with no re-import.

  • Band labels are compared case-insensitively.

    A log that carries both 30m and 30M (both spellings occur in a real logbook) credited the same slot in the awards engine but not in the decode feed's chip gate.

  • Nexus's own tier names are recognised on POTA/SOTA spots.

    TempoFast/TempoDeep/FT1/DX1 fell through to guessing the mode from the frequency, which on a CW-only band segment could invent a CW mode need. This works together with the strongest-need ranking above rather than against it: the band and mode gate decides which of a station's needs count on the surface you are looking at, and the ranking then picks the most valuable of those. So a row is ranked by the best reason to work that station that you can actually act on right now, and its colour, its chip, its screen-reader label and its place in "sort by need" all name that same need.

APRS says which radio it is listening to

Every version