Tilted views: the visible box came from screen rays against the bare
ellipsoid and the 9 tiles kept were the ones nearest its centre, so a view
over high ground (Denver, 1,600 m up) or near the horizon dropped the
streets in front of the camera. At street zooms the rays now meet an
ellipsoid raised to the ground under the camera, hits beyond a range tied
to camera height are ignored, tilted views also keep the ground around the
camera, and tiles are ranked along the line of sight (nadir to screen
centre). Simulated with real Cesium ray maths over Sacramento and Denver
from 2 m to 1,000 m: the nearest visible ground and the screen centre are
always loaded.
Rejected token: Mapillary's 401/403 reached the panel as a raw "Mapillary
tiles HTTP 403" while the pill said ON, and every pan asked Mapillary again.
The proxy now answers a refusal as {keyRejected: true} and holds it for that
token (5 min), and a 429 with its Retry-After (held until then). The panel
reads KEY REJECTED, names MAPILLARY_CLIENT_TOKEN and Provider Settings,
gates its controls like KEY REQUIRED, requests nothing more, and clears the
message when the layer goes off. A rate limit keeps the drawn tiles and
refreshes once the wait is over.
Found in the IC8 review of #768 (P1 tile selection, P2 rejected token).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Fixes from gekh's re-review of 061d269, each with a regression test that
fails on the old code:
- A pan that drops a tile still loading now settles the loading count
(the aborted request no longer owns its key, so its `finally` never
did), and purges the previous zoom's tiles once nothing is pending.
LOADING no longer sticks on the status pill and chip.
- Zoom 0 (above 15,000 km) is a zoom: `zoom == null` instead of `!zoom`.
- A pose that arrives after a photo was closed is dropped, in the
MapillaryJS adapter and in the viewer host, so the marker and sequence
selection do not come back.
- Ground casting passes the tile's signal to the terrain request, so a
cancelled tile stops its height batch instead of holding the queue.
- The server's in-memory tile cache honours the 24 h TTL like the disk.
- The viewer's one-time resize runs for every photo, not just the first.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Both disk caches fed by OVERPASS_UPSTREAMS key only on the query — the Overpass
proxy on the sanitized request body, the installations proxy on the bbox — so an
entry written against one instance is a hit for the same query against another.
After an operator repoints OVERPASS_UPSTREAMS, the previous upstream's answers
keep being served for 7 days, 30 for boundary-class queries, and indefinitely
through the serve-stale path. An install that predates #742 has a cache written
from the public instances that were the default then, and configuring your own
instance does not displace it.
Cached answers now carry a tag identifying the configured upstream set, and a
disk read refuses an entry that does not match it, on the stale path too. With
nothing configured any entry is still accepted: there is no instance to
contradict, and serving last-good data at any age is what the not-configured
path exists for. The tag is a hash of the resolved endpoint list, never the
URLs; it is written to disk and compared on read, and is not sent to clients or
surfaced in errors.
One line is logged once per process when a foreign entry is ignored, naming the
variable and what will happen, as local-receivers.js already does for a
rejected feed.
Cost worth naming: on upgrade every existing entry is untagged, so the first
outage after upgrading serves nothing where it previously served stale data,
until each query is fetched once from the configured instance.
- Lower the Vegvesen default cap to 900, still above the ~850 working
cameras, so the per-pack defaults stay closer to the catalog default.
- Name the flat ground-elevation prior as a known limitation: the feed
is 2D, and mountain-pass cameras can sit below terrain where neither
the point-height prior nor the ground snap resolves.
- README: ~3,900 public cameras.
- Use one HLS figure (~135 working) in CHANGELOG and DATA_SOURCES.
- DATA_SOURCES: list the frame and HLS hosts alongside the catalog host.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
preserve Bilawal's merge history, OpenAI schema-combinator startup fix, and Contacts viewport note
keep the validated voice and analyst behavior while adopting the current provider-neutral feed routes and Google token fallback
Norway's ~850 working road cameras load from one keyless GeoJSON request
to the agency's OGC API view of its DATEX 3.1 CCTV table, so no DATEX
account is needed. Faulty cameras are dropped, every camera sits under a
"Norway" category, and the ~135 that publish HLS play as live video
through the existing in-memory relay with the still as fallback. Frames
and manifests are pinned to each camera's own path on the agency hosts.
CCTV_VEGVESEN_ENABLED=0 disables the pack, CCTV_VEGVESEN_MAX_SOURCES
caps it (nearest the largest cities first) and CCTV_VEGVESEN_VIDEO=0
keeps stills only. Attributed under NLOD 2.0.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Keeps the voice-core prompt rules and adds main's world-query and
vessels-vs-visuals rules; the tool catalog wraps the deixis-aware runner.
Also fixes two things the branch had before the merge:
- Live voice could not start: OpenAI rejects track_entity's top-level
anyOf. Model-facing tools now omit top-level combinators; the runner
and benchmark still require a query or a referent.
- The Contacts-centred analyst answer dropped the viewport coverage note.
The flight, military and vessel feeds were named after the providers the
bundled server uses (/api/opensky, /api/adsblol/mil, /api/ais-live and
their track routes; createOpenSkySource, createAdsbLolSource,
createAisStreamSource). Any server should be able to answer a feed from
its own provider, so the feeds are now:
/api/flights /api/flights/track?icao24=
/api/military /api/military/track?hex=
/api/vessels /api/vessels/track?mmsi=
with createFlightSource, createMilitarySource and createVesselSource. The
old routes are removed. Each server names its provider in X-Feed-Source
(the bundled server sends OpenSky Network, adsb.lol and AISStream, so its
attribution is unchanged) and the military feed's cache state and age in
X-Feed-Cache and X-Feed-Age-Ms. Sources report the provider the server
names and fall back to the feed's name.
b4067f9 moved the OpenSky states/all fetch into fetchOpenSkyStates, with a
10-second limit per attempt and one retry. Keep that and drop this branch's
20-second exchange deadline; the 8 MiB body cap now reads inside the helper
under each attempt's signal. The CHANGELOG bullet moves under [Unreleased],
and CURRENT-STATE describes the per-attempt OpenSky deadline.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Brings in 123 commits from main, including the v0.2.0 and v0.2.1 releases
(MCP server and tool catalog, server security hardening, build hygiene).
Conflicts:
- package.json: kept main's script order and added this PR's
qa:panel-resize and qa:street-level; exports keep both sides (main's
tools/view entries and this PR's street-level and Mapillary entries).
- src/tooling/viteBuild.test.mjs: kept this PR's Mapillary-token define
assertion and main's new wildcard-host allowedHosts assertions.
- CHANGELOG.md: main now has release sections; this PR's entries go under
[Unreleased].
Checks after the merge: doctor, format:check, check:boundaries,
layer-token:check (28 published, 1 new: street-level 0), npm test (5726
pass), build, the Windows job's focused tests (70 pass), qa:panel-resize,
and the keyless street-level gate. The new CSP allows Mapillary (https:
connect/img, blob workers). The keyed gate stops at coverage locally
because the dev .env Mapillary token is invalid (upstream 403), not because
of the merge.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A limit such as 0.5 floored to 0 and disabled the throttle, although only an
explicit 0 is meant to. Positive values below 1 now count as 1.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
/api/google/text-search spends the same Google quota as nearby-places but
was not behind the cross-site gate. It now refuses cross-site requests
before calling Google, and a test covers both paid Google routes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The voice debug log writes the server's timestamp after the client's record,
so a record can no longer replace loggedAt. The dev and preview servers also
send X-Content-Type-Options: nosniff with their shared browser headers.
From the findings in #710.
Co-authored-by: Sunil <[email protected]>
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
The OpenAI endpoints (/api/realtime/token, /api/openai/hud-summary) now
allow 30 requests per minute per client IP and the Google Places endpoints
120, the caps the Pinokio build already ships. GEV_RATELIMIT_OPENAI_PER_MIN
and GEV_RATELIMIT_GOOGLE_PER_MIN still override them and exactly 0 disables
them; an unreadable value falls back to the default rather than to
unlimited.
Salvaged from #683 onto current main.
Co-authored-by: Sameh Khamis <[email protected]>
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
The cost-bearing endpoints (/api/realtime/token, /api/openai/hud-summary,
/api/google/nearby-places) and the voice debug log refuse cross-site browser
requests: a foreign or opaque Origin, a Sec-Fetch-Site other than
same-origin/none, or proxy forwarding headers. Loopback tools without an
Origin and the LAN opt-in keep working. The dev and preview servers send one
shared Content-Security-Policy, and embed-mode framing rewrites only the
frame-ancestors directive of it.
Salvaged from #242 onto current main.
Co-authored-by: Sameh Khamis <[email protected]>
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Voice hears audio directly, and "vessels" is easily heard as "visuals",
which the instructions route to visual styles, so asking for vessels turned
on other layers or a CRT style. The instructions now say a vessels request
without a style name means the ships layer, and a style needs its name.
The voice instructions recommended aircraft_in_area, vessels_in_area and
get_earthquakes, which voice no longer offers, and the docs said voice gets
nearly every query. The instructions now name tools voice has and send
aircraft, ship and hazard questions to analyst_query; a test keeps the
instructions to tools the voice session lists.
Coverage was checked only for packs among the cameras found, so an area
whose cameras were all trimmed away answered zero cameras, complete. The
catalog now reports where each trimmed pack's cameras are, and the tool
marks an answer partial whenever its area overlaps that region.
Camera packs keep their nearest cameras and the catalog cap thins them
further, so London answered 250 cameras while TfL offers many more. The
catalog now records each pack's full count, tags cameras with their pack,
and serves the packs it trimmed; find_cctv_cameras names them.
A stalled global snapshot held the request for over a minute. Each attempt
now gets 10 seconds, a timed-out or failed attempt is retried once, and a
second failure is answered from the stale cache or the regional fallback.
The OAuth token request gets the same limit.
- Voice layer manifest (src/voice/layerManifest.js) generates the layer enums,
aliases and analyst field hints; a test fails when a catalog layer has no entry.
- analyst_query validates fields, operators, units, values and scopes, counts up
to 250,000 loaded records per layer, reports complete/partial/unanswered, tells
LAYER_OFF / NOT_READY / FEED_UNAVAILABLE apart from zero, refuses mismatched
follow-ups and never lets a superseded query replace follow-up memory. Seven
more layers expose analyst records.
- Tool results carry a code-built say line and a display card; the voice card in
the mic control shows captions, plan steps and the result
(src/voice/resultDisplay.js is the one adapter). Slow tools narrate their
current step out of band; holding Space while the assistant talks cuts it off.
Turn spans go to the debug log; caption transcription is metered; the log
omits what was said unless GEV_VOICE_LOG_CONTENT=1.
- Point and ask: the Space-down pointer snapshot resolves "this", "that one" and
"around here" through 'pointer' targets, and "the second one" through
referent:n from the last numbered list.
- Prompt states each policy once (instructions + tools 48.5k -> 46.7k chars).
- Voice benchmark harness (scripts/qa-voice-bench.mjs, scripts/voice-bench/)
shares its phrase table and grading with scripts/qa-voice-routing.mjs.
NHC stamps an advisory with its nominal time and writes it about half an
hour early, so the 5-minute future limit rejected the whole feed each
advisory cycle and the layer showed no storms. Allow 90 minutes ahead.
Traffic roads come from OpenFreeMap vector tiles, or TomTom flow tiles with a
key (Hybrid by default). Military areas use OpenFreeMap polygons with a bundled
worldwide name index; wide views show named points. ALPR reads an hourly
OpenStreetMap extract. Cockpit place context and country, state and US county
outlines use bundled Natural Earth and US Census data. Overpass is used only
when an operator configures their own endpoint.
Also: camera arrivals no longer land below ground, keyless terrain retries on
429, bounded relation geometry for configured Overpass, one shared
OpenStreetMap credit, and separate tile and API transports for embedders.
Co-authored-by: theGaryLarson <[email protected]>
Co-authored-by: Milan Khanal <[email protected]>
fetchTile coalesced concurrent requests for one tile onto a single
upstream fetch, but that fetch ran on the first caller's signal. When the
first request was abandoned (an older pan), every joined request lost a
tile it still needed; and a joined caller that aborted stayed subscribed
until the shared fetch finished.
A flight now owns its AbortController and counts waiters, the pattern the
weather and wind providers use. Each caller's own abort rejects that caller
at once; the upstream fetch is cancelled only when the last waiter leaves,
and a later request after that starts a fresh fetch.
Regression tests (fail on the previous tiles.js, pass now): a joined
request survives the first caller aborting with one upstream request; a
joined caller that aborts leaves at once while the first still gets the
tile; both waiters leaving cancels the owned fetch.
Reported by kvnloo in review.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The proxy already restricted WHERE it would go — https, an allowlisted host, a
station_information/station_status path, no redirects, a 12-second deadline and
a 5 MB body cap — but nothing restricted HOW OFTEN, so the dev server would
relay an unbounded stream of requests at a third-party bikeshare operator on
any caller's behalf, one outbound fetch and one timeout timer each.
Admit 120 requests/minute per client address, global backstop 3x, refused with
429 and `Retry-After` before the target is even parsed. The ceiling comes from
the layer's own demand: `src/layers/bikeshare/registry.js` catalogues 15
systems, station information is cached per city after its first fetch, and
station status refreshes once per city every STATUS_POLL_MS (60s) with
in-flight requests deduplicated — about 15 requests a minute with everything
in range, so 120 is eight times a maximal session.
The method, host, path, redirect, deadline and size guards are unchanged.
Refs #264
Every distinct 5-decimal point was a permanent in-memory entry, a
permanent line in an uncapped `.gev-cache/terrain-heights.json`, and a
live upstream call. Nothing limited the rate, and coordinates were
checked for being finite rather than for being on Earth, so an
out-of-range pair was still a valid cache key.
Four bounds, each following a pattern this repo already uses:
- WGS-84 range is enforced in `parseTerrainPoints`, at the edge, so no
later stage has to. Poles and the antimeridian still parse.
- The point cache holds at most 50,000 entries and evicts oldest-first,
the shape `trackBackfillProxies` uses. Sized from measurement: this
checkout's cache file held 6,189 entries in 1.0 MB of real use, so the
ceiling is eight times that working set. Terrain does not move, so an
evicted point costs one refetch and never a wrong answer.
- A cache file above 24 MB is not read back, mirroring
`rocketLaunchesProxy`. An unmeasurable file is not treated as an
oversized one: without stat, readFile decides, exactly as before.
- The route carries an always-on limiter of 90/min per client, 270
global — the `/api/overpass` band, and an order of magnitude above
measured demand (a full test:track run drove 10 requests, peak 8/min).
The ceilings are constructor options so they can be exercised as
behaviour rather than asserted as constants; production passes none.
They are destructured in the body because `proxyErrorResponses.test.mjs`
slices these proxies to the first column-zero closing brace.
The per-pack defaults already summed to 4,054 before WSDOT (4,304 with
it), past the documented 4,000 cap, so a default install was trimming.
The existing sum check in the DriveBC test listed packs by hand and
missed Calgary and DelDOT. Replace it with a check in
cctvCatalogCap.test.mjs that sums every exported DEFAULT_*_MAX_SOURCES,
and drop the Calgary test that pinned the cap at exactly 4,000.
The loader and normalizer carried the hardware model, but the /sources
response projection and the browser catalog both whitelist fields and
dropped it, so no client consumer ever saw it. Project model and
manufacturer (manufacturer disambiguates bare model strings like "3950")
at both boundaries, and add a test that drives the real middleware and
browser catalog end to end.
WSDOT's layer metadata defines BW (Bothways) plus the diagonals
NE/SE/SW/NW alongside N/E/S/W. Diagonal cameras were being demoted to the
low-confidence id-hash fallback; map them to 45/135/225/315 degrees and
treat BW (not B) as the no-single-facing code. Add CCTV_WSDOT_ENABLED and
CCTV_WSDOT_MAX_SOURCES to .env.example with the other pack controls.
Address review feedback on #595:
- Reject frame links carrying userinfo (user:pass@) in
normalizeCataloniaImageUrl, so catalog data cannot inject credentials
into the server-side frame request on an allowlisted host.
- Fetch cameres.xml over https (verified: same 200 body, no redirect),
refuse redirects with the unread body released, and read the body
through readResponseTextCapped with a 1 MiB CATALONIA_MAX_CATALOG_BYTES
cap instead of an unbounded resp.text().
- Add regression tests for userinfo links on every allowed host, the
refused redirect, and declared and streamed oversize bodies.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
