Repository navigation
Roadmap
No dates except those the sky sets (a monthly summary, a month of lists). ZenithBoard is a spare-time project and any other date would be a fiction. What this page promises is an order.
Where it stands (14 November 2026): 1.0.0 is out. The owner closed the release candidate that day, fourteen days into its last month hands-off (assumé), the fresh install on a blank card redone during it; the same code as 0.99.26. The freeze of 25 October 2026 is over. So this page reads, in order: the lots after 1.0, what is being built now; what waits, item by item; what 1.0 took, kept as the record of the release candidate; what is not planned, and why.
Kept as the record of the release candidate. 1.0.0 was published on 14 November 2026 at the owner's word, before the date below: the fresh install on a blank card done, the month hands-off stopped at fourteen days, the fixed address still never applied. What follows is as it stood during the candidate.
1.0 means validated on real hardware, no caveats in Project status (Release process). The state of it on 26 October 2026, the release candidate 0.99.0 running on the reference Pi since 25 October, 13:51 CET:
| Caveat | How it was lifted |
|---|---|
| Feeding from the reference Pi since 20 October 2026 | |
|
|
Running on the reference Pi: its 24-hour and 7-day graphs seen in the owner's exports, 24 October 2026. Only collectd's memory over weeks is still to be watched, which time does by itself |
/stats, never seen on the real sky |
Seen on the real page: 20/10 08:20 · easyJet EZY17QZ · Manchester → Verona · 335 km, near Cergy-Pontoise |
| Done several times on 25 October 2026, on two Pis (the owner) | |
| Marked experimental, outside 1.0 (the owner, 25 October 2026): no US receiver has run it. It stays in, documented as such; a US user's report can bring it in later |
| Caveat | What lifts it |
|---|---|
| The summaries posted by the service (0.33.0 to 0.38.0, 20–21 October 2026) | Daily (25 October, 08:45) and weekly (Saturday 24 October, 08:00, the day the owner chose) seen going out, the four posts each. The monthly one is left: due Sunday 1 November, 09:00 |
| Posts 2 to 4 and the page's lists over a real month | The lists began on 20 October 2026 (0.31.0): a whole month of them around 20 November |
| A fixed IP address never applied for real |
zenithboard net static … on the reference Pi, then back to DHCP, with a keyboard and screen at hand |
| What 0.54.18 to 0.55.2 added | Seen once on the real Pi: the day's aircraft counted every 5 seconds (against FlightAware's count, a whole day after the update), aviation units, the registration's hashtag on a real post, Hindi, the summaries' post 1 link |
-
A fresh install on a blank card, following the wiki's Fresh install alone, step by step, and finished from the web console (the owner's criterion, 21 October 2026: its checklist ticked line by line; a line that needed anything not written there is a fix to the page and a new try). The owner's blank card of 25 October 2026 met three such lines (the signing-key question, Bluesky in the wizard, where to find the position), fixed in 0.54.18: a new try on 0.99.0 is what is left. Since 0.21.1 the installer is also walked end to end against stand-ins at every change (
tests/test_install.sh). -
One month hands-off, the release candidate's month (the owner, 25 October 2026: l'application passe en RC pour un mois afin de valider un passage en v1.0), social sharing on: no crash in
zenithboard logs, no reboot forced by a watchdog. This one cannot start while the Pi is updated several times a day: it starts with the freeze.
Frozen since 25 October 2026, 13:42 (the owner: on gèle à partir de maintenant, on ne passe qu'en correction): 0.99.0 is the release candidate, the same code as 0.55.2. From it only fixes are made (0.99.1, 0.99.2…): no new feature, no new setting, no new screen, no new language. The hands-off weeks are counted from there, and a fix shipped during them restarts nothing (the owner, 29 October 2026: on fait des corrections, pas des nouvelles fonctions). What is asked during the freeze waits for 1.1, written down below in Waiting for 1.1, unless the owner lets it in: 0.99.6 (30 October 2026) brought AirNav Radar and the Telegram messages, and the 30 days started again with it.
What counts as a fix: something that does not do what the docs say, a crash, a wrong figure, a security or privacy hole, a mistranslation, a doc that a novice cannot follow (the fresh-card try will find some), a provider that moved a file. What does not: a new option, a new tag, a new screen or card, a new language, a new look, an improvement to something that works as documented.
The hands-off clock started on 25 October 2026, 13:51 CET, for 25 November 2026, and starts again with 0.99.6 (the owner, 30 October 2026: repart pour 30 jours de RC à la suite): 30 days from the reference Pi's update to 0.99.6, which ran on 30 October 2026 at 22:27 CET (the wall's journal: 0.99.6 listening, 22:27:08): the month ends on 29 November 2026, 22:27. The fixes do not move it (the owner, 29 October 2026: arrête de pousser la date de la RC, on fait des corrections ici, pas des nouvelles fonctions). 0.99.1, 0.99.2 and 0.99.3 were fixes and the Pi took them in its stride; for two days the date had been pushed to 28 and then 29 November, which is withdrawn. What would stop the clock is what the month is there to catch: a crash in the logs, a reboot forced by a watchdog, a service that stops. Those are written in the field log of Project status, and one of them moves the date.
Swahili and Hindi are published without a native review, said as such in Translating: that is not a caveat for 1.0.
Earliest date: the monthly summary on 1 November, a month of lists around 20 November, the release candidate's 30 days hands-off from the update to 0.99.6 (30 October, 22:27): 29 November 2026 at the earliest, with the fresh card and the fixed address done meanwhile. A caveat still open then moves the date, not the bar.
Not caveats for 1.0, but things the month on the Pi answers by itself. Each ends with a field-log row in Project status, and a fix (0.99.x) only if something is wrong.
| Item | Since | How it ends |
|---|---|---|
| The statistics page's cost on a Pi 3B | 0.37.0's look and period toggle (21 October 2026) approved on the real page; its cost measured on a desk machine only (0.14 s for thirty days) | A toggle through 24 h, 7 days and 30 days on the Pi with sudo zenithboard memory before and after |
| The day's aircraft counted every 5 seconds | 0.54.18 (25 October 2026): /stats was 11% under FlightAware |
A whole day after the update compared with FlightAware's Aircraft reported: a few percent apart is right |
| Debian's daily security fixes | Switched on 18 October 2026 from the menu | A few weeks without a feeder disturbed (8 November) |
| What the wall cannot name or draw | Counted since 18 October 2026 (zenithboard stats gaps, 0.21.0) |
Read on 17 November 2026, procedure below; what it decides is built after 1.0 |
| The posts' maps | Since 0.23.0 (18 October 2026): a real map with the track, town names (0.24.4 to 0.27.0), the photographer's link (0.25.0, if on), units of their own (0.24.3; aviation 0.55.0), the registration as a hashtag (0.55.1). Read so far on the real Pi: the normal and terrain maps, the satellite map once | A few posts on each map (normal, night, satellite, terrain) read on a phone: legible names, credits in place, nothing hidden by the legend |
| The gain | Settled at 48.0 dB, 4–5 % strong messages at rush hour | Nothing to do. Recorded so it is not re-opened |
Two items of After 1.0 are to be built from what the reference Pi measures, not from guesses. The figures are read during the release candidate; what they decide is built after 1.0 (a new lookup and new shapes are features, not fixes). Statistics were switched on there on 18 October 2026; the figures are read on 17 November 2026 with zenithboard stats gaps, and this is the procedure, written down so it does not depend on anyone's memory.
What is being counted. With statistics on, every quarter hour the sampler asks the wall which aircraft are inside its radius and notes, once per aircraft over thirty days: the three-letter callsign prefix of each aircraft shown with a callsign but no airline name, and the type code of each aircraft drawn with the generic silhouette. zenithboard stats gaps prints both lists, most frequent first.
Undercounted before 0.99.13 for imperial and aviation units. Until 0.99.12 the sampler sent RADIUS to the wall as kilometres, whatever the units: a wall set to 10 miles (16.1 km) or 10 nautical miles (18.5 km) was sampled over 10 km, a smaller circle, so stats gaps counted fewer aircraft than the wall shows (in area, about 39% of the circle for miles and 29% for nautical miles). A Pi set to kilometres was counted right. The gaps files of before are kept as they are (nothing can say which aircraft were missed), so on such a Pi read the figures as a floor, or read them again thirty days after the update. The 17 November reading needs 0.99.13 on the Pi; the units a Pi is set to: zenithboard config | grep UNITS (no line means kilometres).
2. Airline names for callsigns without a route
- Read the prefix list. A prefix seen on fewer than three aircraft in thirty days is noise. If no prefix passes that bar, the item is closed: the route lookup already names what matters.
- Otherwise, one lookup per prefix, not per flight: the ICAO airline designator (
EZY,BGA) asked once of adsbdb's airline endpoint if it answers by designator, or taken from an open designator list shipped with the wall if it does not; the answer cached for months in the wall's cache folder, like routes. Off by default, a line in the wall settings menu,zenithboard config-set AIRLINE_LOOKUP 1. - Checked against the real answers for the prefixes in the list before it ships, like the route filter was.
3. More silhouettes
- Read the type-code list. The five to ten most frequent codes get a shape each; the rest keep the generic one.
- Each shape is drawn in
flightinfo/static/shapes.jswith the primitives already there, previewed as a picture here, and shipped only once it has been seen — the rule for everything on the wall. -
aircraft.classifymaps the code to the new shape;tests/test_shapes.jsholds the mapping so a code cannot quietly fall back to generic again.
The reading (14 November 2026, taken by the owner as the reading of the 17th; 28 days of samples).
- Silhouettes: closed. Type codes drawn with the generic silhouette: none: every type code seen had its own silhouette. Nothing to draw; the item leaves After 1.0.
-
Airline names: worth building. Forty prefixes named no airline, every one on five aircraft or more, far over the bar of three:
JFA31,VJT19,WMT17,KLC14,GAC12,PGC,LXAandJAF11,PTNandCTM10, then thirty more between 9 and 5 (TUI,TAY,FJO,AAB,VJH,RUK…). Mostly business and charter operators, and airlines' second certificates, whose flights have no published route, which is why the route answer never named them. -
adsbdb answers by designator (asked on 14 November 2026:
api.adsbdb.com/v0/airline/JFA→ Jetfly Aviation,VJT→ Vistajet,WMT→ Wizz Air Malta,TUI→ TUIfly), so step 2 takes its first way: one lookup per prefix, cached for months. Its country is not to be shown without a check: it gives Canada forVJT, VistaJet's Maltese certificate. The name alone is used; the flag stays the registration's. - Memory, the same paste: 496 MB of 905 MB available with the five networks, Bluesky, Telegram, Tailscale and graphs1090 (409 MB in use; the Bluesky service the largest, 93 MB, peak 129 MB): the figure the two networks of Waiting for 1.1 are to be measured against.
If the figures never come (the Pi was off, statistics were switched off), the two items stay in After 1.0 as they are. Nothing is built on a guess.
The order in which what waits below would be built once 1.0 is out (drawn up on 14 November 2026 at the owner's request: be ready to add features of value). Each feature is a minor version, so a lot is a release. The owner took it the same day (on passe à la 1.1, lot par lot), with the airline names moved into lot 1, and agreed to what is put off after the lots and to what is not planned.
| Lot | What | Why there |
|---|---|---|
| 0. The base (1.1, first) | One table of rules for every value (lib/rules.tsv, the wall's, the receiver's and the sharing settings); units.py and static/units.js, one table test); StatsWindow); flightinfo/saved_state.py); tests/test_pages_smoke.js, the guard for the rest); |
Every lot after it adds settings, units or state. The checks exist in two or three copies that already disagree, the units in five; the two wrong readings of /stats came from windows worked out in several places. Built once here, not paid at each feature |
| 1. Sharing and the wall, the visible gains (released as 1.1.0, 14 November 2026) |
AIRLINE_LOOKUP; the reading of 14 November: forty prefixes, JFA on 31 aircraft); |
Seen every day, small, no new outside service but adsbdb, already asked for the routes. The distance needs lot 0's state helper |
| 2. The wall and the receiver (released as 1.2.0, 15 November 2026) |
|
The data is already there (the squawks heard, thirty days of statistics). Nothing tells today of an antenna or a dongle going weak. MLAT has been fixed on ADSB Exchange since the first release; it touches only the five networks already fed, so it does not wait for the new ones |
| 3. Measured first (1.3) | The watch list of remarkable aircraft | Worth it only if the antenna hears enough of the list: counted before it is built |
| 4. The networks (1.4) | adsb.lol and OpenSky (adsb.lol's MLAT client one more value of lot 2's setting); the Networks block on /stats
|
The least value for the most risk: two more services, installers never run on the reference Pi, terms to read. Memory is not the obstacle (496 MB of 905 MB available on 14 November) |
After these lots, each alone: a setting without a restart (a change of the wall's design); HTTPS for the console; coverage by direction. Slipped in when their file is open: the places index, the console's refusal with its reason, Arabic's chips, the root boundary's test.
What is asked during the freeze, kept here so that nothing is lost, and done after 1.0:
-
Receiver health by trend (asked by the owner, 26 October 2026; built on 15 November 2026, released in 1.2.0:
TELEGRAM_HEALTH,zenithboard telegram health on|off|check, Telegram. Not as written below: the owner's thirty days showed that the rate against the same hour a week earlier cannot be the rule, the messages having fallen by 36 % in a month with the same receiver; three measured rules took its place, deaf, a drop by day, a range that wears away, agreed by the owner). The statistics already keep thirty days of messages per hour. An alert when the rate falls well below the same hour one week earlier would catch a dongle losing strength, a degraded antenna or a cut cable, before a receiver goes quiet for days: the most useful alert for a Pi that runs without anyone watching it. It depends on what the sky looks like, so the threshold waits for the counters to show a normal week (zenithboard stats gapsand the field log), not a guess. Off by default, by command and by menu; the Telegram chat of 0.99.6 is where it would be said. -
Emergency squawk banner on the wall (asked by the owner, 26 October 2026; built on 14 November 2026, released in 1.2.0; its first form seen by the owner on the real wall, reworked the same evening to say which aircraft and where:
SQUAWK_BANNER,zenithboard squawk-banner, the gear's Emergency squawks, The wall). When 7500, 7600 or 7700 is heard anywhere in the antenna's range, not only inside the wall's radius, the wall shows a banner for 30 seconds, longer than its usual time for an aircraft, with the code and its meaning; a checkbox in the wall's settings (the gear), off by default, and a command. Nothing published: an emergency is not a fact to broadcast. The detection is 0.99.6's (flightinfo/telegram.py, heard on two reads in a row); to be previewed on the real wall before it ships. -
Choosing which network does MLAT (asked by the owner, 14 November 2026: not blocked on ADSB Exchange for ever; built the same day, released in 1.2.0; moved to FlightAware and back on the reference Pi the same evening, its MLAT column after a day still to read:
MLAT_NETWORK,zenithboard mlat, Providers. Two things differ from what is written below, after reading each network's own documentation:flightradar24is not a choice, Flightradar24 asking for its MLAT to be off on a receiver that shares with other networks; and removing the network that holds MLAT gives it back to ADSB Exchange and says so, rather than asking). Today the single MLAT client is ADSB Exchange's, written into the code: every install and update, andzenithboard mlat-guard, switch MLAT off in FlightAware (allow-mlat no), Flightradar24 (mlat no) and AirNav Radar (autostart_mlat false). 1.1 makes it a setting,MLAT_NETWORK:adsbx(the default, so nothing changes on a Pi that does not touch it),flightaware,flightradar24,airnav, ornone; still one client at a time, which is what a 1 GB Pi carries (Plane Finder has no MLAT client to choose). What depends on it, all in the same release:-
The command:
sudo zenithboard mlat <network|none>andzenithboard mlat status;mlat-guardkeeps its name and enforces the chosen one instead of ADSB Exchange. -
The guard, the installs and the update: each network's installer and
zenithboard updateread the setting instead of switching the others off; choosing a network that is not installed is refused with the reason; removing the network that holds MLAT says that none is left and asks which takes over. -
Each network's own switch, and back: FlightAware
piaware-config allow-mlat yeswithmlat-results, Flightradar24mlat yes(its feeder decides by itself whether the station qualifies), AirNavautostart_mlat true, ADSB Exchange'sadsbexchange-mlatservice stopped and disabled when it is not the one; every change put back by choosing the former network. -
The menu (1 ADS-B): a line MLAT: which network, with the installed networks only. The web console: the same choice in Receiver, beside One MLAT client only, in the 22 languages.
zenithboard statusnames the network that holds MLAT; today's line off (ADSB Exchange keeps the only one) follows the setting. - Backup and import carry the setting; the docs: Providers, Command reference, Configuration reference, Fresh install.
-
Before it ships, on the reference Pi: MLAT moved to FlightAware and back,
zenithboard statusand FlightAware's page (its MLAT column, 0 since 0.99.6) read after a day; memory with each client (adsbexchange-mlatis 10 MB today). -
In lot 2 (the owner, 14 November 2026), so before the two networks below, whose entry already leaves MLAT to the owner: adsb.lol's
adsblol-mlatbecomes one more value of the same setting.
-
The command:
-
Two more networks to feed: adsb.lol and OpenSky (asked by the owner, 1 November 2026). Each its own switch in the menu and by command, off by default, switched on one at a time, and removable as the five networks are; the antenna's position stays what the feeders already get. Both take what readsb already decodes (its Beast output, port 30005), so nothing changes in the decoder. MLAT stays the owner's choice for each network: off by default, since ADSB Exchange keeps the only MLAT client today, and the MLAT guard would leave a client the owner switched on rather than stop it. What each would bring, read on 1 November 2026:
-
adsb.lol: its script (
adsb.lol/feed.sh) installs two services,adsblol-feedandadsblol-mlat, and needs no account; the data is open (ODbL 1.0). Its page documents no MLAT switch of its own: ZenithBoard would leaveadsblol-mlatdisabled unless asked, and say it. -
OpenSky Network: the
opensky-feederpackage from OpenSky's own signed APT repository, beside readsb; no MLAT client in it. It needs an OpenSky account and a receiver serial, which the owner types in the terminal (a session never creates the account). Its terms give OpenSky a perpetual licence to the data fed and allow its use for non-profit research and education: written beside the switch, so that whoever turns it on knows. - Before it ships:
zenithboard memoryon the reference Pi with each one on (436 MB of 905 MB free today with the five networks, the wall, the posts and Telegram),zenithboard statuslisting them, the menu and the console in the 22 languages, and the pages Providers and Fresh install.
-
adsb.lol: its script (
-
Each network's ranking of the station on
/stats(asked by the owner, 9 November 2026, with FlightAware's Site ranking over 30 days: rank 4786, 76,648 Mode-S aircraft, 6,608,812 positions, MLAT 0). Once a day at most, off by default, the rank, X out of N when the network says N, and the figures the rank is built on (aircraft, positions, with and without MLAT). Each network's own pages and terms, read on 9 November 2026, decide what can be fetched:Network Its ranking Can the Pi fetch it once a day? Out of how many FlightAware flightaware.com/adsb/stats/user/<user>: rank over 30 days, ADS-B, MLAT, UAT and Other aircraft and positions (Other no longer counts toward the rank since April 2025)No. Its terms, item 07: the site only with a human-operated interactive web browser and not with any program… including… "scraping", FlightAware's paid feeds and APIs excepted. No JSON for the rank Not given Flightradar24 My data sharing → Show statistics, behind the owner's sign-in; 30 days, refreshed every 12 hours No. Behind a sign-in, which a session never uses and the Pi would have to keep Not given ADS-B Exchange A public leaderboard, calculations based on 30 days, updated daily, filter ?feed=<name>No, as things stand. The table is built in the browser from /api/…, which its robots.txt disallows, and which answers only with a borrowedRefererheader: getting round that is not doneNot shown Plane Finder Public lists by country, planefinder.net/coverage/receivers/<country>, ranked by the number of position reportsYes, plain HTML, not disallowed by its robots.txt (its terms page could not be read: to read before building). Only if the receiver is listed under a name ZenithBoard can find (a private receiver shows as Anonymous); the columns are not labelled, to be confirmed against the owner's own receiver page Yes, within the country (103 receivers for France on 9 November 2026) AirNav Radar Public station page, airnavradar.com/stations/<serial>: global rank, country rank, score, rangePossibly, plain HTML: its robots.txt allows it, but its llms.txt asks programs to use its (paid) API rather than its pages. To ask AirNav before building Not given So 1.1 would bring, if anything: a Networks block on
/statswith, for each network fed, a link to its ranking page (opened by the person, in a browser, which every network's terms allow), the station's own counts beside it from what ZenithBoard already counts (aircraft per day, positions, MLAT), and the live counters the feeders give on the Pi (fr24feed'smonitor.jsonon port 8754: aircraft tracked, ADS-B and not, MLAT; piaware's status file). A rank fetched automatically only where a network allows it: Plane Finder's country list (X of N in France) and, if AirNav agrees, its station page; each one a switch of its own, once a day at a random minute, the page parsed with care and the block saying not found rather than a wrong figure when the page changes. FlightAware, Flightradar24 and ADS-B Exchange stay a link. Before building: the owner's real pages on each network, to match the figures; the terms of Plane Finder and AirNav read in full. -
The old command names go (done in 1.1.0, 14 November 2026: each is refused with the name to use; 0.99.25 kept them working with a line saying so):
gain check,color,show,style,threads,presets,preset rm|forget,social weekly,social tags|color|sentences|templates,telegram templates. Command reference → Names removed in 1.1 lists them. -
The aircraft's photo in the Telegram messages: dropped (asked by the owner on 31 October 2026, dropped by him the same day). He wanted the photo of the very airframe passing overhead, found by its registration, in each crossing's and squawk's message, with its credit, instead of a link. The only source of that photo is Planespotters, and its photo API terms (read 19 October 2026, Routes, photos and data) allow a picture only loaded straight from the API's address by the screen that shows it, never stored, proxied or passed on; Telegram's
sendPhotostores the picture on Telegram's servers. A photo of the type instead of the airframe (Wikimedia Commons) is not what was asked, so nothing replaces it. The terms could not be read again on 31 October 2026 (the page answered 403 to an automated reader); read again on 1 November 2026, unchanged on this point: image binaries must not be downloaded, stored, or re-hosted on your infrastructure under any circumstance. It would come back only with Planespotters' written permission. -
A watch list of remarkable aircraft (asked by the owner, 1 November 2026, after reading Planefence). plane-alert-db lists aircraft by their ICAO address with an operator and categories (government, police, air forces, remarkable airframes…), under the ODbL 1.0. ZenithBoard would download it to the Pi only, as it does tar1090-db's list of types, credited where it is used, and read only the address, the operator and the categories: never the image columns, whose links point at JetPhotos' and other sites' pictures, which neither the database's licence nor those sites let anyone pass on. A new rule for the posts and the Telegram crossings,
watchlist, besidemilitary,rareand the others, with the categories to keep chosen in the console; the operator's name could stand in the post when the airline is unknown. The person the aircraft is listed for, when the list names one (asked by the owner the same evening): the list gives it for many state aircraft (Gov of Morocco, Mohammed VI), but in free tag columns where a name sits beside a country, a nickname or a joke, and not always right (read on 1 November 2026: an aircraft of the Republic of the Congo listed for the president of the other Congo, presidents who left office years ago). So: the name only from a state aircraft (the list'sGovclass), only when the tag is a name ZenithBoard keeps in a small reviewed table, not every tag; said as the list's (listed for …), never as a fact about who is on board; never the list's editorial tags (Dictator Alert, President4Life and the like); in the Telegram messages first, and in the public Bluesky posts only by its own switch, off. A private person's aircraft (the list's celebrities and business jets) is never named in a post: a public post saying where a named private person is now is not something ZenithBoard does. Off by default, by command and by menu, the console in the 22 languages. Before it is built: how many aircraft of the list the reference Pi actually hears in a month (zenithboard stats gapscould count them from the seen files), so the rule is not built for nothing. -
The heading in the posts and the messages, by default (asked by the owner, 1 November 2026; built on 14 November 2026, waits for 1.1.0: heading south-east, cap sud-est, in the
factualandboardposts and both Telegram messages, the word chosen by the owner the same day over a first draft's arrow, which stays as the{arrow}tag; the phrase is in the posts' catalogue in every language, without a native review; the three sentence shapes are left as written; the heading is left out whenever its room would cost anything else, held by a test over every language). Where the aircraft is heading (heading south-east) in the default text of the Bluesky posts and the Telegram crossings, in the reader's language. The{heading}field already exists for one's own templates (from the aircraft's track, never from the antenna); 1.1 puts it in the default text. Everything must still fit: Bluesky's 300 characters with the text, the hashtags and the map's credit; when it does not, the heading is the first thing left out, before any other part of the text, as the photo's line is today. A test for the longest airline, model, route and town in every language. -
The distance from the antenna in the Telegram messages (asked by the owner, 1 November 2026; built on 14 November 2026, waits for 1.1.0:
TELEGRAM_CLOSEST, off by default, the first message edited at the exit, Telegram. One thing below was wrong and is corrected there: Telegram states no 48-hour limit for a bot's own messages, the limit is for business messages a bot did not send). The closest the aircraft came to the antenna during the crossing, in the units chosen for the messages (km, mi or NM), in the crossing's message: closest 2.4 km. The Telegram messages only, never the Bluesky posts, the wall's pictures or anything public. It changes 0.99.6's rule that the messages never give the distance from the antenna (Telegram): the chat is the owner's, but a group's members would learn it too, so it is a setting of its own, off by default, said as such in the console and the menu. The closest point is known only once the aircraft has left the radius, and the message goes out soon after it enters (0.99.6). The ways weighed on 1 November 2026:- Wait for the exit, one message. The look up, it is overhead moment is lost: the message comes one to two minutes late for an airliner across a 10 km radius, ten or more for a helicopter or a light aircraft circling, and never for one holding inside the radius without a time limit. An aircraft lost from reception inside the radius (low, behind a hill) ends only on a timeout. The crossing must be kept across a restart, or its message is lost.
-
A second message at the exit. The immediate message stays, but each crossing makes two notifications on the phone, twice the messages against Telegram's limits (twenty a minute in a group; the outbox already paces them), and a second line read without its first in a busy chat. Sending it silently (
disable_notification) spares the phone, not the chat. -
The first message edited at the exit (the one kept). The message goes out at the entry as today; at the exit the bot edits its own message (
editMessageText; no time limit is stated for a bot's own messages, read on 14 November 2026) to add closest 2.4 km. One message, one notification (an edit does not ring), nothing late. What it costs: the message's id kept with the crossing (in the state saved after each send, so a restart does not lose it), one more call to Telegram per crossing, and an edit that fails (the message deleted by the owner) is logged once and dropped, the message staying as it was. - A distance predicted at the entry, from the track and the speed: no wait and no edit, but an estimate, wrong for an aircraft that turns or climbs away; not the real distance that was asked for, so not taken.
-
Aircraft never to post (asked by the owner, 1 November 2026, the same reading; built on 14 November 2026, waits for 1.1.0:
SOCIAL_IGNORE,zenithboard social ignore, the menu under Words and rules, one row in the console). The regulars that fill the feed (a school's aircraft, the police helicopter over its own base) left out by registration, ICAO address or callsign:sudo zenithboard social ignore add F-GXXX,remove,list, in the menu and the console. It applies to the Bluesky posts and the Telegram crossings, never to the emergency squawks, which are sent whoever squawks; the wall still shows every aircraft, being a view of the sky and not a feed. Empty by default. -
From the review of 0.99.8 (30–31 October 2026), found while fixing and left for after 1.0 because each is a change of design, not a fix:
-
A setting without a restart (A2b). Today a wall setting restarts the wall service so that the tablets reload (Architecture); a restart loses the aircraft seen since the last quarter-hour sample and sends a burst of lookups to adsbdb, hexdb and Planespotters.
/api/planeswould carryconfig.env's modification time besideserver_id, the wall would reload when it changes, and the wall's commands would only write the setting; a restart stays for a code update,PORTandBIND_ADDRESS. Until then, an option: the restart takes the statistics sample first when statistics are on. -
One validator (A3b). Each value is checked two or three times (the command, the console, the presets, the import), and the copies already disagree: the console's hashtag rule is not
social_hashtags_clean's (100 characters, a letter in each word),summary-at's format is not tied to the period. Either the console asks the command (zenithboard <cmd> --check VALUE) or both come from one table. With it, ADSB Exchange's, FlightAware's, Flightradar24's and Plane Finder's logs in the console's list (new labels, so the 22 languages). -
The towns near a point from an index (A4).
_towns_nearreads the whole list of places at each call; it would be loaded once into one-degree squares, the loader shared by the wall's server, the Telegram service and the split-flap picture. -
Dead code (A5; removed on 14 November 2026, but
local_types, kept with its test until the lots have shown whether one of them calls it, and the ACARS step, which waits for the owner's word):net_default_maskandwifi_default_route_devin the shell,local_typesin the server (unless 1.1 calls it),social_set_weeklyonce its test runszenithboard social weekly on, the two pictures nothing shows (docs/images/console-devices.png,docs/images/stats-fr.png). The step that keeps an older install's ACARS units goes only once the owner confirms no Pi still has them. -
A page-error smoke test (follow-up of T1):
/,/statsand/consoleloaded in headless Chromium on the test server, failing on any page error. Playwright is in CI since 0.99.18. -
The wall's page refused by name too (W2's other way, not taken in 0.99.17):
/and/static/*answer 403 to a name that is not the Pi's, as the wall's data does, so the browser shows the refusal and itsCONSOLE_HOSTScommand at once instead of the wall's WRONG NAME. -
The console's refusal with its reason: the Pi already sends
{error, why}for an action refused here; the console could show Not allowed here: {why} (a new string, so the 22 languages). - Arabic's quick-glance chips: Bluesky and its state read :Bluesky إيقاف, the colon on the wrong side, right to left.
-
A setting without a restart (A2b). Today a wall setting restarts the wall service so that the tablets reload (Architecture); a restart loses the aircraft seen since the last quarter-hour sample and sends a burst of lookups to adsbdb, hexdb and Planespotters.
-
A leaner architecture for 1.1, by value, from the same review:
- One table of rules for every value (key, type, range, choices), read by the command, the console, the presets and the import, so that a check fixed in one place cannot be missed in another (A3b). First part done on 14 November 2026:
lib/rules.tsv, read by the shell (rule_ok) and by Python (flightinfo/rules.py), for the wall's and the receiver's settings; the altitude's two ranges and the radius's three rules became one each. Second part done the same day: the sharing radius, gap, look, map, colour and rules. The hashtags, the place name and the templates keep their own functions (more than a list or a range), and the languages their own list, held together bytests/test_i18n.sh. - Reload the setting instead of restarting the service (A2b).
-
One unit-conversion module per language: done on 14 November 2026:flightinfo/units.pyfor the server, the posts, their pictures and the messages;flightinfo/static/units.jsfor the wall and the statistics;tests/units_table.jsonholds both to the same figures, and a factor written anywhere else fails the tests. - One saved on change state helper for the two posting services, after each send and on SIGTERM, so that once per crossing holds through any restart, in one place.
- The boundary between root and the
zenithboarduser written in Architecture as a rule (0.99.11 did most of it), with a test that walks every path root opens. -
One window helper for the statistics: done on 14 November 2026 (StatsWindowinserver.py): one class decides the slot for/stats' bars and boxes, the farthest aircraft's line, the sky by hour and the day files (S5). - The places index of A4, loaded once and shared.
- The smoke test above, beside the docs' coverage check (every test file named on Testing, 0.99.21).
-
The dead code of A5 removed(14 November 2026;local_typesand the ACARS step kept, see above).
- One table of rules for every value (key, type, range, choices), read by the command, the console, the presets and the import, so that a check fixed in one place cannot be missed in another (A3b). First part done on 14 November 2026:
What was planned before the freeze and is not a fix, in the order it would come. Nothing here starts before 1.0 is out; then Waiting for 1.1 first.
-
Airline names for callsigns without a route: built on 14 November 2026 (
AIRLINE_LOOKUP, off by default, Routes, photos and data; waits for 1.1.0 and for its first run on the real Pi). Kept in memory for thirty days, not in a cache folder as first written here: the wall server writes nowhere. The forty prefixes were checked against adsbdb's real answers (sixteen on the day, the other twenty-four from the owner's list the same evening, 1.1.1): thirty-one right, three wrong and now in the built-in list (JAF,PTN,OCN), two never asked because adsbdb names another company (OOM,ECA), four unknown to it (KMM,LUA,HBJ,FJD). Decided on the reading of 14 November 2026 (forty prefixes on five aircraft or more; adsbdb answers by designator; above). More silhouettes: closed by the same reading (no type code drawn with the generic shape). Decided on the figures (procedure above). Since 0.17.0 every airline with a known route is already named from the route answer, so the lookup is worth it only if bare callsigns are still common on a real wall. -
Coverage by direction on
/stats(which side of the sky the antenna hears best), once there are weeks of real data, on the page alone: the posts never give a direction. -
The web console, further: HTTPS, which passkeys and Face ID need; and the post pictures previewed inside the page rather than in a new tab. A Pi on a home network can have a certificate browsers trust in two ways (the owner's choice of 26 October 2026: both written down, Tailscale first):
-
Tailscale, already in ZenithBoard (
zenithboard tailscale): with MagicDNS and HTTPS switched on in Tailscale's admin console,tailscale cert(ortailscale serve) gives the Pi a real Let's Encrypt certificate for a name likezenithboard.your-tailnet.ts.net, nothing to install on the phones. Limits: only the devices on the tailnet reach it, each with the Tailscale app; the name shows the tailnet's. -
Let's Encrypt by DNS, with the DuckDNS name: the DNS-01 challenge proves the name with a TXT record instead of a connection from outside; DuckDNS sets it through its API, and
acme.shdoes it by itself (dns_duckdns). The name can point at the Pi's home address (192.168.x.x): nothing is opened to the internet. Limits: some home routers refuse a DNS answer that points at a private address (their DNS-rebinding protection) and need an exception for the name; the certificate lasts 90 days, so the Pi renews it on a timer.
Not considered: a self-signed certificate or a local authority (
mkcert), which every phone and computer would have to install by hand, and with which passkeys stay unreliable. -
Tailscale, already in ZenithBoard (
-
978 MHz UAT validated by someone in the United States: experimental, outside 1.0, since 25 October 2026 (written and tested, never run for real).
-
More languages: Tamil and Urdu (asked about on 24 October 2026) only with a native reviewer: 90% could not be promised without one; Hindi came without (0.55.0, the owner's request). Hausa and Yoruba likewise (left out in 0.48.0). Swahili and Hindi wait for their native review (Translating).
Kept so the question is not asked twice. Newest first.
-
Telegram alerts, from the Pi— built in 0.99.6 (30 October 2026), let in during the release candidate by the owner: each crossing of the sharing radius and each emergency squawk (7500, 7600, 7700) anywhere in range, once per flight, to the owner's own chat or group; never to Bluesky. The token in the terminal only, the finer settings in the console, every language. See Telegram. -
A decision on AirNav RadarBox— built in 0.99.6 (30 October 2026) at the owner's word, with its MLAT off (autostart_mlat=false): ADSB Exchange keeps the only MLAT client, so it does not fight the rule. See Providers. -
Posts, small things: a lower threshold of town names for a sparse region; the farthest aircraft named for the whole month— settled 26 October 2026: the region's own list (sudo zenithboard social places on, 0.27.0, 20 October) names towns anywhere, ranked by GeoNames' place type rather than a population threshold; the farthest aircraft is named every day since 0.22.0 (19 October), so the unnamed days leave the thirty-day page around 18 November. -
The web console,— built in 0.39.0 (21 October 2026) at the owner's request, off by default and removable; computers signed in by the phone in 0.40.0 (21 October); an update from the page and a phone sign-in done on two Pis (25 October). See Web console. Never through it, by design: Wi-Fi and fixed-address changes, the uninstall, the Bluesky app password./console -
Translations— the web console,/stats, graphs1090's page and the posts: French in 0.47.0 (21 October 2026), twenty languages in 0.48.0 (22 October), Swahili in 0.53.1 (22 October), Hindi in 0.55.0 (25 October); twenty-two in all, each with its review sheet (Translating, Translation plan). -
Enrich the statistics page— the sky by hour, airlines, models and altitudes (0.31.0, 20 October 2026); cargo, business jets, military and general aviation with flags, ZenithBoard's own look and a 24 h / 7 days / 30 days toggle (0.37.0, 21 October). -
graphs1090— optional since 0.32.0 (20 October 2026), running on the reference Pi (seen 24 October). -
Try Plane Finder— feeding from the reference Pi since 20 October 2026 (Hardware). -
Photos as Planespotters' terms ask— 0.24.5 (19 October 2026): the tablet loads each picture itself, the Pi keeps none. JetPhotos is not used (no API; personal use only). -
Posts with a real map, flags whole, no dark blue— 0.23.0 (18 October 2026),social look dotmap|flap|alternate; the dots-only look left in 0.24.0. -
Wi-Fi hardening— 0.21.0 (18 October 2026):zenithboard wifi harden on, off by default, reversible. -
Per-tablet presets— 0.21.0 (18 October 2026):zenithboard preset add kitchen theme=green mode=flap, then?p=kitchenon that tablet.
Saying no here saves answering the same question twice.
| Not planned | Why |
|---|---|
| Photos from AirTeamImages (asked 30 October 2026) | A paid stock library (about 323 000 pictures). Its terms, read on 30 October 2026, allow low-resolution downloads for trial layouts only and forbid all other reproduction outside a paid usage agreement (personal prints, editorial licences); there is no public API. Showing its pictures on the wall or in the Bluesky posts would need a licence for each use. Planespotters stays the one source: a free API whose terms allow exactly what ZenithBoard does (the picture fetched from its own address, credited and linked) |
| Airline and aircraft-maker logos | They are trademarks, and redrawing one in dots is still the trademark. A free project cannot hand them out, and a wall of 90 logos is a licensing problem, not a feature. Flags stand in for them: the registration country next to the registration (since 0.20.0), the maker's country next to the model |
| Sound on the wall | An airport-hall ambience was tried (0.13.0) and removed an hour later (0.13.1): it did not sound good enough to keep. Nothing nobody has heard is shipped |
| A cloud account, a hosted dashboard, telemetry | The Pi is yours. Nothing should need to phone home for the wall to work. |
| Turning off package signature checking | It has been the quick fix for two different installer problems. It stays off the table: a signature that is not checked is not a signature. |
| Supporting hardware other than a Raspberry Pi | It probably works on other Debian machines; it will not be tested there, so it will not be claimed. |
| A route source that does not need the internet | There is no free, complete and current database of routes by callsign: the public ones are partial, stale, or under licences a GPL project cannot carry, and a stale route shown with confidence is worse than none. Callsign lookups stay online (adsbdb, hexdb as a second opinion), cached, and off when the Pi is |
| A photo in the Bluesky posts | Planespotters' photo API terms forbid storing, re-hosting or passing on a picture, and a link card would upload its thumbnail. A link to the photo page with the photographer named is what is allowed (social photolink on, 0.25.0). JetPhotos has no API and grants personal, non-commercial use of the site only |
| ACARS and Grafana | Removed in 0.30.0: never run on the reference Pi, about 900 MB more than a 1 GB board has, and what Grafana would have shown is on the wall's own statistics page, with graphs1090 as the long-term option (0.32.0). A Pi that had them keeps them until sudo zenithboard acars remove
|
| A paid tier | It is GPL and it stays free. There is a beer link in the README and that is the whole business model. |
| MQTT or an RSS feed of the crossings (looked at 1 November 2026, from Planefence) | What they would carry is already there: the crossings go to Telegram as they happen (a push to the phone, which is what most home-automation recipes use MQTT for), the history is on /stats and in the console, the live sky on the wall and tar1090's map, and Bluesky is itself a public feed. MQTT would also need a broker (Mosquitto) on a 1 GB Pi, or one elsewhere to set up. Comes back if a user shows a use the others do not cover |
| Other feeding networks (the list of the ADS-B Feeder Image's homepage, looked at on 1 November 2026) | Seven networks at most: the five fed today and the two of Waiting for 1.1. Each network is one more service on a 1 GB Pi, one more installer to follow when it changes, one more set of terms. Each of the others, and why it is left out, is just below |
The other networks, one by one (read on their own pages on 1 November 2026):
- Flightradar24, FlightAware, RadarBox (AirNav Radar), Plane Finder, ADSB Exchange: already fed (Providers).
- adsb.fi, airplanes.live: the same kind of open community network as adsb.lol, each with its own feed and MLAT services; one is enough, and adsb.lol is the one asked for.
- Fly Italy ADSB: Docker only (an ultrafeeder container); ZenithBoard installs no Docker on a 1 GB Pi.
- AVDelphi: a feed must be claimed with a membership on the site, and an unclaimed one is refused; a session never creates accounts, and the site offers services built on the data.
- Planespotters' feed (not its photos, which the wall uses): its installer could not be read before running it, and its page says nothing of what is done with the data. Nothing is run blind.
- TheAirTraffic: its terms let it pass the data fed on through a paid API.
- RadarPlane: its address sent the browser to an unrelated site on 1 November 2026, and the Feeder Image's current list no longer has it.
- HPRadar: its site did not answer on 1 November 2026; nothing could be read.
- Plane.watch: a key from its own sign-up site first, and its pages say nothing of what is done with the data.
- ADSBHub: an account and a station key, and no sign of recent activity on its pages.
- RadarVirtuel: an e-mail address as the station's identity and a script of its own; not one of the two asked for.
- 1090MHz UK: a network for the United Kingdom; the reference Pi is in France.
Open an issue. During the freeze a new idea goes to Waiting for 1.1; after 1.0, an item moves up because someone wants it, not because it has waited long enough.
Direction
Getting started
How it works
- Architecture
- Configuration reference
- Command reference
- The wall
- Statistics
- Social sharing
- Telegram
- Web console
Running it
Working on it