Full removal, not a hardcoded flag: _is_dispatch_channel, _DISPATCH_TG_RE, the is_dispatch parameter, and _call_fits_incident's tactical branch are gone. One evaluation path for every channel.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
Owner correction from direct scanning experience: a talkgroup named tac/tactical only sees materially different traffic during a real incident, and that's rare -- the bulk of traffic on any monitored channel, including high-risk stops and pursuits, runs on the main channel regardless of what it's named. _is_dispatch_channel's string match on the talkgroup name is a naming-convention guess, not a detector of actual channel behavior; trusting it here meant a busy single-channel department not literally named 'dispatch' would silently get the more permissive 15-minute window and could reproduce #115's original bug (the gate never firing on the channels it targets).
Always use tg_dispatch_thin_idle_minutes (5 min) in the escape hatch, regardless of talkgroup name. Does NOT touch incident_correlator.py's own fast/thin idle-window selection, which uses the same dichotomy for a different, decision-changing purpose -- bigger blast radius, left for its own review.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
Review of #132 found a blocker: PATCH /calls/{id}/transcript wipes tags/severity/location/units/embedding before re-extraction, but not the new scenes map, and doc_set(merge=True) can only add/overwrite nested map keys, never remove one. A call corrected from 3 scenes to 1 kept scenes.1/scenes.2 with pre-correction transcripts and incident_ids forever -- corrupting the exact per-scene tally #96 exists to make trustworthy, and able to re-feed stale text into #114's summarizer fix if a stale scene's incident_id still names a real incident.
Fix: fstore.doc_update(...,{"scenes": fstore.DELETE_FIELD}) -- a real delete, not a merge over an empty map. Added fstore.DELETE_FIELD (re-exports the real firebase_admin sentinel) and stubbed it in the sandboxed test conftest, which didn't have it. Also softened an overclaiming docstring: the Firestore nested-merge behavior is verified against the doc_set wrapper's pass-through, not against live Firestore.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
Every scene of a multi-scene call correlates independently in upload.py's
scene loop, but every scene's corr_debug was written flat onto the same
shared call doc — scene 2's write silently clobbered scene 1's
corr_path/corr_consensus/etc (#96), and summarizer.py read the whole call's
raw transcript per linked call, mixing text from scenes the incident had
nothing to do with, while ignoring transcript_corrected entirely (#114).
Fix: thread a scene_index from both `for scene in scenes:` loops in
upload.py down through _correlate_with_consensus ->
incident_correlator.preview_correlation/correlate_call -> _build_context ->
ctx["scene_index"]. incident_correlator._apply_and_log now writes, in the
same Firestore call:
- the existing flat corr_* fields, unchanged (last-scene-wins, the safe
backward-compatible default for any reader that doesn't know about
`scenes` yet)
- a new nested `scenes.<scene_index>` entry with {transcript, incident_id,
corr_debug}, via doc_set(..., merge=True). Firestore's
DocumentReference.set(data, merge=True) recursively merges nested map
fields by key (documented SDK behaviour, not assumed) — a write to
scenes.1 merges alongside an existing scenes.0 instead of replacing the
whole `scenes` map.
scene_index defaults to 0 for every caller with no scene concept (the
recorrelation sweep, the no-scenes-extracted orphan-check path), so a plain
single-scene call still gets a one-entry `scenes` map equivalent to reading
its flat fields today.
admin.py's _call_summary exposes the new `scenes` list per call (each entry
carrying the same corr_* field names as the flat fields, so the two shapes
are interchangeable to the tally); the summary tally now iterates each
call's scenes-if-present, else its own flat fields, so a 2-scene call with
two different corr_path values counts as two data points instead of one
blend. New `scene_decision_count` sits next to `linked_call_count` to make
that distinction visible.
summarizer.py's _scene_text_for_incident reads a linked call's `scenes` map
to find the scene(s) whose corr_debug recorded a link into the specific
incident being summarized, joining more than one if several scenes landed
in the same incident. Falls back to transcript_corrected-or-transcript for a
call doc with no `scenes` field (predates this change) — the one-liner half
of #114, worth doing regardless since it stops raw-transcript summaries even
for old-schema docs.
Does not touch #80/#95/#102's existing ctx-threading fixes
(embedding/severity/coords/LLM-prompt-transcript) — correct as-is, out of
scope here.
Tests: 14 new (test_per_scene_call_doc.py, test_summarizer_scene_transcript.py,
additions to test_admin_debug_correlation.py) covering the merge shape,
last-scene-wins flat-field backward compat, the admin tally's per-scene vs
per-call counting (including old-schema fallback), and the summarizer's
scene-specific text selection (including old-schema fallback). Full sandboxed
suite: 364 -> 378 passed, all green.
Fixes server-26#96, server-26#114
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
82% of the classifier's backtest flags were <=5-word transcripts that already exit at skip_reason=transcript_too_short before the classifier ever ran, so shadow mode was on track to observe roughly a fifth of the real catch rate. Compute the verdict once, ahead of that check, and fold it into whichever doc_set already runs (no extra Firestore write). Also add a chatter_classifier_flagged/reason tally spanning both linked calls AND orphans in admin.py's summary block -- the target population is non-events, which land as orphans or single-call incidents, so linked alone undercounts it the same way corr_gate_veto would have without the #126 fix.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
Three live measurement windows and two consensus-layer fixes (#125, #126)
converged on one decision (CORRELATION_REVIEW_0907b.md, _0912.md): stop
iterating the correlator's consensus layer, the actual lever is upstream —
a classifier in scene extraction that recognizes radio housekeeping (roll
call, bare 10-4/10-8/98 acknowledgements, unit check-ins) before it ever
becomes a scene for the correlator to judge.
Adds app/internal/chatter_classifier.py: a pure classify_chatter(transcript)
function recognizing two shapes drawn from hand-labeled examples in the
review docs, cross-referenced against the real dumps — not invented regexes.
Deliberately conservative: anything that doesn't cleanly reduce to a known
shape returns (False, None) and the existing pipeline runs unchanged.
SHADOW MODE ONLY. intelligence.extract_scenes computes the verdict next to
the existing _is_garbage_transcript / transcript_too_short gates and writes
chatter_classifier_verdict / chatter_classifier_reason onto the call doc,
but does not skip extraction. admin.py's correlation-debug _call_summary
surfaces both fields, same pattern as corr_gate_veto (#115/#126), so the
next live window can measure the real-world false-positive rate before
anything is wired to actually skip extraction. TODO(server-26#127) marks
the call site.
Backtest against all three existing dumps (1002 calls): 154 flagged, 0
false positives (no flagged call carries tags, coords, non-routine severity,
or matches any review-doc-named dangerous-to-drop transcript — the major
extinguishing-fire call, geocoded calls, pursuit updates, the Pelham Station
subject check, the property-retrieval call, all individually verified).
tests/test_chatter_classifier.py: real transcripts from the dumps/review
docs in both directions. Sandboxed pytest 332 -> 364, green.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
corr_gate_veto was written to corr_debug but the admin endpoint's whitelist (_call_summary + the summary tally) never surfaced it, so the last commit's whole point -- measuring window #4 instead of guessing -- would have produced nothing to read. Add it to both. Also softened the docstring's remaining overclaim: whether the active-only ctx[recent] limitation explains the 2/24 window-3 misses is unanswered, not confirmed -- read corr_gate_veto next window instead of asserting a guess again.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
Review of #126 found: (1) the escape hatch applied tg_dispatch_thin_idle_minutes (5 min) unconditionally, but incident_correlator's own fast/thin path only uses that on dispatch channels and 15 min on tactical ones via _is_dispatch_channel -- mirrored the same selection here, plus a config.py note flagging the second consumer. (2) the docstring claimed a 'confirmed explanation' for 2 window-3 gate misses that was actually wrong (self-contradictory in its own text); replaced the guess with corr_gate_veto, written into corr_debug on every llm=orphan/rules=new disagreement that escalates, so window #4 can see *why* each one escaped instead of reconstructing it from the raw dump.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
_recent_incident_on_same_talkgroup previously treated ANY same-talkgroup
incident within the 2-hour correlation_window_hours lookback as 'recent',
which disabled the whole LLM-orphan consensus gate on busy dispatch
channels: window #3 (CORRELATION_REVIEW_0912.md) measured 0/24 gate fires
against the exact target shape (rules=new, llm=orphan, tiebreak=new), with
22/24 explained by a same-talkgroup incident existing somewhere in the
prior 2h — nearly guaranteed on channels producing 3-13 incidents/2h.
Now the escape hatch only counts an incident as recent within
settings.tg_dispatch_thin_idle_minutes (5 min), reusing the same recency
bound the fast/thin path already uses for the 'dispatch, thin ack 10-30s
later' case this hatch exists for, instead of inventing a new constant.
Investigated the 2 unexplained misses (no same-tg incident found even by
a naive full-collection timestamp scan): confirmed ctx["recent"] is built
from status=="active" incidents with over-capacity incidents dropped
(_build_context / _drop_capped), not a full collection scan — an incident
that has auto-resolved or hit incident_max_calls/incident_max_duration
within the window is invisible to this check even though it is
chronologically recent. This does not explain the 2 misses (a same-tg
incident was absent by both checks there, so some other
_call_is_substanceless condition must be responsible), but it is a real
gap in the check as written. Documented in the docstring with a
TODO(server-26#115); fixing it needs a new, non-active-filtered Firestore
query, out of scope for this pass.
Tests: added a regression test proving an incident inside the old 2h
window but outside the new 5-minute window now correctly gates (fails on
main, passes here), plus a test proving a truly recent (<5min) same-tg
incident still escapes the gate as intended. Sandboxed pytest: 327 -> 329
passed (2 new tests), all green.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
_call_is_substanceless mirrored has_event_substance but not the creation gate's type-resolved short-circuit, so a routine-severity fire/medical call with no coords/tags/vehicles — or a reassignment (unit pulled to a new job) — could be gated to orphan where rules would open an incident. Bail out of the gate on incident_type or reassignment.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
The gate added in ca1d8fb checked rules_decision["corr_debug"] for a positive
signal, but that dict is empty at preview time for action=="new" (corr_path is
written at apply time). The check was always False, so the gate fired on real
events — replayed against corr_dump_9-7_pm.json it dropped ~36 linked calls
including a major "extinguishing fire", a moderate fire-alarm, geocoded calls
and pursuit updates.
Gate now runs against ctx (fully populated at preview time). It fires ONLY when
the call is substanceless: routine severity, no vehicle/geocode/tag, and no
incident already running on the same talkgroup. Any of those escalates to the
tiebreak instead. The substance predicate (has_event_substance) is factored out
of incident_correlator's creation gate and shared, so the two cannot diverge.
recorrelation_sweep: a call the gate parked gets a longer link-only retry budget
(10 vs 3) — the gate fires before any incident for the job exists, so the
substantive call that justifies linking can land after the standard ~6 min.
Still create_if_new=False.
incident_correlator location path: evaluate every in-radius candidate and link
the nearest that carries corroboration, instead of the first in an unsorted
`recent`. A unit-overlap location link is now tagged "location_unit_overlap" so
it stops merging into the fast path's bucket in the admin fit-signal histogram.
tests/test_consensus_gate.py: replaced the corr_debug-signal cases with ctx
substance cases (severity, coords, tags, vehicles, same-tg incident); added a
nearest-wins location test; the two location guard tests now assert they reach
the new guard. Full drb-c2-core suite 322 -> 325.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
CORRELATION_REVIEW_0907b.md measured that radio housekeeping (unit
check-ins, roll call, 10-8/10-98 clearings) is being promoted to
incidents. Every case reads corr_llm_action=orphan, corr_rules_action=new,
corr_consensus=tiebreak -> new: the cheap LLM correctly reads "not an
incident", the rules engine says `new` only because there is no incident to
link to, and the smart tiebreaker then sides with rules ~21/21. Reframing
the tiebreaker prompt (#116) did nothing. The fix is a consensus-logic gate,
not another prompt.
Fix 1 (routers/upload.py) - LLM-orphan gate in _correlate_with_consensus:
when the cheap LLM says `orphan` and the rules engine says `new` with NO
positive event signal, resolve to `orphan` and skip the tiebreak call
entirely. "No positive signal" = the rules corr_debug carries neither a
positive corr_path (unit-continuity / location / fast/disambig / fast/single)
nor a positive corr_fit_signal (unit_overlap / location_proximity). When it
does carry one, the existing escalation-to-tiebreak is kept so a genuine
event the LLM misreads as orphan still gets the second look. The resolved
outcome records corr_consensus="llm_orphan_gate" (greppable, distinct from
"tiebreak") and keeps corr_llm_reasoning / corr_rules_action /
corr_llm_action populated.
Fix 2 (incident_correlator.py) - tighten corr_path=location: the location
path linked on a bare sub-location_proximity_km (0.5 km) distance with no
unit or content check, which stitched a vehicle lockout to a station-restroom
slip and merged two different churches an hour apart. A location link now
requires unit overlap with the candidate OR a distance under a tighter bar
(_LOCATION_TIGHT_PROXIMITY_KM = 0.2 km). Pursuit incidents keep their
movement-speed-validated wide radius. A surviving location link now also
writes corr_fit_signal (unit_overlap | location_proximity), consistent with
Fix 1's positive-signal set.
Tests: new tests/test_consensus_gate.py (13 cases) - the gate resolves to
orphan without calling tiebreak on a no-signal disagreement; a unit_overlap /
location_proximity / unit-continuity / fast-disambig rules signal still
escalates; llm=link vs rules=new still escalates; the location path drops a
shared-area candidate with neither unit overlap nor tight proximity, links on
unit overlap, and links on tight proximity alone. Full c2-core suite
309 -> 322 passing.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Tbknwttzou4s46PAykmtix
The last leg of the #80/#95 scene-context leak. llm_correlator._call_block
read call_doc's whole-call transcript for every scene, so on a multi-scene
call every scene's cheap-tier and tiebreaker decision was made against text
that also contained the other scenes.
- intelligence.py: each processed[] scene now carries its own "transcript" —
transcript_corrected, else this scene's segments joined, else (single scene)
the whole transcript.
- _build_context / preview_correlation / correlate_call: take a `transcript`
param; _build_context resolves ctx["scene_transcript"] from it, falling
back to the call doc (sweep, single-scene, tests) — the fallback is kept
here, unlike embedding/severity, because a scene always has real text.
- upload.py: both scene loops pass scene["transcript"].
- llm_correlator._call_block: reads ctx["scene_transcript"] (call-doc
fallback retained for test-built ctx).
- recorrelation_sweep: passes the call doc's text explicitly.
- +1 regression test. Full c2-core suite green (296 passed, sandboxed venv).
NOT for merge until the running correlation measurement window closes and its
dump is analysed — deploying a correlator change mid-window would mix old and
new behaviour in the sample.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
intelligence.py writes only the primary scene's embedding and severity to
calls/{id}. _build_context read them back off the call doc, so every
non-primary scene of a multi-scene call was correlated against scene 1's
semantic vector and severity rung: a scene about a different event scored
on the embedding path against the wrong incident, and could inherit a
minor/moderate/major severity it never had, clearing the creation gate on
borrowed weight. Same defect and same fix as the #87 coords leak.
- _build_context / preview_correlation / correlate_call: take embedding and
severity as params; drop the call_doc.get() fallbacks. A scene that
passes none has none, and is judged thin on its own signal.
- upload.py: both scene loops pass scene["embedding"] / scene["severity"];
_correlate_with_consensus forwards them. The no-scene unclassified branch
passes neither (correct: no scene, judged thin).
- recorrelation_sweep: passes the call doc's stored values explicitly
(whole-call re-link, link-only, so a borrowed severity cannot create).
- intelligence.py: SCENE DETECTION prompt tightened toward one scene
(server-26#5, partial) - MULTIPLE only for genuinely separate events,
"when unsure, one scene", plus a not-a-new-scene list.
- test_incident_identity.py: +2 regression tests mirroring the #87 test.
Full c2-core suite green (295 passed). #5 prompt change is unmeasured -
needs a scoped correlation-only window. Known remaining legs, tracked
separately: llm_correlator._call_block still reads the whole-call
transcript per scene; content-divergence veto skips on a None embedding.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
POST /incidents/{id}/summarize was gated by require_service_or_firebase_token,
which accepts any authenticated Firebase user including role "viewer". That
route spends OpenAI credits via the background summarizer. The call-side
equivalent was already moved to require_admin_token; this brings the incident
side in line with it.
The frontend's two "summarize now" buttons on the incident detail page are
already gated behind isAdmin, so this backend change matches existing UI
behavior exactly and does not break any viewer/operator surface — it only
closes direct-API access for non-admins.
Swept every other route in incidents.py: list/get are reads with no spend and
correctly stay open to any signed-in user; create/update/delete/link/unlink
were already require_admin_token. No other sibling route needed changing.
Adds test_incident_summarize_auth.py pinning the dependency wiring directly
(the convention used in test_admin_feature_flags.py), so a future revert back
to the weak dependency fails a test immediately.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Board minutes #62 Decision 2 (server-26#64), due 2026-08-31. CTO draft #60
finding 1 and CISO draft #61 finding 3 reached this independently.
GET/PUT /admin/features accepted only a Firebase admin token, so the unattended
runbook had no headless path and SSHed into the c2-core container to write
config/ai_features with the admin SDK. Moving a platform-wide AI cost switch
required a full container shell, and set_flags() wrote no audit entry either
way, so a flag flip was unattributable however it happened.
- New agent_service_key (AGENT_SERVICE_KEY), deliberately separate from the
Discord bot's service_key. Sharing one key would collapse two principals into
a single unattributable identity in every log line, and the bot has no
business flipping AI flags regardless.
- require_agent_key_or_admin accepts the agent key or a Firebase admin, and
rejects the Discord key. The "key is configured" guard is load-bearing:
compare_digest("", "") is a match, so a deployment that never set the key
would otherwise accept an empty credential.
- set_flags() writes an audit_log entry with before/after values and the actor,
wrapped so an audit failure cannot lose the flag write or 500 the route.
- Cascade helper sets the global doc and every system carrying an ai_flags
override in one call. A global False already beats everything, but a system
False beats a global True, so turning AI *on* could half-apply and leave a
radio system hot after shutoff. It scans for the override rather than
hardcoding the two known system IDs, so a new system cannot silently defeat
it.
- cascade defaults to False. PUT /systems/{id}/ai-flags and the AiFlagsPanel
toggle mean a per-system override is deliberate operator intent; cascading by
default would erase it on any unrelated global flip. The runbook opts in.
Issue items 5 and 6 (retiring the SSH path from drb-worksession.md) are NOT
done here and the runbook is untouched. The credential does not exist in
production yet, so the SSH path is still the only one that works; retiring it
now would break the next unattended run. Owner activation is recorded on #64.
Tests 273 -> 289.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
config/ai_features was not the switch it was documented to be. Three paths
spent money with it off, and one path read it wrong, so per-system opt-outs
did not opt anything out.
- Correlation in the ingest pipeline tested the raw global flag instead of the
per-system resolution. With a system opted out, extraction was skipped but
the no-scenes fallback still correlated the call with empty tags, taking the
thin/recency path and attaching it to whatever incident was most recent on
that system. The opt-out did not disable correlation, it disabled good
correlation and left the worst kind running. (#75)
- Transcript correction ran on every transcribed call gated only by an env var,
spending Gemini tokens and a Places lookup per proposed location. An
"STT-only" window was never STT-only and its cost could not be attributed.
Now behind transcript_correction_enabled. (#76)
- _run_extraction_pipeline and the vocabulary learner, both reachable from
PATCH /calls/{id}/transcript, checked no flags at all. (#76, #81)
The flag resolver now lives in feature_flags.resolve_flags() rather than as a
local helper in upload.py. Three copies of that logic is how #75 happened.
PATCH /calls/{id}/transcript now refuses with 409 when correlation is off.
That route wipes tags, severity, location, units, embedding and unlinks the
call from every incident before queueing re-extraction. Gating extraction
alone would have made it destructive-only in the standing flags-off
configuration: the call left blank and orphaned forever, with the route still
answering 200. The wipe and the rebuild are one transaction in intent, so it
refuses before the first write.
Also: the summarizer's stale-incident sweep is no longer behind
summaries_enabled. It is pure Firestore with no model call in it, and gating
it meant nothing auto-resolved while AI was off - so every incident stayed
active forever and the candidate set every correlation reads kept growing.
transcript_correction_enabled is documented as NOT a pure cost lever. The
corrector is also the noise gate that sets not_speech; with it off, recogniser
noise reaches extraction as a real transcript, comes back thin, and
auto-attaches. Never open an evaluation window with correction off and
correlation on.
14 tests added covering flag precedence, both pipeline paths, the 409, the
correction gate and the summarizer no-op. Suite: 264 passed.
Refs #75, #76, #81, #45.
#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>
audio/mpeg was hardcoded at both points call audio is written and served, from
back when the node produced nothing but 16 kbps MP3. It now uploads FLAC, and a
browser will not play a FLAC body labelled audio/mpeg.
storage.py grows one extension -> Content-Type map, used by the GCS upload and
by /media. Keyed off the object's real extension, so every existing .mp3
recording keeps working with no migration -- and _safe_audio_filename already
accepted .flac, so object naming needed nothing.
Also flags what this costs: /media sends the whole body with Accept-Ranges:
none, which was fine at ~60 KB per call and is not fine at ~1.3 MB/min. Noted
at the header and in DEFERRED.md, whose stated reason for deferring Range
support was the old file size.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three backend pieces the /calls page needs, plus the fix for a debug view that
hid its data exactly when it was wanted.
GET /calls/search — paged, filterable call archive. GET /calls returns every
call in one unordered shot: fine for a node's handful of active calls, useless
as an archive. Only the org scope and the started_at ordering go to Firestore,
since that pair is the one composite index that exists; the rest filters in
Python over a bounded window, the same shape admin.py's debug route uses. The
cursor advances over the scanned window rather than the returned page, or a
sparse filter would re-scan from the same place forever.
Manual attribution. POST /incidents/{id}/calls/{id} only ever wrote the legacy
scalar incident_id, never incident_ids -- which is what the correlator writes
and what the frontend queries with array-contains. A manually attached call was
therefore invisible on the incident page it had just been attached to. It now
maintains both and marks the summary stale. DELETE is new: there was no way to
undo an attachment at all, so a wrong link was permanent.
The debug view no longer filters to AI-enabled systems by default. That filter
emptied the view the moment the flags went off, which is precisely when a
window gets reviewed -- on 2026-08-23 it fell from 100 incidents to 6 between
switching correlation off and opening the tab. ai_systems_only=true restores it.
It also returns a summary block now: corr_path / fit_signal / consensus /
llm_action tallies, transcript coverage on both linked and orphaned calls,
single-call and median-calls-per-incident for fragmentation, max span and
anything past the server-26#22 caps for merging, and the count of incidents
still carrying a fallback "— TGID" title. All of it was being recomputed by
hand from the raw payload on every review.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
84 of the 100 incidents in the 2026-08-23 dump were titled "Ems — TGID 9048"
or "Other — TGID 9600" -- the fallback, not a description. The title is the
incident's name everywhere it appears: list rows, map pins, Discord alerts.
_create_incident builds it from a content tag and a talkgroup label, and the
label was collapsing to "TGID {id}" because talkgroup_name arrived as None.
It is a plain form field on /upload, forwarded untouched into correlation, and
the node only sends it when OP25 had the name in its loaded tags file -- which
is exactly the case C2 can cover from its own systems collection, where all 125
talkgroup definitions live.
The lookup already existed, on the other path: mqtt_handler resolved it from
the system config on call_start. So the call document held the right name while
the pipeline that titles the incident ignored it. That asymmetry is the bug.
internal/talkgroups.py is now the one implementation -- caller's hint, then the
call document, then the system config -- and both paths use it.
_run_intelligence_pipeline resolves once at the funnel /upload and
/calls/{id}/reprocess share, so the dispatch-channel test, scene extraction and
the title all see a real name. When the call document was the thing missing it,
the resolved name is written back, so the archive and the orphan panel stop
showing a bare TGID too.
Also gives fast/thin a corr_fit_signal. It is 63% of all links and was the only
path writing none, so corr_fit_signal was absent on 295 of 309 calls and the
admin debug view's distribution panel read empty -- looking broken when it was
faithfully reporting that the dominant path records nothing. It now says
thin_recency, which is what actually decided it.
Closes server-26#34. Refs server-26#35 -- the tier's 3.5% invocation rate is a
cost/benefit question, not a bug, and stays open.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
/admin/debug/correlation stripped corr_consensus and the corr_llm_* fields
that upload.py's consensus correlator writes onto the call doc, making it
the one tool built to answer "is the LLM correlation tier alive" unable to
answer it (2026-08-19 dump had to infer LLM state from commit dates instead
of reading it off the data). admin.py's _call_summary() now includes
corr_consensus, corr_llm_reasoning, corr_llm_action, corr_rules_action.
The unit-continuity correlation path never wrote corr_matched_units, unlike
fast/single and fast/disambig, so the debug view showed null for a match
that was in fact unit-driven by construction. Now populated unconditionally
on that path (server-26#16).
Also traced the negative corr_incident_idle_min (-4.1 observed) to its root
cause: the re-correlation sweep anchors `now` to the linking call's own
started_at, and that back-dated value was being written straight into the
incident's updated_at, letting it land before the incident's own
started_at. Added _floor_at_started_at() so updated_at can never precede
started_at. (commit 33a247d already fixed the recency *gates* misreading
that negative value; this fixes the write that produced it.) Verified the
skip_reason filter in recorrelation_sweep.py:63 is already correct, no
change needed there.
Added tests for the debug endpoint's LLM field passthrough, the
unit-continuity corr_matched_units fix, and the updated_at floor — each
confirmed to fail when its fix is reverted. 148 passed, 0 failed.
Closeslogan/server-26#24Closeslogan/server-26#16
/upload compared the per-node API key with a plain !=, which short-circuits on
the first differing byte and so leaks a little information about how much of a
guess was correct.
The reason to fix it is less the timing channel itself -- an HTTP round trip is
noisy -- than the inconsistency: enrollment.py and dynsec.py both went out of
their way to use secrets.compare_digest for the same class of credential, so the
codebase contradicted itself on whether this mattered. Now it does not.
Also coalesces a missing api_key field to "" so compare_digest is never handed
None, which would raise TypeError and turn a malformed node_keys document into a
500 instead of a 401.
Closeslogan/server-26#12
#17: severity was written once at _create_incident and never touched again,
so an incident that opened routine and escalated to a working fire stayed
routine forever. _update_incident now merges call_severity into the incident
via _max_severity() on every link.
Severity is monotonic: it only ever rises, never falls. An incident briefly
assessed "major" genuinely was major at that moment; a later, calmer-sounding
call is evidence the situation is winding down, not that the earlier read was
wrong. status/resolved_at exist to retire an incident — severity should stay
as the high-water mark so the worst-first rail, "Major only" filter, and map
colouring never bury a call that was genuinely major. See _max_severity's
docstring in incident_correlator.py for the full argument.
#18: none of the resolution sites wrote resolved_at, so an incident's
lifespan couldn't be reconstructed for the history-scrub feature. Added
resolved_at alongside status="resolved" at all six sites that flip it:
- incident_correlator.py _update_incident (signal-based: units all cleared)
- incident_correlator.py maybe_resolve_parent (master auto-resolve)
- summarizer.py _stale_sweep (90-minute auto-resolve)
- upload.py, both scene-resolution loops (single- and multi-scene)
- calls.py reprocess/correction path
(_update_incident's signal-resolve and maybe_resolve_parent's master-resolve
weren't named in the issue's four call sites, but they set status the same
way and were missing resolved_at too.)
No backfill: existing resolved incidents keep resolved_at = null, which
means "resolved before this field existed," not "never resolved." Backfilling
from updated_at would be a guess dressed up as data.
Tests: added to tests/test_correlator_gate.py, which needs no Firestore for
the pure _max_severity cases and patches fstore for the _update_incident/
maybe_resolve_parent writes. Covers the escalation case (routine -> major),
the no-downgrade case, and resolved_at on both the signal-resolve and
master-resolve paths. 52/52 passing in that file; 83 passed / 10
pre-existing failures for drb-c2-core overall (baseline was 69/10 — the
+14 is exactly the new tests, no regressions).
Fixes#17, #18.
/admin/debug/correlation built its call lookup as {doc["call_id"]: doc}, which
raises KeyError on any stored call missing that field -- and at least one in
production is missing it. One bad document took down the entire view rather
than dropping a single call from it.
The document id is authoritative and always present; the call_id *field* is
written by the upload path and evidently has not always been. Keying off the id
we asked for removes the dependency on the field entirely.
Found while generating a correlation dump server-side, because the UI route this
serves has been unusable tonight.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
/admin/debug/correlation read every incident ever created, sorted them in
Python and kept 20, and separately pulled every call in the orphan window with
no cap. That worked while the collections were small. They are not small now:
Firestore kills an unbounded scan with a 503 and the request never returns, so
the debug view simply spins -- which is also what made the org backfill script
fail earlier tonight, same cause, different caller.
Incidents now come back pre-sorted from Firestore with a limit, and the orphan
scan is capped at 3000 documents. Both queries order on the single field they
already filter or sort by (updated_at, ended_at), so neither needs a composite
index -- worth preserving, since the index file from the tenancy work has not
been deployed.
Capping introduces a way to be wrong quietly: a truncated window looks exactly
like a quiet night. The payload now carries incidents_window_exhausted and
orphan_scan_truncated so a short result announces itself instead of being read
as a correlation improvement.
The AI-system filter still runs in Python, so the incident window is 10x the
requested limit rather than the limit itself.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The previous commit shipped Firestore rules that reference an org_id claim
nothing issues yet, and an org_id filter nothing writes yet - this is the
commit that makes both real. Backend half of SAAS_PLAN.md B2/B2b/B2c.
Data model: organizations/{org_id} and org_members/{uid} are new
collections (models.py OrganizationRecord/OrgMember). org_id is now an
Optional field on NodeRecord, SystemRecord, CallRecord, IncidentRecord,
AlertRule, and AlertEvent - optional because every existing document
predates it; scripts/backfill_org_id.py (written, not run - it touches
production Firestore and Firebase Auth claims) is what closes that gap
later. plan_id/subscription_status/stripe_* on OrganizationRecord are
deliberately None: no billing or pricing model has been decided, so this is
a seam, not a promise. app/internal/tenancy.py holds FOUNDING_ORG_ID, the
org every pre-tenancy document and every legacy enrollment path resolves
into.
Where org_id comes from, end to end: a customer's node enrolls with a
per-org token (new enrollment_tokens/{token_hash} collection, minted via
POST /org/enrollment-tokens - new routers/org.py) instead of the old
fleet-wide ENROLLMENT_TOKEN, which still works as a fallback that resolves
to FOUNDING_ORG_ID so an already-deployed node's .env doesn't start failing
today. The node's org_id then flows onto every call it produces
(mqtt_handler.py's call_start/call_end, upload.py's /upload handler all
resolve it from the node doc), and onto every incident correlated from
those calls (incident_correlator.py's _create_incident/_create_master_incident).
That last one is the part that isn't just a read filter: _build_context's
`all_active = collection_list("incidents", status="active")` fed every
correlation candidate - fast-path talkgroup match, unit-continuity,
disambiguation - from the entire incidents collection, unscoped. Without
scoping it to the call's own org_id, a call from org A could link into an
incident org B already owns, which is a cross-tenant data merge at
correlation time, not just an over-broad read. Same shape of bug in
alerter.py: rule matching pulled every enabled alert_rule regardless of
org, so org A's keyword rule could fire (and POST org A's Discord webhook)
on org B's radio traffic. Both now resolve org_id from the call doc itself
rather than threading a new parameter through every caller.
Every list/get route gained org scoping via a new resolve_caller_org_id()
helper in internal/auth.py, which handles the three credential shapes those
routes accept (service key, node api_key, Firebase user) uniformly and
returns None (unrestricted) for the service key and platform admins -
preserving today's single-org behaviour exactly while closing the leak for
everyone else: GET /nodes, /systems, /calls, /incidents, /alerts,
/alert-rules. Write routes for nodes/systems (approve, create, delete, etc.)
deliberately stay platform-admin-only for now rather than being loosened to
org-owner/operator - that's a real gap called out in SAAS_PLAN.md 2.4's
"should be" column, but it's a separate authorization redesign the 12-item
build order doesn't actually enumerate, and doing it half-considered here
risked being exactly the "half-applied filter is worse than none" failure
mode the plan warns about. Today's founding org keeps working unchanged;
loosening node/system management to org owners is follow-up work, flagged
rather than guessed at.
Also closed the four spend/access-attack routes SAAS_PLAN.md B2c called out
by file and line: POST /calls/{id}/reprocess is now admin-only (was any
signed-in viewer looping the Whisper+Gemini pipeline for free - DEFERRED.md
had this as a live, independent-of-SaaS exploit) plus a per-call rate
limiter as a second guard; POST /alerts/{id}/acknowledge now checks the
alert's org_id; GET /admin/features moved from require_firebase_token to
require_admin_token; and trips.py's four unauthenticated mutation routes
(create_trip, update_trip_tags, create_event, update_event) are now
restricted to the founding org (or the bot's service key, or a platform
admin) - trips has no org_id of its own and isn't getting one, since
[[trips-feature-intentional]] says it's an internal utility riding along on
this stack, not a tenant-scoped product surface.
New public-but-scoped seam: POST /auth/signup (routers/links.py, alongside
the existing /auth/link* routes) provisions an organizations doc and an
owner org_members doc for a just-created Firebase user, then sets their
org_id/org_role claims - idempotent, so a double-submit doesn't create two
orgs. This is the only route that turns "has a Firebase account" into "can
read anything," which is what the frontend AuthProvider no-claim guard
(next commit) is built around.
Also new: GET/PATCH /org for the organization profile (closes the disabled
"Save changes" button noted in DEFERRED.md - there was no organizations
concept to save into before this), and POST /waitlist (public, source-IP
rate-limited, not coupled to any plan or tier - the commercial model is
still an open decision per SAAS_PLAN.md section 6).
Verified: all touched files py_compile clean; c2-core pytest is 69
passed / 10 failed, matching the documented pre-existing baseline exactly
(DEFERRED.md - mqtt_handler/node_sweeper test-vs-code drift, unrelated to
this change) - no new failures. flake8 --max-line-length=120 shows no new
violations in any touched file (checked each new E501/E221/E30x against
`git diff` to confirm it predates this commit); c2-core has no CI lint gate
regardless (CLAUDE.md - flake8 only runs in Client CI).
No new environment variables. Firestore composite indexes for the queries
this introduces were already shipped in the previous commit
(infra/firestore/firestore.indexes.json).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two independent sources of garbage in the AI pipeline, both visible in the
2026-08-16 correlation dump.
1. Hallucinated transcripts. The Whisper prompt opened with an enumerated run
of ten-codes: 10-4, 10-23, 10-20, 10-97 and so on. Whisper treats prompt
text as preceding transcript, so on noisy or silent audio it continued the
series, emitting transcripts that count upward from 10-4 to 10-99. The
existing no_speech_prob filter could not catch these: the model is highly
confident in text it invented by continuing a pattern.
The prompt no longer contains a series to extend, and _is_degenerate()
rejects the three shapes this failure takes: ascending ten-code runs, one
phrase looping, and near-identical segments across a whole recording.
Verified against 13 transcripts from production: all four known
hallucinations rejected, all nine real ones kept, including terse traffic
containing legitimate codes.
2. Duplicate recordings. node-002 and node-PI-2 both cover TG 9048 and both
uploaded the same transmissions, ~1.1s apart. Nine pairs appeared in one
dump. Each was transcribed, billed and correlated twice, and the resulting
incident listed two units where there was one.
Canonical selection is by earliest started_at, tie-broken on call_id, NOT
by upload order: upload order varies with encode time and network latency,
so it would make the authoritative recording non-deterministic. Call
documents are created from MQTT call_start before uploads arrive, so both
nodes independently reach the same verdict. The loser keeps its audio (it
may be the cleaner capture) but is excluded from STT, correlation, the
re-correlation sweep and the orphan debug view.
Also fixes _sync_transcribe returning a bare None when OPENAI_API_KEY is
missing, where the caller unpacks two values. A missing key surfaced as a
misleading "Transcription failed" instead of the real warning.
Adds tests/test_dedup.py (15 cases). dedup.py reaches Firestore through an
injected callable so it stays importable without firebase-admin present.
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.
Edge nodes are deployed to arbitrary locations by arbitrary people, so the
broker has to be reachable from the internet and secured on its own merits
rather than by a VPN.
Three defects made that impossible. The broker only had a plaintext 1883
listener; every node shared one drb-node password; and the ACL pattern used
%c, the client-supplied client id, so any holder of that shared password
could set client_id to another node and take over its namespace. The comment
claiming this cryptographically prevented cross-node access was wrong and is
gone.
Authentication now uses mosquitto 2.x's built-in dynamic-security plugin on
the stock eclipse-mosquitto image. c2-core administers it over the control
topic, creating each node's client on approval with username=<node_id> and
password=<its node_keys api_key>, attached to a role whose ACL is nodes/%u/#
against the authenticated username. One credential, one revocation point.
An HTTP-callback plugin was implemented first and rejected: that project is
archived upstream, which is not an acceptable dependency on an
internet-facing broker.
Because dynsec state is a second source of truth alongside Firestore,
approve/reissue/delete now write to the broker first and surface a 502
rather than drifting, and c2-core reconciles every approved node into dynsec
on startup.
Adds node self-enrollment (POST /nodes/enroll, GET /nodes/{id}/credentials)
so a new node can obtain its key over HTTPS without an operator handling
secrets by hand. Enrolling an already-approved node_id is refused on the
fleet token alone — otherwise a leaked token plus a guessable id would let
an attacker steal a live node's key before the real node asked for it.
Pickup secrets are stored hashed and returned once, and the endpoint is rate
limited per source IP.
Infrastructure: an 8883 TLS listener fed by Caddy's certificate via a
systemd path unit, a firewall rule for it, and Caddy now 404s /internal/*
so the api vhost cannot proxy internal routes.
Also fixes CORS, which allowed https://app.<domain> while the frontend is
served on the bare domain — every call from the portal would have failed —
and widens the vault gitignore to a glob, since ansible-vault leaves
backup siblings that the exact-name rule left committable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
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.
Switch from legacy Places textsearch and Directions APIs (disabled on
this project) to Places API (New) and Routes API (New). Both places.py
and the assistant's _places_search helper updated. Also fixes uid()
recursive self-call in trips page and adds Places API response logging.
auth.py
secrets.compare_digest replaces == for service key comparison (timing-safe)
Added require_service_key — bot-only endpoints (trip/event join/leave)
Added require_service_key_or_admin — node commands/config (bot via service key OR dashboard admin via Firebase)
Added _RateLimiter with three shared instances: trip_chat_limiter (20/5min per user), summarize_limiter (5/10min per incident), bootstrap_limiter (2/hr per system)
nodes.py
send_command and assign_system now require require_service_key_or_admin — the Discord bot can still call them via service key, but regular Firebase users are blocked
tokens.py
add_token, flush_tokens, set_preferred_system, delete_token all require require_admin_token
Token masking changed from token[:10] + "…" + token[-4:] to "•••" + token[-4:]
systems.py
All write endpoints (create, update, delete, ai-flags, ten-codes, vocabulary writes, bootstrap) now require require_admin_token
bootstrap_vocabulary also calls bootstrap_limiter.check(system_id)
incidents.py
POST /incidents/summarize (bulk) now requires require_admin_token
POST /incidents/{id}/summarize now calls summarize_limiter.check(incident_id)
trips.py
join_trip, leave_trip, join_event, leave_event require require_service_key — only the Discord bot can set Discord attendee identity
delete_trip, delete_event require require_service_key_or_admin
trip_chat rate-limited per caller UID, history stripped to user/assistant roles only, user message truncated to 2000 chars, Maps query strings capped at 200 chars
upload.py
Rejects files larger than settings.upload_max_bytes (default 100MB) with 413
storage.py
_safe_audio_filename() derives GCS object name from call_id + allowlisted extension, completely ignoring the client-supplied filename
config.py
Added upload_max_bytes: int = 100 * 1024 * 1024
Both Dockerfiles — python:3.14-slim → python:3.12-slim
New /trips router with full CRUD, attendee management, and nested
events. Events validate date is within parent trip range and inherit
trip location when not explicitly set. Leaving a trip cascades
removal from all its events.
New TripCommands cog with /trip create, list, view, delete, join,
leave and /trip event add, remove, join, leave. Event autocomplete
is scoped to the selected trip. Enforces must-be-on-trip rule for
event joins with a clear error message.
Refactor incident_correlator.py to a decision/commit split (preview_correlation
/ apply_correlation) so the rules engine and LLM can both produce decisions before
anything is written to Firestore.
Add llm_correlator.py: cheap Gemini Flash first-pass + Gemini Pro tiebreaker.
Wire _correlate_with_consensus in upload.py — rules-only fallback when key is
absent or call is thin; agreed/tiebreak consensus written to corr_debug.
- Geocoding: reject GEOMETRIC_CENTER/APPROXIMATE results — vague location strings
(regions, city centroids) were resolving to node-area coords and creating false
proximity matches that merged unrelated incidents
- Thin path: on dispatch channels with multiple active incidents, skip attachment
rather than guessing — "10-4" with 3 active incidents is genuinely ambiguous
- Short transcripts (≤5 words) now write skip_reason="transcript_too_short" to
the call doc, matching garbage transcript behavior
- upload.py no-scenes fallback now checks skip_reason before running correlation —
flagged calls (garbage, too short) no longer attach via thin path
- Update Server README to reflect current project purpose, goals, and pipeline
- Cap unit-continuity path at 20 min idle (unit_continuity_max_idle_minutes)
- Block time_fallback and unit-continuity matching on reassignment calls
- Expand reassignment detection to cover unit-initiated self-reassignment
- Skip GPT extraction entirely for transcripts ≤5 words (prevents hallucinated tags/units)
- Reduce geocode_max_km from 75 to 40 to reject far-out-of-area results
- Include county in geocoding query for tighter jurisdiction anchoring
Logan CusanoandClaude Sonnet 5