Live trains from 30 keyless operator feeds: GPS positions from 29 and
Irish Rail's station reports, shown as the operators publish them. 28
work out of the box; Renfe long and medium distance and KTMB appear once
their pack is built, because their vehicles name no route.
Packs are opt-in (scripts/build-rail-paths.mjs): the operator's own
route lines, stop names, next stops and trip-update delays for live
trains. Nothing downloads when the app starts, no train is placed from a
timetable, and no request carries a deployer's key.
- The look-at longitude wraps fully, so a tilted camera at a pole gives a
point on the globe. The strict off-globe check on street_level is undone:
points are clamped again, as before.
- A posted view that fails is answered, and the views behind it still apply.
- A posted view waits at most 2 s for the terrain under its camera, then
keeps its own estimate.
- A camera taken during that wait stays where it is: the rest of the view
applies, and only its photos stand down.
- show_street_level keeps a search or a press on the globe made during its
own flight: it adopts only its own navigation stamps on landing, and a
gesture is never forgiven.
- Releasing a held viewer frees the old source's adapter without resetting
the panel, so a replacement's error, outcome and armed OPEN survive; a
cross-source replacement drops the old photo's details and marker.
- The action holds the open photo only for its own switch-off: a source
switched by hand during the load stops it and reports cancelled.
- A posted view waiting on terrain watches the camera and stands down after a
navigation or a globe press; a flat or straight-down camera asks no terrain.
- An arming link opens the panel without persisting the layout, and a posted
view's street-level step runs as programmatic.
- An off-globe street_level point throws TypeError.
- zh-Hant uses the file's wording for the new strings; README and
CURRENT-STATE count 35 tools.
A provider that threw while geocoding (offline, an upstream down) escaped
the action as a bare error. show_street_level now reports it as its
location stage ("Could not arrive at …", the cause in detail), and nothing
is switched or looked up; fly_to_location itself is unchanged.
The voice card showed English for show_street_level failures the catalog
did not know: a missing key, a map or source that could not be switched on,
the layer, invalid coordinates, an imagery fallback and the filter clause.
They now translate like the other action errors (exact messages and
patterns in speech.js), in all twelve languages; the spoken line stays
English, and the filter clause names the filters as their chips do.
AIS Vessels are keyed by the id their live source gives, with the MMSI as an
attribute: an MMSI can be shared, spoofed or reused, so two vessels with one
MMSI are now two vessels. Looking a vessel up or selecting it by MMSI still
finds it.
A share link made while a vessel is selected opens that vessel. A UUID vessel
id travels as 16 bytes in a new thing field (13). A shared thing that is not
reporting now says so in a sentence per kind, with its id (ICAO, NORAD, MMSI
or vessel UUID) on a second line.
The radio and push-to-talk helpers shared by both voice providers now pass
English catalog text, and the OpenAI controller words its detail when shown,
like the Gemini session. A token failure without a server message carries
its status as sourceParams, so it is worded again with the number. The docs
say sourceDetail is the message as given, catalog text when it is the app's
own.
The Gemini session translated each status and error message once, when it
happened, so after a language change the voice control kept showing it in
the old language. The state event now also carries the English catalog text
as sourceDetail, and the control words it when shown. The local server's
known token errors stay catalog text until then instead of being translated
on arrival.
The Gemini token route now answers a missing key with code API_KEY_MISSING,
the session carries it to the error state as missingKey, and the voice
control hides "Check microphone permission" for that error, as it already
does for OpenAI. The OpenAI controller sends the same flag, so the shared
control and the controller agree whichever updates last.
The voice error was matched against the translated key message to decide
whether to show the microphone hint, so switching language brought the hint
back and left the message in the old language. The error now carries a
missing-key flag into the status, which words the message at render time.
A controller test covers the language switch, and CURRENT-STATE.md documents
the keyless voice response.
The decision no longer reports not-entity-context or channel-closed, which
its only caller rules out first, nor pointer-scale-not-useful, which the
crop path already checks and logs. A closed channel returns before any
record, as on main. The crop decision is pointer-crop-eligible, since the
crop checks after it decide whether the pointer is on ground, and a sender
called on a closed channel logs channel-lost.
- layerStateTokenLedger.test: the retired-token reuse case now offers the
retired token (02) to the newcomer, not News's 00.
- he.json: the catalog's singular imperative, {count} first in the plural,
Wikidata read as one source, and smoother legend and info text.
- layerManifest: News after Shipwrecks, so the prompt-budget comment stays
next to the entries it explains.
- CURRENT-STATE: Prettier-formatted (table column, blank line before a heading).
Main added the News layer at the same places: both layers are kept in the
layer names, panel order (News stays last of Events), standalone sources,
voice manifest, credits, DATA_SOURCES and the CURRENT-STATE table. News took
share-link token 00 first, so Shipwrecks moves to 01 (ledger, registry,
CURRENT-STATE and the token tests). The catalog now has 40 layers and the
share-link registry 38. Shipwreck strings are added to the new Hebrew
catalog; the other catalogs keep both sides' strings.
- Comments in the action, the layer, embed, the Mapillary policy and the
controls say why in fewer words; restatements of the code are gone.
- CURRENT-STATE's "Show me X on Street View" section is a list by topic and
no longer says the browser gate aims at coverageLines.
- TOOLS.md's street-level view paragraph is shorter.
- A link's photo step compares with the navigation token its own restore
flight owned, so a navigation after the flight landed (while the layers
still load) keeps the camera.
- A request that switches the source keeps the open photo up until it is
replaced (holdViewer), so EXPAND stays across sources.
- A lookup given up while it looked leaves the open photo and the panel as
they were, also when it found nothing or failed.
- A failed replacement shows its error inside the expanded viewer.
- A replacing lookup stops a playing Mapillary sequence before the found
photo opens.
- The action refuses coordinates outside the globe, keeps a name given with
coordinates as its label, and fly: false keeps the camera only for
coordinates (its test-only entity and query branches are gone).
- Numbered entities carry their label; a view's street-level point is a pair.
- Mapillary's nearest search always asks for 50 m and 100 images.
- Tests: the lifecycle reproductions, PLAY on replacement, out-of-range
coordinates, half a point, and the named label; three layer tests share
enabledLayer; the second payload digest is gone.
- Docs: the tool count, the panel freshness rule, armed: false, the photo
replacement, the browser-gate comment and a shorter changelog fragment.
- A pasted or start-up link opens its photos only when its restore's camera
landed: a drag or search during the restore flight keeps the camera.
- A posted view whose photo step was cancelled or superseded no longer closes
the viewer or flies; one that found a photo it could not load keeps the error.
- openNearest gains `replace` (keep the open photo while it looks, close it
only when nothing opens, so EXPAND stays) and `proceed()` (asked before a
found photo loads); the action uses both, so a navigation, a gesture or a
hiding mode during the lookup loads no billed panorama.
- Mapillary's timeout message is a msg() string with all eleven translations.
Name routeId among the canonical analyst IDs in CURRENT-STATE, say what the
changelog entry changes for users, and drop the comment that repeated the
canonical-field doc.
A follow-up analyst_query filters on the routeId it was handed back, so a
route ID with repeated spaces or over 180 characters must stay byte-for-byte
raw like the other canonical fields.
The earlier main merge replaced `import { interruptGlobeRotation } from
'../cameraVerbs.js'` with the displayText import while keeping the
`interruptGlobeRotation('voice-zoom-intent')` call, so voice zoom actions
threw ReferenceError. The same merge replaced main's Overpass reservation
paragraph in docs/CURRENT-STATE.md; restore it next to the new text.
Conflicts:
- docs/CURRENT-STATE.md: keep main's hidden-tab polling paragraph and the
display-text normalization paragraph.
- scripts/package-boundaries.json: keep main's voice panel and text chat
modules and add src/data/displayText.js.
- src/voice/gevActions.test.mjs: keep main's set_panel_open and partial
analyst card tests and the display-text tests.
A throw from the crop capture now has a test that it closes its decision with an error outcome. Direct calls to the crop and image senders pin the channel-closed and pointer-scale-not-useful skips. The docs say a send decision requests an image, and when a pointer snapshot counts as already requested.
Street Level: keep the PR's billedPerImage contract note beside main's keyReady, and both sides' layer tests (armed OPEN, failing source) and (keyless default lighting).
Third review pass on #835. No runtime change.
- Controller test: viewport decisions reach the controller's debug log.
- Policy test: a second request for a snapshot still being captured is
logged pointer-already-requested and is not captured twice.
- CURRENT-STATE: pointer-not-fresh also covers a missing snapshot,
already-requested covers an earlier attempt that threw, and the image
step runs when any tool output of the response was sent.
- Changelog: describe what is recorded rather than a fixed miscount.
Second review pass on #835's diagnostics. Behaviour is unchanged.
- Report viewport_context.result as soon as the send returns, so a throw
in the retained-frame bookkeeping cannot log a real send as "error".
- List the decision reasons and when each field is present in
CURRENT-STATE, note that the rate limit can split a pair, and say which
codes are defensive.
- Changelog: the decision is logged when an entity-context answer reaches
the image step, not for every answer.
Review follow-up for #835. What is captured, sent and kept is unchanged.
- Log nothing for tool results other than get_entity_context, so other
tools stop spending the shared debug-log budget.
- Give every decision and outcome the same contextId and image kind, so
overlapping requests pair up.
- Log why a crop decision did not send (stale, off ground, off screen,
expired, pose changed, already requested) instead of leaving a lone
"pointer-ground-crop" that reads as sent.
- Split capture-invalidated into session-reset, superseded, channel-lost
and frame-stale.
- Close the pair with "error" when a step throws, then rethrow.
- Document the events in CURRENT-STATE and reword the changelog fragment.
Wildlife Sightings (#1591) published share token 9, so News takes the
next free token, 00. The token tests read the ledger before both where
they were written against it, and the README counts both layers.
Main's wildlife-sightings layer took share-link token 9, so shipwrecks
takes the next free token, 00. Locale, docs, format-scope and voice
manifest conflicts keep both sides. Ledger and registry tests now model
the ledger before shipwrecks took 00, and layer counts include both
new layers.
Co-Authored-By: Claude Opus 5.5 <[email protected]>