in push-to-talk, answer a Gemini tool call that arrives with no input, reply
or tool result outstanding with a failure instead of running it, and refuse
every other call in the reply it starts, including one after unprompted
speech; the reply stays tracked so the next input retires it and its late
boundary cannot close the user's turn
in push-to-talk, a caption that arrives after its turn ended no longer opens
a reply, so the next request is not retired with its tool calls discarded
open mic cannot tell either case from a new reply and stays unchanged
check packet ownership after the narration callback before queuing tools,
and raise the per-connection call history to 4096 IDs
move the pointer turn event, context item and announcement into
pointerContext so the OpenAI and Gemini controllers share them, and drop the
Gemini token route's copy of the shared exact-Origin comparison
- A background catalog refresh that rejects now keeps the old catalog for
another TTL, as an empty refresh already did, so a refresh that keeps
failing is retried once per TTL instead of on every request.
- Trim notes are keyed by the source list they describe, and /sources asks
for the list it served, so a request answered during a refresh never pairs
the old cameras with the new list's notes.
- Tests: a failed background refresh (one warning, no unhandled rejection,
no retry before the TTL, trim notes stay with their list) and an empty
background refresh (old catalog kept, no upstream fetch before the TTL).
The race assertion's message now names the failure; its timer is cleared.
- The Tarktee test no longer imports the new constant, so on main it fails
on its own assertion instead of taking the whole file down.
- precompute-cctv-heights waits 60 s for a cold catalog, past Tarktee's 30 s.
- CURRENT-STATE notes the 30 s bound and the retry spacing.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
refreshCctvFocusStyles passes cameraColorFor as baseColorFor, but only
refreshCoverageStyles was tested. The new rendering.test.mjs runs the
focus pass through createRendering with stubbed focus services and
checks that an idle video camera fades from violet, a still from cyan
and the active camera from amber. It fails if the focus callback goes
back to the old active/idle check. Also adds the missing semicolon on
the policy.js import in cctv.test.mjs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Drive drawProjectionFrame with a fake runtime and a recording canvas:
a 640x480 still draws at its fitted rectangle over the matte, the same
pixels at the same size skip the redraw, and the same pixels at a new
aspect redraw at the new fit. The last case fails on the pixel-only
signature; the first fails if drawImage stretches to the canvas again.
Expected rectangles now read the canvas size from policy.js.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The parity test compared getWorldPacket with voice, which now calls
getWorldPacket. Compare both against a fixed payload captured from main's
get_current_view_state before the extraction, key order included, and
cover a non-finite now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A default parameter only covers undefined, so now: null aged every feed
against the epoch and reported stale data as just now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- State FL511's terms the same way everywhere: individual, personal use
only, and any other use (a public deployment included) needs FDOT's
express written consent. .env.example, README, dataCredits, the source
licence string, the catalog comment and the changelog fragment said only
"individual, non-commercial use".
- Page the ArcGIS catalog on exceededTransferLimit alone and advance the
offset by the rows each page returned, so a lower server-side max
record count no longer cuts the catalog short at the first short page.
- Order pages by OBJECTID_1 (this layer's object id; a plain OBJECTID is
rejected with HTTP 400) so offset paging is stable.
- Raise the page ceiling to 40 and warn when it is hit.
- Tests: short pages with exceededTransferLimit keep paging, orderByFields
is sent, and a service that ignores resultOffset stops at the ceiling.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
aircraft_in_area stats summarized altitude over airborne aircraft but
speed over every match, so parked aircraft pulled the mean speed toward
zero. Both figures now use the same airborne subset, and the on-ground
fixture is stationary so the test can tell the difference.
Reserved AIS ship-type codes 100 to 255 now read as unknown, as type 0
does, instead of showing up in the breakdown as their own type.
Adds tests for an area where every aircraft is on the ground and for
find_vessel type and destination normalization, and notes in the
changelog fragment that every vessel answer is now normalized.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Attach the background-refresh warning where the refresh starts, so a
failed refresh logs once instead of once per request served meanwhile.
The test now checks that requests during a refresh share it, and turns
the hang off before releasing the fetches so a retrying loader can't
stall it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An expired catalog made every /api/cctv request (camera list and frames)
wait for the full refresh, which lasts as long as the slowest of the live
packs: Tarktee's DATEX feeds took 2 to 16.6 s across ten requests. With
Tarktee's timeout raised to 30 s, that wait could reach 30 s at each
15-minute refresh. The cached list is now returned while a single-flight
refresh replaces it in the background; only a first load, with nothing
to serve, waits. A background refresh that throws is logged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The Estonian Transpordiamet feeds answered in 2 to 16.6 s across ten
requests (120 and 160 KB), and both must arrive for any camera to load,
so the shared 15 s catalog timeout dropped the whole pack on about a
third of loads ("Tarktee DATEX download error: The operation was aborted
due to timeout"), for the 15 minutes until the next catalog refresh.
TARKTEE_FETCH_TIMEOUT_MS gives the two requests 30 s; other packs keep
15 s.
Live: 175 cameras in 3.3 s.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
AIS sends ship types as ITU-R M.1371 codes, so vessels_in_area grouped
"70", "74" and "79" as three types, and type: "tanker" matched nothing on
feeds that send codes. Codes now read as the family the vessel cards already
show (CARGO, TANKER, TUG), for rows, group_by and the type filter. The code
table moves to src/data/vesselTypes.js, which has no i18n import, so the
server-side tools can use it; vesselLabels re-exports it.
A destination with no letters ("0", "---") now counts as unknown.
Live, Singapore: "most by type: CARGO (3), TANKER (3), PASSENGER (1)"; "how
many tankers near Singapore" found 3.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
ICAO flight-plan Item 7 caps aircraft identification at seven characters, so
the flight number after the designator is one to four characters, not five.
Philippine RP-C1234 and Cuban CU-T1240 registrations read like a designator
plus a flight number once the hyphen is dropped; they now count as unknown.
Cubana's CUB123 still counts as CUB.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- dataCredits `gbfs`: names the seven PBSC operators, with BiciMAD's
required "Powered by EMT de Madrid" linking to emtmadrid.es.
- dataCredits `jcdecaux-gbfs`: adds Valenbisi and Sevici.
- DATA_SOURCES.md: a BiciMAD row (Madrid's open-data catalogue publishes
the feed under EMT's terms, attribution required), one row for the other
six PBSC feeds (no licence published; attribution to the operator), and
the JCDecaux row widened to Valencia and Seville.
- CURRENT-STATE.md: the Bikeshare row, and the underscore and
geofenced-capacity rules in "GBFS upstream bounds".
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Adds the station-based public bike systems of seven Spanish cities on
PBSC and two on JCDecaux, 1,843 stations in all (live feeds, 2026-10-17):
- PBSC (`<city>.publicbikesystem.net`, GBFS v2 like Bicing): BiciMAD
(Madrid), Bizi Zaragoza, BIKI (Valladolid), BiciCoruña, Dbizi
(Donostia-San Sebastián), Bilbao Bizi and BicinRivas (Rivas-Vaciamadrid).
- JCDecaux (`api.cyclocity.fr/contracts/…`, like Lillestrøm): Valenbisi
(Valencia) and Sevici (Seville).
Both hosts are already allowed by the proxy. Every system uses a 40 km
load radius; the farthest station is 11.7 km from its centre (BiciMAD).
The radius also keeps out what two feeds list away from their city:
Sevici's workshop at (0,0) and lab in Madrid, and BiciCoruña's planned
stations with no position yet.
spainCitiesSystems.test.mjs checks the registry and proxy for each feed,
that the outermost live stations activate their system and sit inside
its radius, that central Madrid activates BiciMAD and BicinRivas, that
the far-away stations are not loaded through the real ingestion path,
and that JCDecaux names arrive without underscores and PBSC capacities
as docks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PBSC adds the bikes a station accepts in the geofenced area around its
docks to `capacity`, and reports that part as `geofenced_capacity`, which
is not a GBFS field. Bilbao Bizi publishes 39 docks as capacity 839 with
geofenced_capacity 800, so its stations would read as nearly empty: red,
drawn at maximum size, and carded "39 avail · 0 docks · 839 cap".
BiciMAD (671 of 678 stations, about twice the docks) and Dbizi (11 of 70)
do the same.
The shared parser now reads `capacity − geofenced_capacity` when the
extension is present and no larger than the total, so the layer, its
colour ratio and get_bike_share all see the docks. No registered system
changes: none of the 52 publishes a geofenced capacity above 0 (live
feeds, 2026-10-17, all 13,397 capacities identical).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
JCDecaux's Spanish contracts prefix every station name with "_", and
Valenbisi joins its words with it ("_PLAZA_MUSICO_LOPEZ_CHAVARRI"), so
the card and HUD would show the raw identifier. The shared parser now
reads a run of underscores, with the spaces around it, as one word break
and trims the ends. It applies to `name`, the `short_name` fallback and
GBFS v3 localized names, so the layer and get_bike_share both read it.
No registered system changes: run over all 52 live station_information
feeds on 2026-10-17, the old and new parsers give identical names for
all 13,397 stations. Other spacing stays as the operator published it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
OpenSky, the default civil feed, reports no operator, so group_by: operator
said "no operator reported". aircraft_in_area now also takes
group_by: 'airline', read from the ICAO designator in each callsign: three
letters followed by a flight number that starts with a digit (SWA1234,
UAL12A). Registrations used as callsigns (N123AB) count as unknown. Rows are
unchanged; the designator is derived for the breakdown only.
Live, Texas: "785 aircraft ...; most by airline: AAL (89), SWA (65), UAL (52);
283 unknown."
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Said Turksever