Skip to content

Project status

Regis edited this page Nov 19, 2026 · 139 revisions

Project status

Last updated: 15 November 2026, version 1.2.2. 1.2.0 (lot 2) was published on 15 November 2026 at the owner's word. On the Pi before it went out: the squawk banner's first form on the real wall, MLAT moved to FlightAware and back. The banner's second form was seen on the real wall on 15 November at 01:48 (1.2.1, its test). Not on the Pi yet: a real emergency squawk, FlightAware's MLAT column after a day (MLAT is with FlightAware since 14 November, 23:45), AirNav Radar's MLAT switch, the receiver alert on a real failure. About 1.1: 1.1.0 was published the day it was written, at the owner's word (je sors la 1.1.0 maintenant), before any of it ran on the reference Pi. Since then, on the Pi (field log, 14 November 2026, 21:16 and 21:35): both services start on 1.1 and read 1.0's state; a crossing message was sent and then completed with its closest distance by a real edit in Telegram. Bluesky posts by 1.1.0 went out from 21:33 (field log, 22:03). Not seen yet on the Pi: the heading read in a post, the airline lookup, the list of aircraft never to post, a restart in the middle of a followed crossing. About 1.0.0: The owner closed the release candidate on 14 November 2026 (je ferme ma RC à 14 jours, assumé): the fresh install on a blank card was redone during it; the month hands-off stopped at fourteen days of thirty, and a fixed IP address has still never been applied on a real Pi. Those two are the limits 1.0.0 is published with.

This page separates what has been proven on a real Raspberry Pi from what has only been written and unit-tested. The difference matters: every test in this project runs against stand-in programs, so a green suite proves the logic and says nothing about the hardware.

Proven on hardware

One Raspberry Pi 3B, outdoors, on Wi-Fi, with an R820T dongle.

Part What is known
Install and the menus Run end to end on a Pi 3B
Telegram messages (0.99.6) Set up on the reference Pi on 30 October 2026, 22:39 to 22:40: the bot answered, the chat was picked from those that wrote to it (/start sent again: one from 26 October was too old for Telegram to list), the connection message arrived. Crossings sent that evening in French, every crossing inside 10 km: KL31L at 22:53, then S66373, LG68A and S66381 (23:05 to 23:14), each with the model, registration and flag, the route, the altitude, speed and climb, the time and the place, and the ADSB Exchange link; the console's Telegram card as documented. Not yet seen: a real emergency squawk, rare by nature (sudo zenithboard telegram test squawk send shows its shape)
AirNav Radar (0.99.6) Installed on the reference Pi on 30 October 2026 (Debian 13: AirNav's repository accepted with its pinned key): zenithboard status shows rbfeeder running, MLAT off with one MLAT client in all (ADSB Exchange's), and its station's serial with its page on radarbox.com. 9 MB of memory (15 at its peak)
graphs1090 (0.32.0) Installed from the console on the reference Pi 3B on 21 October 2026 (about a minute) and running in ZenithBoard's look: 212 MB for every service with it, 518 MB available (21:02)
The web console (0.39.0) In use on the reference Pi since 21 October 2026: the .path unit and the root helper run its changes (graphs1090 installed from the page, statistics, the MLAT guard), an iPhone paired by QR code, a PC signed in with the password, a backup made and downloaded (12:57)
readsb + the wall Running continuously; the split-flap look is the one in daily use
Feeding ADSB Exchange Feed received and confirmed
ADSB Exchange's stats (0.52.0) adsbx-stats on run; Stats Ok and the Anywhere map on their page; 6 MB
Gain tuning zenithboard gain status reads real statistics; settled at 48.0 dB
The 24-hour gain log A full day recorded, 96 measurements, nothing lost and nothing overrun. See Gain tuning
Aircraft photos Real photos from Planespotters on the wall
The Wi-Fi menu The Pi is on Wi-Fi, joined through the menu
Feeding FlightAware and Flightradar24 Both feeding; FlightAware's coverage chart confirms it
Domain name (DuckDNS) Set up and updating
Route lookups Working against adsbdb. The wrong answers it produced are what led to the plausibility filter and the hexdb second opinion in 0.9.2
Reliability menu All four options turned on and left on
Hardware watchdog Installed and surviving reboots
Network watchdog Did real work on a real outage and diagnosed it: three network restarts three hours apart, no reboot loop, and its log named the cause
Persistent logs Working, after one sudo journalctl --flush the first time
The systemd sandbox The wall server runs inside it; systemd-analyze security gives 1.3 (OK)
Signed updates Checked against GitHub's current key, silent once confirmed
zenithboard health Reports power, temperature, memory, card and the previous boot correctly; since 0.34.0 the card's size too (3.5 GB used of 61.0 GB, the card: 62.5 GB)
zenithboard memory (0.34.0) Service by service on the reference Pi: 155 to 212 MB for all the services with four feeds, social sharing and the wall (180 MB on 21 October, 619 MB available); Plane Finder 19 to 43 MB, growing while it runs. See Hardware
Feeding Plane Finder Since 20 October 2026; its page on port 30053 shows the aircraft decoded (KLM31L, PH-HSE, Amsterdam → Geneva)
The statistics page Running since 0.16.0; the lists of 0.31.0 and 0.37.0's look, toggle, cargo, business jets and flags seen on the real page (45 cargo aircraft, DHL 11, ASL Airlines Belgium 10…)
social summary weekly test Run on the reference Pi (0.33.0, 0.35.0): the right totals (6 437 aircraft, 354.6 million messages, busiest Friday), the lists left out until they cover the week (0.34.1)
Social sharing on Bluesky Posting on its own since 17 October 2026 under ALL, both looks, the four maps, the colours in turn; every crossing of a 10 km radius once since 0.20.0. Memory measured by the service itself
The web console An update run from the page, and a computer signed in by the phone (its QR code, then yes on the phone): done several times on 25 October 2026, on two Pis
Room on a 1 GB board With four feeders, the wall, statistics and social sharing: 575 to 654 MB of 1 GB still available, 34–35 °C, no throttling (20 October 2026; 656 MB with three feeders on 17 October). See Hardware

Written and unit-tested, not yet run for real

Treat these as likely-working rather than known-working, and please report what you find.

Part What is missing
978 MHz UAT (United States) No US hardware has run it. Experimental, outside 1.0 (the owner, 25 October 2026)
Fixed IP address The nmcli calls are tested; a real address change was not run. Keep a keyboard and screen at hand
The farthest aircraft named on /stats Seen on the real page (20/10 08:20 · easyJet EZY17QZ · Manchester → Verona · 335 km, near Cergy-Pontoise, the owner's export of 24 October 2026); the thread's cost on a Pi 3 is not measured on its own
A fresh install from the README install.sh has been run end to end in a container against a scripted whiptail and stand-ins (0.21.1, tests/test_install.sh): flags, wizard, every menu it can reach. Not by that test: the deployed wall answering (curl is a stand-in there; the server has its own tests) and the uninstall (a source check only; corrected 0.54.1, it was credited with both). A blank card on a real Pi has not been done
graphs1090 Running on the reference Pi (its graphs in the owner's exports, 24 October 2026); collectd's memory over weeks is not measured
The summaries posted by the service (0.33.0–0.38.0) weekly test ran on the Pi; the first scheduled one went out on 21 October 2026 (daily, 09:20), post 1 only, its lists left out by the rule fixed in 0.40.4. The daily and thirty-day summaries and the queue shared by all of them (0.38.0) are tested with a simulated clock only. The first weekly post is due on Sunday 25 October 19:00 if switched on; the airline lists cover a whole week from 1 November
Posts 2 to 4 and the page's lists with real data over weeks Cargo by callsign prefix (40 operators checked against published sources), military by database mark and address block, countries by ICAO address block: checked on made-up and tar1090-db data; a month of the real sky is still to come
The page's last 24 hours (0.37.0) Their lists need the last-seen column written since 0.37.0: complete from a day after the update

Known rough edges

  • Feeder download addresses are pinned in lib/versions.sh and were copied from each provider's page on 6 October 2026. Providers move files. A 404 during install usually means one line in that file needs updating. When a provider changes something says what happens and what to do.
  • FlightAware and Flightradar24 keep your antenna position on their own websites. zenithboard location cannot change that for you; it tells you where to go.
  • The airline name list covers about ninety airlines. Since 0.17.0 the rest are named from the route lookup once one of their flights has been looked up; with routes off they show the callsign.
  • Routes are looked up by callsign, not by flight. A callsign can be reused for the other direction. The filter drops the impossible answers; it cannot invent the right one.

Audit of 22-23 October 2026: what is left

The audit (build, tests, security, novice journeys, translations) ran on 0.53.3. Everything of high and medium severity is shipped (0.53.6 to 0.54.7, see the CHANGELOG), and so is most of the low severity. What follows is the rest, in batches that go together, each batch one pull request with its tests, docs and (for the web console) its translations into every language in the same commit.

Before starting a batch: read the CHANGELOG from 0.54.7 up, git log, and the open pull requests; then grep each item below in the code. Line numbers are those of 0.53.3 and have moved; an item already fixed is crossed out of this list in the same pull request that fixes the rest. The owner's standing instruction (23 October 2026): do the low-severity findings batch by batch, novice journeys first, merge each through the squash merge, report once per batch.

Batch 1: the console's words: shipped in 0.54.8; its optional findings, and two more of the console's (a lost phone's sign-outs, the gain log read in the console), in 0.54.15 (lot A, the owner's choice of 25 October 2026).

Batch 2: the Bluesky menu and wizard: shipped in 0.54.9; its optional findings in 0.54.16 (lot B).

Batch 3: Esc, numbers and the last menu details: shipped in 0.54.10; its optional finding (a Wi-Fi password typed again) in 0.54.16 (lot B).

Batch 4: the wall: shipped in 0.54.11.

Batch 5: robustness and the tools: shipped in 0.54.12; its optional findings, with the Bluesky service waiting for the clock, in 0.54.17 (lot C).

Batch 6: the tests: shipped in 0.54.13 and 0.54.14. Nothing left.

Batch 7: the docs: shipped with 0.54.13's documentation.

Decided against (the reviewers' majority), unless something new shows up: moving the security updates line in the console; a clock-generation counter for sessions; the QR on a tiny link; the framing headers of the wall without a CSP check in Chromium; a tests/lib.sh shared harness (each suite stands alone on purpose); a full menu-walk suite for the network screens; time.sleep(0.4) of console.py; flags.js having no module.exports; the RTL literals spread over four files.

Field log

Short notes from the Pi that is actually outside. Kept because the second time something happens, the first time is useful.

When What happened
20 November 2026, 00:46 1.2.2 on the reference Pi (update quick between 00:24 and 00:40): the Bluesky service says running ZenithBoard 1.2.2 at 00:29:37, and the owner restarted both posting services again at 00:46. No post yet since: the night over a 5 km radius. To read in two or three days: journalctl -u zenithboard-social --since "2026-11-20 00:30" | grep "no route", where still on its way should be nearly gone. Two more posts before the update, MTL363C and MYW5060, were callsigns neither database knows
19 November 2026, 21:46 Why posts go out without a route: the owner's social log of three days (17 to 19 November, 30 posts without one), after his screenshots of the 19th. 13 callsigns unknown to adsbdb and to hexdb (business jets and charters: NJE919Y, SVW15RZ, OYO13, TJD801, FYL61AN, AWH51A…), 5 not airline callsigns (N302KC, 3AMSR, 39b985…), 6 with two answers for other flights, rightly refused (each checked on the 19th: UAL3907 Hong Kong → Guam and Brussels → Newark over the Moselle heading south; WMT6X Mykonos → Rome twice; AIA252F Paris → Paris and Newcastle → Amsterdam; BCS2219 Stansted → Leipzig and Newcastle → Leipzig; BAW709 Zurich → London and Larnaca → London while flying south-east), and 6 that went out with the answer still on its way, one lookup in the queue: ABR4VA, TJD801, BAW709, SWR5MP, RYR3WB, TAY8SL. Those six are a bug, fixed in 1.2.2: the two-minute wait for a lookup was timed per airframe from the first time it was ever seen waiting, and never reset, so an aircraft back hours or days later did not wait at all (ABR4VA on the 18th then the 19th, TJD801 at 10:11 then 17:27). SWR5MP would have had London → Zurich: hexdb has it, and the rule accepts it at Trémery
15 November 2026, 01:48 The owner sees the banner's second form on the real wall, on 1.2.1 (rp3:8080/?squawk_test=7700, split-flap, red, an empty sky so the invented aircraft): 7700 EMERGENCY / FLT TEST123 / REG F-TEST / TEST AIRLINE / AIRBUS A320 / ALT 10670 M / PARIS / →ROME / THIS IS A TEST, the dotted silhouette and the sky scene in the corner as for any aircraft. zenithboard memory, twenty minutes after the paste before: 544 MB available of 905; the Telegram service still 35 MB, the Bluesky service 36 MB, the wall 41 MB (32 MB just after its restart), piaware 35 MB with FlightAware's MLAT client; load 0.84. zenithboard health: nothing alarming, 30 °C, the same boot since 14 November 17:45. Still to read: FlightAware's MLAT column after a day (MLAT with FlightAware since 14 November, 23:45); the Telegram service's memory over a day (20 MB on 14 November at 17:08, 35 MB since 1.2.0); a real emergency squawk; AirNav Radar's MLAT switch; the receiver alert on a real failure
15 November 2026, 01:31 The owner's paste after updating to 1.2.0 (telegram health check, status, memory, and the wall at rp3:8080/?squawk_test=7700). The receiver alert is ON and its check, on the Pi's own thirty days (2 877 samples), says what the session's replay of his file said: 8 November 03:30 Alerte récepteur : presque rien entendu depuis 30 minutes (374 messages…), 05:15 Le récepteur entend de nouveau normalement, Now: nothing to report. MLAT with FlightAware since 23:45: MLAT is done by: FlightAware, one client, piaware 35 MB in three processes; its MLAT column is still to read after a day. Memory: 554 MB available of 905, all services 368 MB; the Telegram service 35 MB (20 MB on 14 November at 17:08), the Bluesky service 36 MB, the wall 32 MB; load 2.20 just after the update. The banner test showed the plain form (EMERGENCY SQUAWK / 7700 / GENERAL / EMERGENCY / TEST123): at 01:30 no aircraft was inside the wall's radius to stand in, so what he had come to see, which aircraft and where, was not shown. 1.2.1: the test invents an aircraft when the sky is empty
15 November 2026, 00:47 The owner has 1.2.0 published (enchaîne 1.2.0), with lot 2 complete on main: the squawk banner, the MLAT network, the receiver alert. He left MLAT with FlightAware at 23:45 the evening before, to read its page after a day. To read after the update: rp3:8080/?squawk_test=7700 (the banner's second form), zenithboard telegram health check, zenithboard status, zenithboard memory
15 November 2026, 00:23 The owner's stats.csv, thirty days (16 October to 14 November 2026, 2 877 quarter-hour samples), read for the receiver alert's thresholds. Messages: 65 million a day in mid-October, 41 this week, 36 % fewer. Messages per aircraft in view by day (08:00 to 20:00): 10 400 to 12 100 a quarter hour, unchanged. Aircraft with a position in view: 63 on average, then 37 to 44; distinct aircraft a day: 2 745, then about 2 200. Median range: 308 km, then 278; never under 94 % of its fourteen days before. An hour against the same hour a week earlier: 0.55 at the lowest by day, 0.11 by night. One real event: 8 November, 02:45 to 04:00, 12, 374, 878 and 1 709 messages a quarter hour (4 320 at the least on every other night); its cause is not known. The owner's reading of the fall: fewer flights, with the price of kerosene; not checked here. He agreed the three rules (ok pour tes règles)
14 November 2026, 23:49 The owner sees the squawk banner on the real wall (rp3:8080/?squawk_test=7700, split-flap, white; the gear's Emergency squawks: Banner, 30 s): EMERGENCY SQUAWK / 7700 / GENERAL / EMERGENCY / TEST123. His word: test is ok, but it must tell me what plane and where, with picture as usual. The banner becomes the aircraft's own screen with the code on its first row and the place near it on its last (the same evening; not yet seen by him). He also looked for stats.csv under /var/lib: it is in his home folder, /home/pi/stats.csv
14 November 2026, 23:29 The owner moves MLAT to FlightAware and back, on main after 1.1.2 (sudo zenithboard mlat flightaware, zenithboard status, sudo zenithboard mlat adsbx, with piaware's log from FlightAware's site). At 23:26 piaware restarts, says multilateration data requested, starts fa-mlat-client 0.2.13 (UDP to FlightAware, Beast input from 127.0.0.1:30005, results to ::1:30104); status says MLAT is done by: FlightAware, MLAT clients running: 1 (OK: max 1), AirNav MLAT off (FlightAware keeps the only one). At 23:28 mlat adsbx: piaware says multilateration data no longer required, disabling mlat client, and one client runs again. One wrong line: right after the change the command said MLAT clients running: 0, counted three seconds after piaware's restart, before its client was up; it now waits up to twenty seconds. Not read: FlightAware's MLAT column after a day (it held MLAT two minutes), memory with its client, AirNav's switch. The owner also made ~/stats.csv on the Pi, for the health alert's threshold; it has not reached the session yet
14 November 2026, 22:09 The owner's journal of the wall and the Bluesky service from 18:55 to 21:35, for the silence noted at 22:03. No sign of a fault: not one line of trouble (no the wall did not answer, no could not post), only the starts: the wall at 19:26, 20:15, 20:43 and 21:14 (the evening's updates), the service some forty-five seconds after each. The process that had posted at 18:58 posted nothing more in its last twenty-nine minutes, before the first update, and 1.1.0 nothing in its first nineteen: the silence began and ended outside the restarts. Read as a quiet sky over a 5 km radius; the journal cannot prove no aircraft crossed, /stats shows the hours
14 November 2026, 22:03 The owner's zenithboard logs social, filtered on the posts. 1.1.0 posts: PJZ600 at 21:33 (with the reason it has no route: adsbdb gave a route this aircraft cannot be on, hexdb did not know the callsign), RYR4VM 21:41, KLM77E 21:50, KLC79M 21:54, four in twenty-two minutes, each once: the state written by the new helper after a post holds. The silence of 18:58 to 21:33 stays unexplained: no post under the three processes started at 19:27, 20:16 and 20:44 (all saying 1.0.0), none of which ran a full hour, so none wrote its hourly line; the posting code of 1.1.0 is the same as theirs for an aircraft crossing, and it posts. A quiet sky, the wall not answering during those restarts, or something else: the journal of that window would say
14 November 2026, 21:35 The owner's screenshot of his chat: the closest distance works against Telegram itself. PJZ600 (Premium Jet, a Global 5000 HB-JAM, 12 500 m, 921 km/h), message at 21:33 from 1.1.0 with his own crossing template, completed by the bot's edit with au plus près : 1,2 km as its last line; so the message's id was kept, the aircraft followed and editMessageText accepted. The same picture shows a line holding only →: his template writes the route as {from} → {to} without brackets, and this jet has no route. 1.1.2 drops a line left with only the template's own signs. His template also puts the heading in his own place (sud-est), so the default text's cap sud-est is not what this message shows
14 November 2026, 21:16 The owner's first paste after updating to 1.1.0 (health, social status, telegram status, both journals). Both services started on 1.1.0 at 21:14 and 21:15 and read the state 1.0.0 had written, through the new helper: history since 17 October, 58 posts in the last day, 139 types, range record 485 km; Telegram 45 crossings today, 393 and 1 squawk in seven days. The new lines are in the statuses (Aircraft never to post: none, Closest distance … off). Health: nothing alarming, 504 MB free of 904 MB, 33 °C, the Bluesky service 36 MB. Not seen yet: a post or a message made by 1.1.0 (so neither the heading nor a state written by the helper after a post), and a stop of the 1.1.0 services (the write on SIGTERM). Not explained by the paste: no post between 18:58 and 21:16, across restarts at 19:27, 20:16 and 20:44 whose cause it does not say, with the gap at none and every crossing posted; a quiet sky or a fault cannot be told apart from it, the hourly line (last hour: N looks, M with a candidate) needing an hour without a restart. Under 1.0.0, at 18:50 and 19:01, Telegram answered 502 Bad Gateway and a read timed out: the crossing messages about RYR5LV and EJU5435 were dropped after ten minutes, as designed
14 November 2026, 21:13 The owner's zenithboard stats gaps as text: the forty prefixes without an airline name, from JFA on 31 aircraft to eight on 5. The twenty-four not yet read were asked of adsbdb: seventeen right, OCN wrong (O Air; the aircraft are Discover Airlines'), OOM and ECA named after other companies (Zoom Airlines, Eurocypria Airlines), KMM, LUA, HBJ, FJD unknown to it. 1.1.1: OCN in the built-in list, OOM and ECA never asked
14 November 2026, 21:00 The owner has 1.1.0 published (je sors la 1.1.0 maintenant, to test it by sudo zenithboard update), choosing that over trying main on the Pi first. Nothing of lots 0 and 1 has run on the reference Pi at that time: the state helper of the two posting services, the airline lookup, the heading, the aircraft never to post, the closest distance edited into the Telegram message. To read after the update: zenithboard health, sudo zenithboard social status, sudo zenithboard telegram status, zenithboard logs social, zenithboard logs telegram
14 November 2026, evening adsbdb's airline endpoint read for sixteen of the owner's forty prefixes without an airline name: fourteen right, two wrong (JAF Jetairfly, PTN a Brazilian airline; the aircraft heard are TUI fly Belgium's and Platoon Aviation's), both put in the built-in list. Telegram's documentation read for editMessageText: no time limit stated for a bot's own messages (the Roadmap said 48 hours)
14 November 2026, 17:50 The owner puts the choice of the MLAT network in lot 2 (passe mlat en lot 2), and has zenithboard.app say 1.0 is out (its notice in the 22 languages, the same day)
14 November 2026, 17:42 The owner: MLAT must not stay on ADSB Exchange for ever; the network that does it is to be chosen, with what depends on it in the menu, the console and the command. Not in a lot, so written under Waiting for 1.1 with its dependencies (the guard, the installs, the update, status, backup, the 22 languages) and proposed first in lot 4, before adsb.lol and OpenSky
14 November 2026, 17:25 The owner closes the release candidate and publishes 1.0.0: the fresh install on a blank card a bien été faite during the candidate; the hands-off month (started again on 30 October, 22:27, with 0.99.6) is stopped at fourteen days, assumé. The freeze ends: 1.1 follows, lot by lot (Roadmap). He agreed the same day to what is put off after the lots and to what is not planned. Still open and said as such: the fixed IP address never applied on a real Pi
14 November 2026, 17:17 The owner asked for the work after the release candidate in lots, justified, and put the airline names in lot 1 (je suis d'accord, lot 1). Written in the Roadmap, The lots after 1.0: the base first (proposed, his word awaited), then the visible gains, the wall and the receiver's health, the watch list once measured, the networks last
14 November 2026, 17:08 The owner's zenithboard stats gaps (28 days of samples; he takes it as the reading planned for the 17th) and zenithboard memory. No type code drawn with the generic silhouette. Forty callsign prefixes without an airline name, each on 5 aircraft or more: JFA 31, VJT 19, WMT 17, KLC 14, GAC 12… adsbdb's airline endpoint asked the same day: it answers by designator (JFA Jetfly Aviation, VJT Vistajet, WMT Wizz Air Malta, TUI TUIfly), its country wrong for VJT. Memory: 496 MB of 905 MB available, 409 MB in use, swap 115 MB; Bluesky 93 MB (peak 129), the wall 79 MB, Tailscale 38 MB, Telegram 20 MB; load 0.41. Decided in the Roadmap: silhouettes closed, the airline lookup to build after 1.0
11 November 2026, 22:24 The owner ran sudo zenithboard gain status then gain check (48.0 dB, 414 832 messages in 15 minutes, 0.1 % very strong at night, farthest 268 km): the same screen twice, what for? Every command looked over for doubles: eleven old spellings or aliases (gain check, color, show, style, threads, presets, preset rm/forget, social weekly, social tags/color/sentences/templates, telegram templates). 0.99.25 keeps one name each; the old ones still work with a line saying so, until 1.1. Close but not doubles, kept: status (the settings and services) and health (the checks); stats show (the figures) and stats status (the service); update quick (a shortcut of update only zenithboard flightinfo)
10 November 2026, 18:11 The owner's /stats on his phone: 2 306 distinct aircraft in 24 hours, 6 661 over 7 days, 10 719 over 30 days, busiest days 2 525 (06/11) and 2 963 (01/11): 2 000 a day times 7 is never 6 600, the figures are wrong. Checked on his own day files (the backup of 21 October): 16–20 October, 13 097 aircraft added day by day for 6 186 different aircraft; 445 seen all five days, 611 four, 899 three, 1 500 two. The figures were right, the label read as the days added up; his FlightAware gives the other count (76 648 over 30 days, about 2 550 a day added up). 0.99.24 shows both
9 November 2026, 07:29 The owner, with FlightAware's Site ranking over 30 days for regis57: rank 4786, total 76.648, 76,648 ADS-B Mode-S aircraft and 6,608,812 positions, MLAT 0 aircraft, 0 positions (as set since 0.99.6: ADS-B Exchange keeps the only MLAT client), Other 2,695 aircraft and 2,078,227 positions. Asked: each network's ranking on /stats. Studied the same day, written in Waiting for 1.1: only Plane Finder and possibly AirNav Radar let a program read theirs
2 November 2026, 19:26 The owner: neither the wall nor SSH answer at the Pi's address, but http://rp3:8080 works (his phone's screenshot: the split-flap wall, EWG7LW Palma → Cologne). The router had given the Pi another address; the name followed it. No ZenithBoard change touches the address (checked: nothing in 0.99.x changes the network settings). Written in Reliability and diagnostics; a DHCP reservation in the router keeps it fixed
1 November 2026, 18:31 The owner, with a screenshot of the ADS-B Feeder Image's homepage (nineteen networks): adsb.lol and OpenSky for 1.1, each its own switch, MLAT left to the owner; the others looked at and left out. Each read on its own pages that evening: adsb.lol's script installs adsblol-feed and adsblol-mlat, no account, ODbL data; OpenSky's opensky-feeder package needs an account and a receiver serial and has no MLAT client; RadarPlane's address redirected to an unrelated site; HPRadar did not answer. In the Roadmap, Waiting for 1.1 and Not planned. Nothing built: the freeze
31 October 2026, after updating to 0.99.21 The owner's paste. ls -ld /var/lib/zenithboard: drwxrws--t root zenithboard, mode 3771 as 0.99.11 sets it; ls /var/lib/zenithboard/* as pi says No such file or directory because others may cross the folder but not list it (o=--x), which is meant: sudo ls -l /var/lib/zenithboard lists it. journalctl --list-boots on Bookworm: the header IDX BOOT ID FIRST ENTRY LAST ENTRY, then one row, boot 0 since Wed 2026-10-21 08:33:18 CEST: the format 0.99.13 reads (one boot, so no earlier boot, never ABRUPTLY). sudo zenithboard telegram status: ON, crossings on the 10 km sharing radius, squawks 7500/7600/7700 anywhere, French, aviation units; Sent today: 42 crossing(s), 0 squawk(s); last 7 days: 49 and 0, Telegram being on since 30 October, so the 7 days are today and yesterday (as 0.99.16 counts them); the service active since 2026-10-31 09:10:48 CET. sudo zenithboard stats gaps over 14 days of samples: the callsign prefixes shown without an airline name, most frequent first, JFA 17, KLC 12, WMT 10, JAF 9, TAY 8, VJT 7, TUI 7 (data for the reading of 17 November; nothing built from it during the freeze). The command is sudo zenithboard stats gaps, not sudo stats gaps
31 October 2026, 00:17 The owner: is it said, in the menu and the console, that RadarBox is now airnavradar.com? Only partly: the menu said AirNav Radar (RadarBox), the console AirNav Radar, and both linked the station on radarbox.com, which answers 302 to airnavradar.com. 0.99.8 links airnavradar.com and says formerly RadarBox in the menu and the console (22 languages); previewed on the console's overview in French and Japanese
31 October 2026, 00:03 zenithboard status and zenithboard memory on 0.99.7: every feeder running, AirNav Radar among them with its MLAT off and its station named, one MLAT client in all; 44 aircraft in the decoder. Memory with all five networks, the wall, statistics, graphs1090, Bluesky, Telegram, the console and Tailscale: 332 MB for all the services, 436 MB available of 905, 90 MB of swap, load 0.85 on four cores. The wall 56 MB, Plane Finder 54, Bluesky 36 (109 at its peak), Tailscale 36, Telegram 22, collectd 22, Flightradar24 17, rbfeeder 9. The wall's journal dates the clock: 0.99.6 started at 22:27:08 on 30 October, so the month ends on 29 November, 22:27. AirNav Radar moves to the proven table
30 October 2026, 23:32 The owner's screen recording of telegram setup (22:41) and his phone: the bot answered at once; the chat list stayed empty until a new /start (his of 26 October was too old for Telegram's list), then you: Regis; messages switched on. The first crossings arrived in French the same evening (KL31L 22:53; S66373 23:05; LG68A and S66381 23:14), on the lock screen as a notification too. Telegram moves to the proven table, the squawks still to be seen
30 October 2026, 22:47 The owner: say how to claim the Pi on AirNav Radar (sudo rbfeeder --showkey --no-start, then where the key goes). Read on AirNav's claim page that evening: an account, then a Sharing Key field and a Claim! button; the key is not put in the address. 0.99.6's terminal gave the page alone, and said the claim worked from the Pi's public IP, which is FlightAware's way. 0.99.7 prints the steps. Asked the same evening whether AirTeamImages could add photos beside Planespotters: its terms (read that evening) forbid any reproduction outside a paid usage agreement and it has no public API; not planned (Roadmap)
30 October 2026, evening The owner: two features during the release candidate (c'est mon projet, je suis en RC maintenant mais on va faire deux choses), AirNav Radar as a fifth network with its MLAT off, and Telegram messages (the crossings of the sharing radius, the emergency squawks anywhere in range); then repart pour 30 jours de RC à la suite. 0.99.6. The hands-off clock starts again when the reference Pi runs 0.99.6: 30 days from that update (29 November 2026 if it is on 30 October). Its time goes here
30 October 2026, 09:05 Counted by hand on the Pi, 24–30 October: cat seen-* | sort -u | wc -l gives 6 855 distinct aircraft, cat seen-* | wc -l 16 994 aircraft-days; each aircraft came back on 2.5 days of the seven on average. The page's 6 838 at 08:19 was right. The same morning CodeQL's note #84 (a variable set twice in read_stats, left by 0.99.5) was cleared, no change in behaviour
30 October 2026, 08:19 The owner's /stats#7d: 6 838 distinct aircraft over the week against about 2 600 a day (he expected more), messages 340.5 M with busiest hour 18:00, 19 587 531, the farthest box 183 NM on 30/10 at 00:30 while the line below gave SQ308 at 262 NM on 25/10, busiest day 29/10 with 2 826. Checked: the week's distinct count is right (the union of the day files: the same airframes fly over day after day, as on 18–24 October); the busiest day is the day file's exact count; the messages are the decoder's quarter hours added up. Two wrong: the farthest box read the decoder's quarter-hour figure only, and the busiest hour added up seven days. 0.99.5
29 October 2026, 23:02 The social log on 0.99.3 (zenithboard logs social | grep "no route"), 13:52 to 21:41: twenty-two posts without a route, each with its reason. Seven registrations or addresses (T7NYD, PHVER, 399127…), never looked up by design, written adsbdb was not asked; eleven unknown to both services (CTM1277, BNJ1357, JFA41T…); four with a route neither could make fit (WMT6X, QAJ2, VLG6025). And one with no line: EJU81FR at 21:27, empty route, while hexdb had Brussels → Nice (asked by hand at 23:10). It went out after the two-minute wait with hexdb's answer still on its way, a case 0.99.3 did not write. 0.99.4: that case is written with the queue length, and a callsign that is not an airline's is said so
29 October 2026, 21:20 The owner: the fixes do not move the release candidate's date (arrête de pousser la date de la RC, on fait des corrections ici, pas des nouvelles fonctions). The hands-off clock runs from 25 October, 13:51 CET to 25 November 2026, as set on 25 October; the restarts written for 0.99.2 and 0.99.3 (28 then 29 November) are withdrawn
27 October 2026, 01:09 The owner, on his iPhone: 0.99.1 ended the console's reload at each tap on a section (c'est bon)
29 October 2026, 09:20 Five posts in twenty minutes with an empty route row: EZY9004, RYR95AL, EZY41AY, RYR73JA, RYR26XQ. Not 0.99.2's check: each asked by hand of adsbdb and hexdb. EZY9004 (Bristol → Madrid, Liverpool → Luton) and RYR95AL (Bergamo → Cologne from both, the aircraft flying south) rightly refused. EZY41AY and RYR26XQ unknown to adsbdb, known to hexdb (Luton → Zurich, Stansted → Perugia), which was asked only when adsbdb was wrong. RYR73JA: adsbdb's Charleroi → Turin is plausible but was not there at 09:02; the journal holds no failure between 08:40 and 09:10, and an unknown callsign or a full queue writes nothing. 0.99.3 (the owner's choice): hexdb asked when adsbdb does not know; a post without a route says why
28 October 2026, 19:29 The reference Pi runs 0.99.2 (ZenithBoard FlightInfo 0.99.2 listening, 19:29:27)
28 October 2026, 19:10 The owner's post of CCA897 at 18:22: Madrid → São Paulo for an Air China 787-9 still flying Beijing → Madrid (Google: due in Madrid 20:09), and ALT 38000 F on the picture. 0.99.2: a callsign's other leg is no longer believed; a row too long drops a space, not a letter
27 October 2026, 00:44 The owner's iPhone, away from home through Tailscale: the console blinked, every tap on a section reloading the whole page (Chrome's progress bar, a second of black), the home-screen app and the Chrome tab alike; /stats steady, the computer steady, the wall's server not restarting (its log). The bottom tabs also ran off the screen in French. 0.99.1 shows a section without the browser navigating, and keeps each label in its tab. A console fix: the hands-off clock goes on
25 October 2026, 16:52 The owner: the release candidate runs for one month to validate 1.0. The hands-off clock of 13:51 now ends on 25 November 2026 (it was two to four weeks); 1.0 that day at the earliest, if the open caveats are lifted. Said on GitHub (README, wiki) and on zenithboard.app
25 October 2026, 14:06 The owner: an update run from the console and a computer signed in by the phone, done several times today on two Pis. The web console's caveat is lifted
25 October 2026, 13:51 CET The hands-off clock starts (the owner): the reference Pi on 0.99.0, left alone from now. Two weeks on 8 November, four on 22 November. The same minute, CodeQL's two notes on the tests (a file read without with, in tests/test_server.py) fixed in the tests alone: nothing for the Pi, the clock goes on
25 October 2026, 13:42 Feature freeze (the owner): 0.99.0, the release candidate for 1.0, the same code as 0.55.2; fixes only from now on, features wait for 1.1. The hands-off weeks start when the reference Pi runs 0.99.0 and is left alone: their start date goes here
25 October 2026, 13:31 What 1.0 takes, reviewed with the owner: graphs1090, the farthest aircraft and the daily and weekly summaries seen on the Pi; UAT marked experimental, outside 1.0; a feature freeze (0.99.x) before the hands-off weeks; earliest late November. Roadmap
25 October 2026, 13:10 The owner's screenshot of Sharing → Summaries: links to posts 2 to 4, none to post 1 (All summaries runs the terminal's test). 0.55.2 links post 1, from a picture of its own
25 October 2026, 12:36 The owner: the registration as a hashtag in every post, to count on Bluesky how often an aircraft crossed, inside the 300 characters, with the hashtags and the custom templates too. 0.55.1
25 October 2026, 11:03 (same notes) Aviation units (kt, ft, NM) asked for wherever units are: 0.55.0 adds them as a third choice, on the wall, the posts, the console, the menus and /stats, which follows the wall's units for its distances since. Hindi for the console: 0.55.0, the pages and the posts, without a native review (the owner's decision; Tamil and Urdu wait for one)
25 October 2026, 11:03 The owner's notes from a fresh card and a week of figures (his folder zenithboard fixing stats). The key question on a fresh card read as an error (the expired key listed first): GitHub's current key now ships with ZenithBoard and is trusted without a question. ADSB Exchange's update printed the station's UUID in the pasted output: hidden. /stats counted 2 554 aircraft on 23 October where FlightAware counted 2 878 (11% fewer, every day alike): the quarter-hour sample missed what crossed between two; the wall's five-second look now adds them. The week's 6 673 is distinct aircraft over seven days, not the sum of the days. The week's messages, 386 530 3…, were cut in their box: now 386,5 M. The first steps showed the summaries switched on with no Bluesky account: Start here now offers the account, and the step shows no switch without one. "16 not upgraded": the system's own updates, said at the end of an update with the command. The Pages grouped and folding, the other devices folded, the triangle given room
23 October 2026, 08:21 The owner on the 41 rows à trancher: take the first proposal. 0.53.5 does, all 41, each checked to keep the English's placeholders and tags
23 October 2026, 08:01 The owner sent reviewed sheets for all 20 languages (194 rows, 153 with a correction). After --dry, 149 corrections were taken, 2 left out (one inexact, one an instruction), and 3 done by hand from the reviewers' notes. The 41 rows marked à trancher wait for his word. The first --dry also found a bug in tools/i18n.py: the sheet reader guessed doublequote=False from the header line, so plural rows came out ending in ""}. Four catalogues were broken that way before the tests caught it; they were put back and applied again with the reader fixed (0.53.4)
22 October 2026, 23:35 The owner, from his phone, on his own template in the console: saved, it shows and joins the posts' rotation; how to take it out, an emptied field or a button? Both worked already (an emptied field saved deletes them; unticking your own keeps them unused), but he asked for a real button. 0.53.3: Delete them all and Put back the previous ones, each asked first, in the console and the menu, and social template undo. Seen also: his template wrote at in a French post; a template's own words are his and are not translated
22 October 2026, 21:24 The owner: publish Swahili. Held back since 0.48.0 for a native review that has not come; 0.53.1 offers it for the web pages and the posts, its sheet left open for a Swahili speaker's corrections. The draft.txt mechanism stays, for a language to hold back later
22 October 2026, 20:58 The owner: add "install the app" in the browser for /console. 0.53.0: a manifest, the icons, a service worker that keeps nothing, and a card in System that says how for the browser in hand. Chrome's own Install button needs https and the Pi answers in plain http at home, so on the Pi's home address the browser's menu does it (Chrome: Install page as app; Safari: Add to Home Screen). Checked in Chromium on a secure address: the manifest read with no error, no installability error, the service worker registered on /console. Not yet tried on the owner's phone or computer
22 October 2026, 14:00 The posts in Indonesian and French, on the real account: both read right to the owner. One question: pukul also means to hit; is it right? It is: before a clock time, pukul is the standard word for at (the hour) (KBBI's own examples: pukul 17.00), and jam is a duration. What was wrong is beside it: pukul 13:50. Indonesian spelling (EYD) separates hours and minutes with a full stop, pukul 13.50, and so does CLDR, for Finnish too. The units were English everywhere as well: 839 km/h where Indonesian writes km/j, Russian км/ч, Dutch km/u. 0.52.1 writes the time of day and the metric units as each language does (feet and knots stay ft and kt). English is unchanged: 5 610 posts and summaries compared with 0.52.0, byte for byte
22 October 2026, 13:51 ADSB Exchange's stats on the real Pi, 0.52.0. Their myip page now has Feed Details filled: a Feed UID with Stats Ok, and the two ADS-B Exchange Anywhere links (Map and Stats). The Feed UID there is a short code of their own, not the UUID that ZenithBoard's adsbx-stats url puts in its link. zenithboard status lists ADSB Exchange stats running without the link; adsbx-stats status says ON, active and where the link is. zenithboard memory (RSS, without sudo): adsbexchange-stats 6 MB, two processes (their loop and its sleep). 558 MB available of 905, 165 MB of swap used, 242 threads, load 0.78 / 0.97 / 0.75, with every feed, graphs1090 and Tailscale running. What it uploads was not measured
22 October 2026, 12:46 The owner's screenshot of ADSB Exchange's myip page: ADS-B Connected, Feed Ok, MLAT Connected, Feed Ok, and under Feed Details: Feed UID Not Found, No Stats, You do not have the stats package configured. His question: why ADSB Exchange does not offer this. The feed is fine. That panel is fed by their stats package, a second program that ZenithBoard had never installed. Their script and service were read before anything was written: every 5 seconds the feed's aircraft list goes to adsbexchange.com/api/receive/, with the UUID and the Pi's power warnings. 0.52.0 offers it, off by default: adsbx-stats on, menu 1, the console. Not yet run on the Pi: what it costs a 3B (memory, upload) will be measured there
22 October 2026, 10:48 The owner on 0.50.0: what he meant was the Translation plan's your own message, with tags (step 3), not a post written from the console: take that out. He asked for the posts in the language he chooses, the built-in templates translated, and his own templates with a reminder of the tags as one more selectable style. He chose a setting of its own, English until set; the aircraft posts and the summaries in this release; his templates as a sixth style. 0.51.0 does that: SOCIAL_LANG (social lang, the console, the menu), every word of the posts and summaries through a catalogue in the 20 languages, CLDR's numbers, dates and countries, social template with its tags. The written post of 0.50.0 is gone. The pictures keep their English labels for now (steps 4 and 5)
22 October 2026, 09:30 The owner: no field lets me write a custom Bluesky message, although you gave me the commands. Asked which he meant, he chose both answers offered: a written post (his own words, text only) and the automatic posts' words in the console. 0.50.0 does both: social say and Sharing → Write a post, the first post the console can send, after a confirmation that shows the words; and the place name, radius, rules, town names and hashtags as fields. The templates with tags of the Translation plan (step 3, your own message, with tags) are a different thing, still to come
22 October 2026, 09:11 The daily summary went out, four posts from 08:45, by the owner's look at the feed. The time was set to 08:45 at 08:26 (daily summary now every day at 08:45 in the journal), with last still the plain day 2026-10-21, and the service restarted by hand at 08:27 on 0.49.1. The same loop run here, against stand-ins, gave 08:45, 08:47, 08:49 and 08:51. The airline and type lists covered the last 24 hours this time. why before 08:45 kept the summary lines and listed the starts apart, as 0.49.1 meant to. Its starts also correct the row below: the service started 0.49.0 at 08:13, eight minutes after that why, then 0.49.0 at 08:22 and 08:24 and 0.49.1 at 08:26. The likely reading: the update to 0.49.0 was still running at 08:05 and restarted the service at its end, rather than having stopped
22 October 2026, 08:05 sudo zenithboard social summary why, after the update to 0.49.0. The cause of 21 October: the daily summary's post 1 went out at 09:20 under 0.40.0 (the journal has its Bluesky address). Posts 2 to 4 were skipped, as designed: the airline and type lists do not cover the last 24 hours yet (counted since 2026-10-20, shown from the summary of 2026-10-22). Under the one-a-day rule, that post made the day done, so the times moved later (21:00, 23:35) did nothing, silently. That post is no longer on Bluesky (its address opens nothing; the posts of 22 October open). Also seen: the service still ran 0.47.0, since 00:11, with the files at 0.49.0. An update restarts it at its very end, so that update stopped before its end, or the files came another way; asked how it was run (09:11: more likely still running, the row above). 0.49.1: why says so with the restart command and lists the starts apart. Twenty-four starts on 21 October had filled thirty-two of its forty lines
22 October 2026, 07:58 The social service's journal from 21 October 21:27 to 22 October 07:49 (still 0.47.0 at that hour): aircraft posted all night, not one line about a summary, no error, no traceback. The service restarted six times that evening, once per update (0.43.0 at 21:29 to 0.47.0 at 00:11). At 23:35, with the daily summary set to 23:35 and the service up since 23:19 (0.46.0), nothing was even attempted: before 0.48.0 a summary falling due wrote nothing until it was posted or skipped, so a summary already marked as scheduled for that day explains the silence, and the hour before 21:27 is not in the paste. Waiting for sudo zenithboard social summary why on 0.49.0, which reads two days of the journal (it came at 08:05, the row above)
22 October 2026, 00:53 No summary in the Bluesky feed on 21 October, and zenithboard logs social | grep -i summary empty (the last hundred lines only). The owner: the day lock is in the way, the schedule is enough to keep a summary from going twice. 0.49.0: one summary per scheduled slot (a time moved to later today goes today; a time already past, or a summary switched on after its time, goes from the next one) and sudo zenithboard social summary why, which reads two days of the service's journal; the cause of the 21st waits for its output
21 October 2026, 23:35 The daily summary, its time moved to 23:35 (twice, from the console), did not go out at 23:35; Try drew the posts, as it should, without posting. 0.48.0 guessed that a 21:00 one had gone out earlier the same day (one a day, by design): corrected on 22 October, 00:53, nothing went out as a summary on 21 October (the owner's look at the feed; `logs social
21 October 2026, 23:06 The owner's answers to the translation plan: both Chinese scripts, English's flag the UK's and the US's merged, no unreviewed mark but a review document per language with the English as reference (corrections he puts back), the terminal output in English, and Italian too. 0.47.0: the framework and French on the console, /stats and graphs1090's page, previewed at desk and phone width with a stand-in
21 October 2026, 22:25 Tailscale works on the reference Pi ("ça marche"). The owner asked for it in the console too, and opened the next project, translations, starting with an analysis: Translation plan, measured on 0.45.0 (about 1 000 strings, 8 000 words per language; Noto CJK +91 MB on the card and +4.4 MB while a card is drawn). 0.46.0: System → Tailscale in the console, previewed with a stand-in
21 October 2026, 21:42 The owner asked for Tailscale in the menu, with how to use it and how to take it away. 0.45.0: 3 Connectivity → 4 Tailscale and zenithboard tailscale on|off|status|remove|console, the install steps of Tailscale's own page (pkgs.tailscale.com, read the same evening), tested against stand-ins; not yet run on the Pi
21 October 2026, 21:02 On the reference Pi 3B (0.43.0): graphs1090 in ZenithBoard's look, all services running (212 MB for all of them, 518 MB available, 44 MB of swap), health OK, both watchdogs and the kept logs on. The owner moved the data to another card and it followed; the networks' identities were not tried yet; the position, the altitude and the units "did not follow" a restore (imported, not applied: fixed). He asked for graphs1090's look by default, a simpler move written for non-specialists with WinSCP, and a menu by use case: ten-entry menus, and the console switched on before ADS-B and the wall "because it is obvious to you, not to everyone". His /stats screenshot still carried the 0.42 caption of the sky by hour: a page loaded before the update. Asked why the console turned black and white: it follows the wall's colour, set to white. 0.44.0
21 October 2026, 17:55 A new card on a Pi 4 Model B (2 GB): zenithboard health OK (50 °C, 1 661 MB free), 188 MB in use with nothing installed but the system. Switching the console on before the wall stopped on chown: invalid user: zenithboard (the wall's user is made by menu 2): 0.43.0 makes that user where it is first needed and says the console needs the wall. The owner asked for graphs1090 in ZenithBoard's look, readable, and for /stats to follow the 24 h / 7 days / 30 days toggle everywhere, in the order of the summary's posts, each airline's flag first. Built and checked on the reference Pi's backup of 12:51 (its real statistics and graphs1090 history), at desk and phone width. 0.43.0
21 October 2026, 12:57 The first backup on the reference Pi, made from the console in 0.42.0: 24 MB, downloaded. Read back part by part: 572 statistics samples (15 October 14:00 to 21 October 12:45), seven days of aircraft lists, the posting history, 68 graphs1090 databases whose last point is 12:50, the kept logs from 8 October to 12:51 (13 boots), the settings and the ten features on; no network identity, as a console backup should be. The owner asked whether such files pile up on the Pi, how they sort, and where the import is. 0.42.1: backup show checks and says each part's span; backup list, backup clean --keep N; the console's copy (one, replaced) can be deleted and is downloaded under a name with the host and the date; the card says where the import is and why. 0.42.1
21 October 2026, 12:17 Before writing the new card, the owner asked whether moving FlightAware and the others would be held back by the MAC address, and for the logs to follow with a way to switch them on again, from the menu or the console. Measured in the code: settings export left the statistics, graphs1090's history, the kept logs, the posting history and the identities behind, and an import brought STATS=1 or NETWATCH=1 back with no timer behind them (the watchdog and the kept logs are not in config.env at all). 0.42.0: backup make / show / restore. Built and tested against stand-ins; not yet run on the Pi. 0.42.0
21 October 2026, 11:00 The owner is writing a new card from Raspberry Pi's own image and asked for every step, from the card to a setup finished in the web console, and made it a 1.0 proof. Written as Fresh install (21 steps, the feeders' identities to keep, a checklist); the console's first steps grew from five to nine and can be opened again. Not yet followed on a real blank card. 0.41.0
21 October 2026, 10:40 With 0.40.4 the four posts of the 24 hours went out (summary daily post to post4): 1 138 aircraft, 90 cargo, 45 business jets (Germany, Portugal, United States…), 6 military (United States: C-17 ×3, Learjet 35; Germany: Unknown type ×2), 1 helicopter, 6 light aircraft. The owner asked for flags on general aviation, the German unknowns named, and the period said plainly on each card (21 OCTOBER read as that day). Seen too: Bluesky's ALT badge covering ZENITHBOARD. 0.40.5
21 October 2026, 10:06 The first scheduled summary went out (daily, 09:20, on Bluesky): 933 aircraft, 59.5 million messages, busiest hour 17:00, 42 crossings posted — but with no airline, no model and none of posts 2 to 4, the lists having begun the day before (the rule of 0.34.1 wanted them to cover the whole period). 0.40.4: shown from half the period, saying since when
21 October 2026, 09:40 The owner: what 0.40.2 did for the buttons holds for the whole interface. 0.40.3: every control's output in its own card
21 October 2026, 09:29 The owner clicked Health in the console and did not know where its answer was (the Activity drawer). 0.40.2: a button's output appears under it, in its card
21 October 2026, 09:10 The owner showed the Security tab: ten CodeQL notes open (eight Empty except, one mixed returns, one in the vendored QR code library). Fixed in 0.40.1, the library left out of the analysis with the reason; and the running version shown on the console's sign-in page, linked to its release notes
21 October 2026, 08:47 The owner said yes to signing a computer in with the phone: QR code on the computer, the same four-digit code on the iPhone, yes, signed in without the password. 0.40.0, off by default (console phone-signin on); previewed with a computer and a phone side by side, not yet on the Pi
21 October 2026, 08:34 The console in use on the Pi (0.39.2), recorded: the iPhone paired by QR code (08:23), the PC signed in with the password, statistics off and on, MLAT guard, graphs1090 installed from the page (about a minute), the summaries' switches. Seen: changes made while graphs1090 installed waited, and the page blamed the helper for it, repeatedly; a past job clicked in Activity opened its output out of sight. 0.39.3. The owner asked why the PC is not signed in by a QR code already used: a code signs in the phone that scans it, once; signing a computer in with the phone is written down as a proposal
21 October 2026, 08:05 First try of the console on the Pi: neither the password nor the QR code worked, this pairing code is used or out of date. The wall had been restarted by the update before console on made its folder, so its sandbox kept the folder read-only and every code was refused (failed codes then count against sign-in tries too). 0.39.2: the code no longer depends on that write, console on restarts the wall, the page says when the folder is read-only. The owner also asked whether the terminal is needed at every visit: no, and sessions now last while used (a week with the password, a month between uses for a phone)
21 October 2026, 07:56 sudo zenithboard memory after 0.39.0: 180 MB for all the services, 619 MB available, load 1.1. Plane Finder at 43 MB (27 the evening before), the one service the update did not restart: to watch. Two labels were cut or missing in the WHAT column (name on the network (mDN, unattended-upgrades unnamed): 0.39.1
21 October 2026, 07:34 The owner asked for a break-glass procedure for the console, step by step, and a way to start again. The terminal is the way in, by design; the Web console page now lists each case (password forgotten, locked out, phone lost, a job queued or stuck, a setting to undo, the page not loading) with its commands, and how to start again from zero or take the console out. Documentation only
21 October 2026, 01:06 The owner asked for a web console to try, removable if it does not satisfy, "right the first time": a wizard, ZenithBoard's look, a password or Face ID / a QR code on his iPhone ("careful with the black background"), a phone layout, a menu with no endless scrolling, links to the pages that exist or will. 0.39.0, built while he slept. Face ID / passkeys are not possible over plain HTTP at a home address (WebAuthn needs HTTPS under a domain); a one-time QR code, dark on white, signs the phone in instead
21 October 2026, 00:56 End of the evening: 0.31.0 to 0.38.0 shipped in one night (statistics, graphs1090, the summaries and their four posts, real flags, the page's new look and toggle, health and memory). The owner asked for the Roadmap and this page to say where things stand: Plane Finder moves to proven, the summaries and graphs1090 join Now, and a web console is written down as a proposal, not started
21 October 2026, 00:25 The owner liked the page ("so zenithboard") and asked for the countries column aligned with the other two and named most registered in; and for summaries every 24 hours, 7 days and 30 days, each at a chosen day and time, with the posts picked, and what two at the same time would do. 0.38.0; previewed on made-up statistics
20 October 2026, 23:58 sudo zenithboard memory again: 155 MB for all the services after the evening's restarts (social 22 MB, Plane Finder 19, the wall 19), 653 MB available; /social/preview2–4 now say why they have no picture. The owner shared ZenithBoard's banners (the name in amber dots, the split-flap OVERHEAD NOW) and asked for the statistics page in that look, with a 24 h / 7 days / 30 days toggle that every chart follows except the sky by hour, and the weekly posts' lists on it, provided it does not overload the Pi: 0.37.0. Previewed at desk and phone width on made-up statistics
20 October 2026, 23:43 From the owner's pastes. 0.34.0's new screens on the Pi: health says Card: 6% full: 3.5 GB used of 61.0 GB, 54.9 GB free (the card: 62.5 GB); memory gives 212 MB for all the services, Plane Finder 27 MB (in Hardware). 0.35.0's weekly test: the first post without airlines and models (lists not covering the week), the second skipped, so /social/preview2 answered a bare not found: 0.36.1 says why. The /stats lists were small (Ryanair 27): one day of lists, as the page says; TAY was shown bare, now ASL Airlines Belgium from the cargo list. The owner found the dot flags of 0.36.0's cards crude: real flags at scale in 0.36.1. Redrawing them showed that 0.36.0's search over the country table gave the United Kingdom's military block (43C000) to no country, the table's blocks being nested: fixed and checked against 200 000 addresses
20 October 2026, 23:26 The owner sketched the second weekly card: each cargo airline with its most seen model, and the business jets' countries of registration with flags; and asked for two more weekly posts, military aircraft by country and general aviation. 0.36.0. The countries come from the ICAO address blocks (Annex 10, as tabulated by tar1090), the military mark from tar1090's database (the first flag of an entry: 10 military, as tar1090 reads it) and tar1090-db's military address ranges; checked on the real database (23 065 entries marked 10). Previewed on made-up statistics; not yet on the Pi
20 October 2026, 23:06 The owner asked for the models most seen beside the airlines on the weekly card, and a second weekly post two minutes later: cargo airlines and business jets by make and model. 0.35.0. The cargo list's 40 codes were each checked against a published source; Turkish, Qatar and Emirates freighters fly under their passenger prefixes (THY6147, QTR8486, UAE9807 are 777Fs), so a callsign cannot count them. Previewed on made-up statistics; not yet on the Pi
20 October 2026, 22:53 First social weekly test on the reference Pi (0.33.0): 14–20 October: 6 435 aircraft heard, 353.3 million messages decoded. Busiest day: Friday, 2 745 aircraft. Busiest hour: 17:00. Most seen: Ryanair (14), British Airways (8), easyJet (8). Top type: BOEING 737-800. The owner saw at once that 14 Ryanair cannot be right in a week of 6 435 aircraft: the airline lists had begun with 0.31.0 two hours earlier. 0.34.1 leaves the airlines and the type out until their lists cover the week, and writes the type as the posts do
20 October 2026, 22:42 Plane Finder installed: the reference Pi now feeds four networks (ADSB Exchange, FlightAware, Flightradar24, Plane Finder), and Plane Finder's own page showed its feed. zenithboard health then: 575 MB free, no swap, 34 °C, power OK, card 6 % full, previous boot ended normally. The owner asked for the card's size on that screen and for a command showing threads and memory: 0.34.0 (Card: ... GB used of ... GB, zenithboard memory)
21 October 2026, 00:30 The weekly summary post (0.33.0), the last part of the owner's choice when Grafana left. Previewed with the real code on thirty days of made-up statistics, amber and turquoise. Not yet on the reference Pi: sudo zenithboard social weekly test there is the first real one
20 October 2026, 23:30 graphs1090 offered as an option (0.32.0), the second half of the owner's choice when Grafana left. Its installer and uninstaller were read in full (commit 2f96314); running it in the session's own machine was not allowed, so on and off are tested against stand-ins only. Not yet run on the reference Pi: what zenithboard graphs status says there (collectd's memory, the size in /run) goes here
20 October 2026, 23:00 With ACARS and Grafana gone, what Grafana would have shown went onto the wall's own statistics page (0.31.0): the sky by hour over fourteen days, the airlines and types most seen, how high they fly. Checked against the real server with thirty days of made-up samples, light, dark and phone width. Not yet on the reference Pi: the lists start filling on its first quarter hour after the update
20 October 2026, 22:00 The owner decided to take ACARS out, and Grafana with it once its use as a statistics board was weighed (100 to 200 MB on a Pi 3B, a third-party plugin, publishing only by opening the Pi to the internet or through a cloud account): 0.30.0. The reference Pi never had them installed
20 October 2026, 21:30 The owner asked for other words than the factual post; five shapes proposed on RYR5ZK, all chosen, with the reason said first. 0.29.0
20 October 2026, 21:00 Four posts of the evening (RYR5ZK, RYR6JR, SKV615, EJU86EN): on RYR5ZK the north of the map was full of Belgian and Dutch towns unrelated to a track from Namur to Luxembourg. 0.28.0 names along the track first, with three large cities around for orientation (the owner chose that over the track alone). First try: the towns along a track through Belgium filled the 26 names before any city around; three places are now kept for them
20 October 2026, 19:05 0.27.3 on the Pi, first post: RYR5236 over Metz on the light map, 26 names (the most a map holds), spread over the whole strip, Pont-à-Mousson, Nomeny and Rémilly in the south where the towns alone left it empty. Now the dot-matrix example in the README and Social sharing
20 October 2026, 18:50 SWR3AG on the satellite map: seven towns named, all on the edges, nothing in the middle. A target of six names instead of three, tried first, changed nothing (seven already). 0.27.3 names a village wherever no name is within 120 pixels: Boulay-Moselle, Bouzonville, Rémilly, Verny, Morhange, Château-Salins, Dieuze on a preview of that track, with the region's list made in the browser from the same nine GeoNames files (55 164 places, as on the Pi)
20 October 2026, 18:43 First post with the region's own list: CFE844 (BA CityFlyer, London → Milan) on the dark map names Hayange, Florange, Villerupt and Aubange, none of which the shipped list of 5 000 had on a map with this many towns. Now the dark-map example in Social sharing. 0.27.2 on the Pi made the list from nine files, 55 164 places, as expected
20 October 2026, 18:30 0.27.2, from the line above: the countries are chosen by their places of 500 or more within 400 km, not by one-degree cells with a margin. Around Trémery: AT BE CH DE FR IT LI LU NL, 24 MB; CZ and GB are no longer downloaded. Run social places on again to see it; the list kept is the same
20 October 2026, 18:20 0.27.1 on the reference Pi: social places on downloaded eleven country files (AT BE CH CZ DE FR GB IT LI LU NL, 29.4 MB in all) and kept 55 164 places in 676 KB, the same count as the test run in the browser. Two files gave no place within 400 km (CZ and GB: their cells reach within the margin, their places do not). The social service afterwards: 30 MB of memory, as before
20 October 2026, night The owner asked whether the town names would work for a user in Indonesia. Measured: GeoNames knows the population of 540 of Indonesia's 258 055 populated places (ID.txt); the shipped lists name 479 places of 5 000 there against 2 074 in France. Test maps with the shipped lists named nothing over East Sumba and Pontianak. 0.27.0 makes a region's own list on the Pi from GeoNames' country files, ranked by population or administrative rank; checked at twelve places on four continents with the real files and tiles (see Social sharing)
20 October 2026, evening DAH1217 (Air Algérie, Metz-Nancy → Algiers, 3 030 m) went out on the light map with no town named. Reproduced with the real code: two minutes of track make a 20 km by 40 km strip, east of the Moselle when the aircraft is a few kilometres east of the receiver; every town of 5 000 is in the valley, just beyond the edge. 0.26.2 adds the villages of 1 000 to 4 999 (GeoNames) on a map with fewer than three names: Vigy, Courcelles-Chaussy, Boulay-Moselle there. Asked for by the owner
20 October 2026, afternoon The owner saw the town names twice, a few pixels apart, on the light map (EWG8QR over Nancy): 0.25.3 drew ZenithBoard's names over EOX's OpenStreetMap layer, which writes its own. EOX's layer list (WMTSCapabilities) has no OpenStreetMap layer without labels; terrain-light and streets have none. 0.26.1 builds the light and dark maps from those two; previewed with the real code and tiles on EWG8QR's track, approved by the owner
20 October 2026, noon EWG8QR (4D2087) went out with no registration, though Planespotters had a photo of 9H-EWG. Checked: adsbdb unknown aircraft, hexdb n/a; tar1090's database knows it. On the Pi: tar1090's database at /usr/local/share/tar1090/git-db/db, version 3.14.1721; readsb does not load it (no --db-file). 0.26.0 reads it, off by default: sudo zenithboard data localdb on
20 October 2026, late morning Eight posts made with 0.25.3 (10:42 to 11:14): the names read on the two OpenStreetMap maps, none is cut by a track any more, Frankfurt-Hahn → Palma de Mallorca names towns. One left: KLM105 read Amsterdam Schiphol → Cape Town in the text. 0.25.4 keeps the town alone when the name opens on the town by the airport
20 October 2026 A recording of fifteen posts made since 0.25.2 (00:35 to 07:51), on every map: the smaller, more numerous names read well on the terrain and satellite maps. Three things to put right, in 0.25.3. The track cut letters out of the towns it crossed (N ncy, Lu embourg, Pon -à-Mousson): the names are now drawn over it. The OpenStreetMap maps' own labels were an unreadable grey at the post's size, the dark one especially: those two maps get the same names. RYR7CW read Brussels South Charleroi Airport → Turin Airport (Caselle Airport): hexdb's second-opinion route gives airports without a town (checked on its real answers for EBCI and LIMF), so the town is now taken from the name: Charleroi → Turin. Previewed with the real code and EOX's real tiles before merging. Brussels Air stays cut: the boards are fourteen letters wide
19 October 2026 The owner asked why another Bluesky bot shows Planespotters and JetPhotos pictures and his posts cannot, and wanted to stay legitimate. Planespotters' photo API terms, read that day: credit next to the picture, the picture a plain link to its page, the image fetched straight from the API's address by the screen that shows it, never stored, proxied or passed on, the answer cached 24 hours at most. The wall's credit and link were right; its proxy (the Pi fetched each picture and served it from a memory cache) was not. 0.24.5 removes it. JetPhotos: personal, non-commercial use of the site only, no API: not used
18 October 2026, evening Three more posts looked at (TRA21C terrain, NJE221Q satellite, EWG09NC normal map). The normal map names its towns; the terrain map named only Amnéville in forty kilometres, the satellite map only Grand Est and Épinal. 0.24.4 names the towns of 5 000 or more itself on those two. The owner asked whether the scale was right: measured on the posts, TRA21C (997 km/h, 3 min, 50 km) drew 48 km against its 10 km bar, EWG09NC (926 km/h, 6 min, 93 km) about 98 km, NJE221Q (928 km/h, 7 min, 108 km) about 102 km: right within the speed's own variation
18 October 2026, late afternoon Three real dotmap posts looked at (JAF1DY night map, KLM50X satellite, RYR1CA terrain): the maps, credits, flags and legend as intended; the owner found the track line far too thick. 0.24.1 draws it at 3 pixels with a 6-pixel dark edge, from 5 and 11. The README now shows the RYR1CA post
18 October 2026, afternoon 0.23.0 on the Pi: social status says dotmap and flap in turn, the colours in turn, the maps in turn, 140 posts in the last day, 41 MB at rest and 102 MB at most for the service (Node 100 MB). The owner confirmed the dark blue gone everywhere and asked for the dots-only look to leave the choices too (0.24.0: it is gone; a setting still on it posts dotmap). The statistics page named the day's farthest as 18/10 13:09 · TF-ICE · BOEING 737 MAX 8 · 11 890 m · 322 km, with no flight and no route: at 322 km the decoder had the positions before the callsign. 0.24.0 keeps the callsigns heard in the last hour and fills it in. /social/preview still showed a picture from an old version: the page shows the last social test, and none had been run since
18 October 2026, morning A Bluesky post: Eurowings EWG11S, D-AIDO, in the dot-matrix look. The German flag beside the registration had lost its right-hand edge under the track panel's frame (the panel started at dot 86, where the flag's outline ends). The owner asked for a real map in place of the dots-only one on the right, in the same rotation as the split-flap posts, and for the dark blue to go (turquoise stays). 0.23.0: the panel starts at 88; a third look, dotmap, lays the split-flap look's map beside the dot wall, and alternate turns between it and flap; the dark blue is gone, a setting still on it named by the update
18 October 2026, 00:15 Two screen recordings of a walk through every menu, after the 0.20.1 update. Seen: Uninstall everything sat between Update and Exit as entry 7, one Enter from a Yes; the ACARS tag [NOT recommended on this Pi] was cut off at the screen's edge; status called a 1 GB Pi a low-memory board to skip ACARS on, the day after that board was measured comfortable; a full update ran apt on every feeder for a change to ZenithBoard alone; no way to keep the settings as a file. The Debian patches were switched on during the recording. 0.21.0 answers each: a Maintenance entry with save, restore and an uninstall that asks for the typed word; a quick update; a shorter tag; the measured wording
17 October 2026, 21:00 Would the card do better in a Pi 4 (2 GB) that was to hand? The figures said no: 656 MB of 1 GB available with readsb, three feeders, the wall and social sharing, 36 °C, no under-voltage or throttling in seven and a half hours. The Pi 4 stayed in its drawer. The hardware table, which said estimates, not yet measured, now starts from these figures, and a Hardware page says what moving a card to another Pi would change
17 October 2026, night The owner's rule for the feed, with a 10 km sharing radius: every aircraft that crosses it, once per crossing, nothing capped — the same aircraft not again five minutes after it leaves, but its return flight, or the next day's, yes. 0.20.0 makes that the behaviour: a crossing ends an hour after the aircraft left the radius, a new callsign is a new flight, one that left before its turn is still posted, and the default gap is none. The Bluesky session is now reused, since a sign-in per post could pass the account's 300 a day. The registration's flag moves beside the REG row, on the wall and in the posts: beside the airline, a German flag read as Lufthansa's country
17 October 2026, evening Four hours of the service on its own (still 0.19.7: the two updates after it not yet installed): 19 posts between 16:02 and 19:51 under ALL with the ten-minute floor, about five an hour; 35 to 36 of the 238 looks an hour found a candidate; 42 MB at rest, 104 MB at most. The hourly line said Node 104 MB at most too — Python's figure, which the kernel credits to a child started by vfork. Node now reports its own (0.19.10)
17 October 2026, five o'clock Two split-flap posts, RYR8526 and EWG6ZP, went out with empty route rows. Not the look: asked by hand, adsbdb gave other flights under those callsigns (Berlin → Stansted, Salzburg → Aalborg), which the wall rightly refuses for an aircraft over Metz heading south-east or north; hexdb.io, the second opinion, gave the right ones (Charleroi → Reggio Calabria, Menorca → Cologne), but a few seconds after the post. The wait for lookups now covers the second opinion too (0.19.9)
17 October 2026, four o'clock The first post with every 0.19 change in it, EWG9NU at 16:02, eighty seconds after the service started: all nine rows filled (the wait for adsbdb did its job), →COLOGNE by its city, the first step of the varied pictures (dots, amber), no credit line since there was no map. One flaw on a phone: altitude 8 at the end of a line and 600 m on the next. Numbers now hold together (0.19.8)
17 October 2026, quarter to four First satellite post, a Bellanca Decathlon (LX-FUN): the empty airline and route rows are right for a light aircraft, but the map credit took three lines of the post. It is now one short line at the end, the names as links; the full text stays on the picture and in the alt text (0.19.6)
17 October 2026, twenty to four Five minutes after the restart the service posted KLM24V with empty registration, model and route rows: its track, followed from far out, was long enough at once, while the wall had only just started asking adsbdb about it, which it does when an aircraft comes inside the radius it is asked about. The wall now says which lookups are on their way, and the service waits for them (up to two minutes). The owner also asked for posts that vary: the two looks in turn, the colours in turn, the maps in turn (0.19.5)
17 October 2026, half past three After the update the service ran 0.19.3 at once (restarted by the update, its first line said so), sudo social status showed the 14:21 post, four aircraft posted, 24 types noted and a 333 km range record, and as pi it said it could not read rather than never. One claim did not hold: systemctl status still had no Memory line. Raspberry Pi OS boots with the kernel's memory controller off, so systemd can neither show nor cap the service's memory. The service now measures it itself (0.19.4)
17 October 2026, three o'clock The Pi's own log, read line by line. The social service had run since 13:15: the updates that followed had installed new code without restarting it, so it still ran 0.19.0. social status run as pi said no account and Last post: never on a Pi that had posted twenty minutes before — pi cannot read the account file or the state folder, and the status did not say so. The gap had been set to one minute for a test, which explained the post that looked too early. Memory: 650 MB free on a Pi 3B, the service's own use not shown. All four fixed in 0.19.3, and the route rows now give the city (→SAN JOSE, not →JUAN SANTAMAR)
17 October 2026, half past two The service posted by itself for the first time — a Lufthansa A340 under heavy, split-flap with the night map, read on a phone. It came 48 minutes after the previous post, which looked like the two-hour gap failing; reading the code for that found a real fault — the service kept its memory in RAM and wrote it back every fifteen seconds over what Post now had written — fixed in 0.19.2. The post itself was not that fault: the gap had been set to one minute for a test, as the Pi's own log showed an hour later
17 October 2026, two o'clock Service on, radius 50 km, nothing posted in an hour while test always showed something: no military, helicopter or heavy had passed, and rare/farthest were still warming up — as designed, but the rules list made it hard to see, and its last line, any, read like a sixth kind of aircraft. It is now ALL at the end of the list, and ALL is every line at once: the special aircraft first, the ordinary traffic otherwise (0.19.1)
17 October 2026, late afternoon The first posts read on a phone: 10 670 m over Trémery was taken for a distance from Trémery, and the track's legend (10 MIN over 10KM) for a riddle. The text now says altitude 10 670 m · above Trémery, and the legend is a key. The owner sent a mock-up of the dot-matrix picture (flight and registration labelled), asked for the split-flap look with a map and a choice of map, and dropped Mastodon. The map could not come from OpenStreetMap's own servers — their policy rules out automated fetching — and EOX's tile service, free with attribution and a link back, took its place; the tiles were checked in a real browser, layer by layer, before the code used them. The split-flap picture is drawn with Pillow, not a browser, so it works on a Pi 3B (0.19.0)
17 October 2026, afternoon The first real post went out on Bluesky (an Etihad A380, from Post now). Read on a phone, it asked for changes: the text now starts with the flight (Flight Etihad #ETD11K) and ends with the place (above Trémery); the picture has a smaller callsign with the registration under it, and the time with its GMT offset, summer time included. The scale bar's 10KM had pushed it out of the frame, and an A380's ten minutes ran past the edge: the scale now goes to 500 km per bar and always fits (0.18.3)
17 October 2026, noon No automatic post: reading the service again showed rare counting looks instead of aircraft and farthest comparing an aircraft with its own record — neither could fire, while test, which judges differently, said they would. Fixed in 0.18.2; the real service loop now posts a Dash 8 under rare in a test that feeds it a minute of a real-looking track
17 October 2026, late morning The first social test on the real Pi stopped with Permission denied on /run/zenithboard/social: the folder is systemd's, made for the service, and the command line ran outside it. Fixed in 0.18.1 the same hour
17 October 2026 Social sharing (0.18.0): Bluesky and Mastodon, a picture of the wall with the aircraft's track, drawn by the Pi without a browser. Decided against X (paid API since February 2026), Threads and Instagram (both fetch the picture from a public address). The Planespotters photo stays off the posts: their terms forbid re-hosting it. Mocked up in a browser and checked against the real renderer before being written; the networks are tested against fakes until the first real post
15 October 2026, night (4) The gear panel gained a link to the statistics page, shown only while statistics are on (0.17.4)
15 October 2026, night (3) A Pi showed the menu at 0.17.2 and the wall at 0.17.1 at the same time: updating only the ZenithBoard code deployed the new files without restarting the wall server, which runs the old ones until it is told otherwise. Every update now asks the wall what it is running and puts it right (0.17.3)
15 October 2026, night (2) The wall's gear panel and the statistics page now name the version they run, linked to its release notes (0.17.2)
15 October 2026, night Two faults the real wall showed that no test had: the union flag, being nearly all dark navy, was invisible on the black wall (British Airways looked flagless), and the model row scrolled AIRBUS A-380-800 where A-380-800 would have stood still. Both fixed in 0.17.1, both now held by a test. Neither was reproducible from the code alone — the first needed someone to look at the wall, the second needed a real Etihad A380
15 October 2026, late Airline names learned from the route answer (0.17.0): the wall had been throwing away a name adsbdb sends with every route. Checked against the real answer for EZY1234, read in a browser the same day
15 October 2026, evening (2) The statistics page was smeared on its left edge once an hour passed a million messages: the axis drew about three hundred labels on top of each other. Seen on the real Pi's first busy afternoon, fixed in 0.16.3 and tested against the real axis function
15 October 2026, evening zenithboard security on the real Pi after 0.16.2: only two warnings, Debian patches not switched on and SSH passwords, the second kept on purpose (SSH is reached from home only). All five sandboxed services and every listening port recognised. The statistics portal is running
15 October 2026, afternoon Statistics (0.16.0): a 15-minute sample, the /stats page and zenithboard stats, off by default. The page was checked in a real browser against the real server with thirty days of made-up samples before being merged
15 October 2026 Second security pass (0.15.0): the wall answers home networks only, every service sandboxed, Grafana's default password closed, optional daily Debian security patches, a release per version, and zenithboard security to check the running Pi. Tested here with stubs and offline unit scoring; Grafana itself and the Pi's own systemd-analyze are the two things only the real machine can confirm
14 October 2026, night The wall gained the registration and maker flags (0.14.0). The manufacturer logo that was asked for alongside them was refused: a trademark redrawn in dots is still the trademark. The country where the airframe is built carries the same meaning and is free to draw
14 October 2026, evening A generated airport-hall ambience was added (0.13.0) and removed again an hour later (0.13.1): it did not sound good enough to keep. Turquoise and blue joined the wall colours the same evening and stayed
13 October 2026, morning No aircraft photos on the wall, only the drawn plane. The address check added in 0.11.1 allowed planespotters.net, but the pictures are served from plnspttrs.net: every photo had been refused in silence since. Fixed in 0.12.4. The test that should have caught it had been written from a guessed address instead of the real one
13 October 2026, morning main is now protected by a branch ruleset: no deletion, no force push, signed commits only, changes go through a pull request merged by squash, and the checks test, analyze (python) and analyze (javascript) must pass. The first attempt named the checks analyse with an s, which do not exist, so GitHub refused every merge until the spelling was fixed
12 October 2026, evening Security pass, all on the real Pi: the wall server's sandbox scores 1.3 in systemd-analyze security and the service runs normally inside it. The first signed-update checks found two bugs of their own — a download folder deleted before use (curl: (23), fixed in 0.12.1) and only GitHub's expired key being recorded, so every update asked again (fixed in 0.12.3). After that, updates install silently. CodeQL's first run found a real header-injection path (fixed in 0.12.1/0.12.2); two deliberate file permissions were dismissed as Won't fix with the reason written down
12 October 2026 The new monthly link check found, on its first run, that readsb-uninstall.sh has not existed upstream for some time: removing readsb had been doing nothing at all. It now refuses out loud and prints the manual commands (0.11.2)
11 October 2026 First full day of the gain log: 96 measurements at 48.0 dB, median 2.83 % strong messages, range 271–336 km, 65 million messages. Four quarter hours went over 5 %, and the two before dawn turned out to be an artefact of low traffic rather than overload — the finding that a quiet hour cannot judge the gain in either direction
10 October 2026 Cause found: the Wi-Fi extender had a schedule turning it off from 02:00 to 06:00. Nothing wrong with the Pi, the signal or the weather. Schedule removed; the night of the 10th is clean from end to end
9 October 2026, 01:00 Wi-Fi lost. The network watchdog restarted the network and the Pi recovered on its own, nobody touching it. Three restarts that night, three hours apart — the gap rule working, not three faults. The diagnostic lines added in 0.10.2 are what identified it: the saved network invisible while other networks were still in range, which points at the access point rather than at the Pi
8 October 2026 Gain walked up and down the real R820T steps. 48.0 dB gives 4–5 % strong messages at rush hour, inside the 1–5 % guideline. Left there
October 2026 A TV browser's top bar covered the settings gear. Moved to the bottom-right corner in 0.10.3

Clone this wiki locally