#36 — the correction pass shipped in 58efdbd was right, its reference-data
shape was not. One shape now, at both scopes, every field nullable:
area_context: { municipality?, county?, state?,
center?, radius_km?, resolved_from?, resolved_at?,
local_knowledge?: [{term, meaning}] }
`state` closes the ambiguity that made "Ossining" a national guess.
`local_knowledge` replaces roads[]/landmarks[], which could not hold
intersections, schools or nicknames and carried no meanings — `11-X-ray` is
useless alone, `11-X-ray — MTA PD patrol unit` is what a corrector can act on.
Pre-#36 roads[]/landmarks[] are read forward as bare terms so nothing an
operator already entered is lost.
Nullability is the mechanism: which scope gets filled is the operator's
declaration of how homogeneous the system is. One town — fill it once at system
level. Statewide — leave it blank and fill each talkgroup.
The backend owns the derived anchor. PUT /systems/{id} merges config.talkgroups[]
against what is stored instead of writing the client's blob verbatim, which
would have erased the anchor and the pending queue — the same defect as the
ten_codes wipe.
#37 — Maps as a verifier, not as prompt stuffing. The corrector emits its
location nouns; each is geocoded against the talkgroup's anchor, and on a miss
we look for a sound-alike that does resolve there, correct to it, and propose
{term, meaning} to that talkgroup. Cost scales with location nouns, not calls.
No anchor means SKIP. An area too wide to discriminate stores no anchor at all,
because a statewide radius would confirm anything inside it — verification that
passes everything is worse than none, since it reads as a check in the data.
Also re-anchors _geocode_location, which rejected results >40km from the NODE
(server-26#6). An antenna is not a jurisdiction; distance-from-node was always
a stand-in for the anchor and is now only the fallback.
The induction loop proposes at talkgroup level and never promotes. Blast
radius: a wrong term on a channel misleads that channel, the same term
system-wide misleads one 400km away on a statewide system.
38 new tests; 240 pass. Frontend typechecks clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Correction existed, but as a line in intelligence.py's EXTRACTION_PROMPT --
which put it in the wrong place twice over. The same model call that extracted
units, location and severity emitted the correction afterwards, so extraction
reasoned over text already known to be wrong; and it sat behind
correlation_enabled, so during a cost-controlled STT-only window nothing was
ever corrected at all. That is the normal state during development.
internal/transcript_correction.py is now its own pass, between the degenerate
filter and the Firestore write. It receives an already-produced transcript plus
a reference list, so unlike a Whisper prompt it has no series to extend -- the
distinction that keeps vocabulary out of the recogniser's prompt, where an
enumerated ten-code list once made it hallucinate ten-code runs.
Reference data is merged from the talkgroup and the system, TALKGROUP FIRST. A
system spanning several counties can have a talkgroup covering one
municipality, and that municipality's streets must not be buried under a
county-wide list. A single-municipality system is the degenerate case: populate
the system level and every talkgroup inherits it. Area context is now SET --
municipality, county, roads, landmarks, on both scopes -- rather than guessed
from talkgroup names, which is what vocabulary_learner did and which is close
to useless across multiple counties.
Segments are corrected too, not just the joined text. extract_scenes builds its
prompt from numbered segments whenever there is more than one, so a correction
that only fixed the transcript would have been discarded on exactly the
multi-transmission calls carrying the most content. Alignment is enforced: an
array of the wrong length or type is dropped whole, because scenes map back to
transmissions by index and a shifted array would misattribute audio silently.
Whisper is also retried once on degenerate output. Call e49ea32c produced a
56-word ten-code counting run on one attempt and ordinary speech on the next --
same clip, same temperature=0 -- so a hallucination is a coin-flip, and
discarding on the first bad roll threw away a recoverable transcript.
Two things found on the way:
PUT /systems/{id} wiped ten_codes on every save. The systems form sends only
{name, type, config}, and model_dump() wrote every omitted field as its default
over the top. Now exclude_unset. area_context would have been the next victim,
which is why it gets its own route alongside ten-codes rather than a field on
that payload.
Closes server-26#36.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Declare the fields the backend already writes and the UI was discarding:
CallRecord gains location_coords, units, vehicles, cleared_units,
duplicate_of, srcaddr (intelligence.py ~315-327); IncidentRecord gains
units_active, units_cleared, location_mentions, last_thin_at
(incident_correlator.py _attach, ~1270-1300).
Filter duplicate_of client-side in useCalls.ts's three hooks (useCalls,
useCallsByIncident, useActiveCalls) so a call flagged as a second node's
recording of the same transmission no longer renders twice. Client-side
rather than a where() clause to avoid a new composite index.
Removes the two now-resolved DEFERRED.md entries (dedup/useCalls,
lib/types.ts field gaps).
Per UI_REDESIGN.md chunk 3.
The UI worked but read as an operator console: no public face, no way to
describe or sell the thing, and no account surface beyond the node list. This
adds the missing halves and reorganises what was already there around the
incident, which is the unit of value the rest of the pipeline is built to
produce.
A shared design system replaces per-page styling: components/ui (Button, Card,
Badge, EmptyState, Skeleton, PageHeader), a type scale and shadow set in the
Tailwind config, and light-mode tokens in globals.css. The existing
html:not(.dark) remap mechanism is extended rather than replaced -- a parallel
theming system would have been two sources of truth for the same colours.
Public marketing pages (/, /features, /pricing, /faq) load without a session.
middleware.ts gained a PUBLIC_PATHS allowlist to permit that; it remains a UX
redirect and is still NOT an authorisation boundary, which the comment there
says explicitly. Real enforcement is unchanged and still lives server-side in
c2-core's auth.py. Chrome switching is done by pathname in ChromeSwitcher
instead of by route group, because a route group would have collided on / and
forced most of app/ to move for no behavioural gain.
Billing and API keys ship as typed stubs, not integrations. lib/billing.ts and
lib/apiKeys.ts define the data model and the screens consume it, but every
mutating call throws with a message naming the backend route that has to exist
first, and the sample data is labelled as sample. Nothing here can charge
anyone or mint a real credential -- picking a payment processor and holding its
keys is a decision for a human, and a half-wired checkout is worse than an
obviously absent one.
The severity work from the c2-core change lands here too. severity is now a
filter and sort dimension on the incident list rather than decoration, since
a busy dispatch channel is only readable if you can collapse it to moderate and
above. routine gets a muted treatment because it is the majority of traffic,
legacy "unknown" still renders nothing, and TypeBadge handles the new "other"
incident type. Severity rendering moved into lib/severity.tsx so the incident
list, incident detail and call rows cannot drift apart.
Deliberately not touched: calls, map, alerts, nodes, systems, tokens, trips and
admin. They already share the palette and stay coherent, and rewriting them
would have buried the parts that actually needed to change. No colour tokens
were renamed, so nothing regressed there.
Verified with tsc --noEmit (npm run typecheck), clean. No runtime verification
was possible and none was done. No new environment variables.
upload_audio() could only sign a URL when GCP_CREDENTIALS_PATH pointed at a
service-account key file. The deployed VM runs on Application Default
Credentials with no key file, so every upload silently took the fallback
branch and returned a bare gs:// URI. That broke two things at once:
* Browsers cannot fetch a gs:// URI, so no recording was ever playable.
* _public_url_to_gcs_uri() only matched https://storage.googleapis.com/ and
returned None for it, so `if gcs_uri:` in the upload path was always false
and transcription never ran. Nothing was logged, which is why this looked
like an OpenAI credits problem rather than a storage one.
The fallback also interpolated the client-supplied filename instead of the
call_id-derived safe name, so the URI did not even name the object written.
Calls now store only the canonical gs:// location. A short-lived playback link
is minted per read as an HMAC over (call_id, expiry) keyed by SERVICE_KEY, and
audio is served from the private bucket by the new /media route. An <audio src>
cannot carry an Authorization header, so the link has to be the credential;
that router is therefore public with the check done inline, as enrollment.py
already does. Signing GCS URLs from the VM would have needed a
serviceAccountTokenCreator grant on its own service account — this avoids the
IAM change entirely and keeps the bucket private.
gcs_uri_for_call() reconstructs the object name from call_id, so recordings
made before this fix are reachable again without a data migration.
Frontend rows come straight from Firestore via onSnapshot and never see a
server-minted field, so CallRow fetches the link lazily on expand.
Also removes the last long-lived (1 year) signed URL and the log line that
printed it.
Trip-level tags: admins configure available tags in the trip header (inline add/remove pills). The AI can also create new tags via the add_tag tool.
Event tags: selectable in the Add Event modal, shown as colored pills on event cards in the timeline, and on AI suggestion cards.
AI integration: sees available tags in its system prompt, applies them when proposing events, can create new ones with add_tag.
Discord: tags shown as inline code blocks under each event in /trip view.
Colors: auto-assigned from an 8-color palette by tag index, consistent everywhere.
lib/types.ts — TripRecord and TripEvent types
lib/c2api.ts — getTrips, getTrip, createTrip, deleteTrip, createTripEvent, deleteTripEvent
lib/useTrips.ts — Firestore realtime hook on the trips collection, ordered by start date
app/trips/page.tsx — List page split into Upcoming / Past sections, card click navigates to detail, "+ New Trip" modal for admins with all fields including date range and maps link
app/trips/[id]/page.tsx — Detail page fetched via C2 API (gets trip + events in one call), day-by-day itinerary with time, location, maps link, notes, and Discord attendees. Add Event modal (date constrained to trip range). Admin-only delete trip + remove event.
components/Nav.tsx — Trips link added to the nav
app/map/page.tsx
Removed IncidentCard component and the incidents grid below the map — the on-map sidebar inside MapView is the single display
Moved kiosk exit button from top-3 left-3 (overlapping zoom controls) to bottom-[5.5rem] left-3
components/MapView.tsx
Fixed popup "View incident →" link — adds stopPropagation() + window.location.href to prevent Leaflet intercepting the click
Added "View details →" link on each sidebar incident card so you can navigate from the map panel without opening a popup
Added "News Alerts" overlay layer (placeholder, ready for RSS/feed integration)
lib/types.ts
Added preferred_token_id?: string | null to SystemRecord
lib/c2api.ts
Added setPreferredToken(tokenId, systemId) calling PUT /tokens/{tokenId}/prefer/{systemId} (backend already existed)
app/systems/page.tsx
Added PreferredTokenPanel component — loads the token pool lazily on expand, shows radio buttons to set/clear the preferred token, displayed on each system card above the AI flags panel
When the induction loop proposes a new vocabulary term, it now records
which sampled call(s) most likely produced the suggestion. Admins see
a collapsible "▶ source" player under each pending term showing the
audio clip and transcript, so they can hear what was actually said
before approving or dismissing.
- vocabulary_learner: track sampled call docs, attach source_call_ids
to each pending term via word-overlap search with fallback
- types: VocabularyPendingTerm.source_call_ids?: string[]
- c2api: add getCall(id) using existing GET /calls/{call_id} endpoint
- VocabularyPanel: SourceCallPlayer component — lazy-loads call on
first expand, shows audio controls + transcript snippet
incident_correlator.py — full rewrite: always runs on every call, fetches all active incidents cross-type, fast path collects all talkgroup matches and disambiguates by unit/vehicle overlap → location proximity → embedding, new location proximity path, slow path requires location corroboration, "Auto:" stripped from titles, "auto-generated" tag added, units/vehicles now accumulated on update
intelligence.py — resolved field in GPT schema, returned as 5th value
upload.py — both pipelines unpack 5-tuple, always call correlate, auto-resolve on resolved=True
summarizer.py — stale sweep runs each tick, resolves incidents idle for 90+ minutes
config.py — correlation_window_hours=2, embedding_similarity_threshold=0.93, location_proximity_km=0.5, incident_auto_resolve_minutes=90
Logan CusanoandClaude Opus 5