A) _render() now emits multi-line format matching Fire/Roads style:
emoji prefix M{mag} — place / Depth · coords / TSUNAMI WARNING
B) Environment.tsx usgs_quake panel replaced — dead min_magnitude/bbox
controls removed, wired to real adapter_config keys: global_mag_floor,
regional_mag_floor, regional_radius_mi, escalate_mag_floor,
broadcast_pager_alerts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move work zone settings out of itd_511 into dedicated wzdx adapter:
- config.py: add WZDxConfig dataclass with feed settings
- defaults.py: migrate 3 work_zone keys to wzdx namespace (broadcast,
min_severity, sub_types) + add ADAPTER_META entry
- incident_handler.py: work zone gate reads adapter_config.wzdx
- Environment.tsx: full WzdxConfig state/load/save/discard, native feed
fields when feed_source!=central, broadcast settings panel
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add 'wzdx' adapter key with its own META entry and render block.
Move work zone controls (enable toggle, min severity, sub-types)
out of the roads511 panel into the new WZDx tab. Data still
loads/saves via /api/adapter-config/itd_511 using the existing
roads511Config state. The wzdx panel mirrors roads511 enabled and
feed_source since they share the same backend adapter.
Bundle: D045j2lq -> BiMKNe5L.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
defaults.py: add work_zone_enabled (bool, default false),
work_zone_min_severity (str, default Minor), work_zone_sub_types
(json, default [road_works, lane_closed, road_closed]) to itd_511.
incident_handler.py: replace hardcoded work_zone return None with
adapter_config-driven gate. Resolve sub_type and event_sev before
the work_zone check so severity and sub-type filters apply. Non-work-zone
events keep the existing min_severity / enabled_categories / enabled_sub_types
filters unchanged.
Environment.tsx: add work_zone_enabled, work_zone_min_severity,
work_zone_sub_types to Roads511Config. Load/save/discard wired. Work Zones
section in roads511 panel with enable toggle, min severity dropdown, and
sub-type checkboxes (visible only when enabled).
Bundle: KLGUZQYL -> D045j2lq.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Includes NWS broadcast filter controls from 31c464c that were missing
from the previous bundle.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
consumer.py: return None immediately for work_zone/road_closure/road_incident
categories instead of routing through format_work_zone_mesh.
incident_handler.py: add work_zone kind to _parse_itd_511_incident and return
None immediately so itd_511 work_zone events never reach change-detection.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add NwsConfig adapter config (broadcast_severities, duplicate_allowed_after_seconds)
with load/save/discard/change-detection wiring. When feed_source=central, hide
native-only fields (User Agent, Tick Seconds) and show Broadcast Filters section
with severity checkboxes and re-broadcast cooldown input.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Drop NWSheadline entirely. Build line 2 from expires_epoch (formatted
to local America/Boise time with timezone abbrev) and first areaDesc
segment (truncated to 50 chars at word boundary).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove all headline casing (title-case, sentence-case, TZ regex). Pass
the NWSheadline string through as-is from the CAP parameters, only
applying word-boundary truncation to 80 UTF-8 bytes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace .title() with manual sentence-casing: capitalize first char,
lowercase the rest, then re-uppercase timezone abbreviations (MDT, MST,
PDT, PST, CDT, CST, EDT, EST, UTC) via regex.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fix 1: wind.lower() so 60 MPH winds becomes 60 mph winds.
Fix 2: rstrip trailing period/comma/space from locations text.
Fix 3: bearing is direction storm moves TOWARD, not FROM — remove
the +180 flip and use (deg+22.5)/45 for correct compass bucketing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Cut at last space before 80 UTF-8 bytes instead of hard-slicing at 77
chars with trailing dots.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Drop the instruction line (line 6) from NWS wire output entirely.
Remove the _SAME_INSTRUCTION dict that was added in 503c16d.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Short, actionable instructions keyed by SAME event code (TOR -> "Seek
shelter immediately.", SVR -> "Move indoors now.", etc.). Falls back to
the CAP instruction field when no SAME code matches. Truncates at 40
UTF-8 bytes to keep wire size compact for mesh broadcast.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace single-line _render() with structured 6-line format:
L1: SAME emoji + event type + NWS office (from WMO identifier)
L2: area (first areaDesc segment, max 60 chars)
L3: hazard (from HAZARD.../TORNADO... or maxWindGust/maxHailSize params)
L4: impact (from IMPACT... in description)
L5: expires
L6: instruction (max 80 chars)
Add module-level helpers: _SAME_EMOJI, _NWS_OFFICE_SHORT, _nws_office(),
_parse_nws_description(). Emoji prefers SAME event code, falls back to
_emoji_for_event() substring match. All _render() call sites pass d=d.
Update test to match new format (coordinates removed from wire).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The enrichment pipeline writes to d._enriched, not d._enrichment.
Fix both _parse_state_511_incident and _parse_itd_511_incident.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ITD 511 sometimes sends lowercase direction strings (e.g. "east"
instead of "East"). Add lowercase and abbreviated lowercase keys
so the renderer resolves them without falling through to raw echo.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add min_magnitude dropdown (1-4), drop_non_present and
drop_zero_magnitude toggles to the TomTom Traffic adapter card.
State loads from /api/adapter-config/tomtom_incidents on mount
and saves changed keys on save, following the same pattern as
the WFIGS and fires config panels.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add tomtom_incidents.min_magnitude setting (default 4 = severe)
to adapter_config registry. Replace the hardcoded magnitude==0
drop check with a config-driven floor that silently drops any
TomTom event below the configured threshold.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Let road stay None when road_numbers is absent so the renderer
uses the from → to segment format instead of clobbering it with
the raw from string.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Line 2 now falls back to from_loc → to_loc when road is absent
(common for TomTom street-level incidents without road_numbers).
Line 3 renders length (meters from TomTom) as human-readable
distance (mi or m) alongside delay and lanes_affected.
Add length field to _parse_tomtom_incident return dict.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove the prefix parameter from _render() and all callers — the
New:/Update: labels are no longer surfaced in the multi-line format.
Add comment field extraction to _parse_state_511_incident and
_parse_itd_511_incident return dicts. Render comment as line 3b
when it provides additional context beyond lanes_affected and is
<=140 chars, skipping duplicates.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace single-line _render() with structured multi-line output:
Line 1: emoji + display name + city/state anchor
Line 2: road + full direction (Eastbound) + mile marker
Line 3: lanes affected + delay
Line 4: cause (if non-default)
Add _SUB_TYPE_DISPLAY and _DIRECTION_LONG mappings. Extend
_parse_state_511_incident and _parse_itd_511_incident return dicts
with lanes_affected, cause, description, and mile_marker fields.
Add mile_marker: None to _parse_tomtom_incident for consistency.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Events with severity=immediate skip the per-toggle cooldown check
entirely — they are already rate-controlled by source handler change
detection. Also set cooldown_seconds default to 0 (disabled).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
POST /api/debug/clear-cooldowns clears both in-memory toggle
cooldown map and SQLite dispatcher_cooldowns table.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Disable all standalone FIRMS broadcast paths (cluster, halt, spotting)
by inserting early return None. Growth events now use the shared WFIGS
_render() for consistent multi-line format with movement data, and set
_severity_override=immediate + _cooldown_suffix for per-fire cooldown.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Each WFIGS fire event now carries its irwin_id as _cooldown_suffix in
event.data. The dispatcher incorporates this suffix into the cooldown
key region field, giving each fire its own independent cooldown slot
instead of sharing one per (toggle, category, region).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove _build_prompt, _llm_render, _terse_fallback and all LLM backend
references. render_digest() now queries the fires table directly and
builds a structured multi-line wire: header with count, up to 5 fires
with acres/containment/anchor, and a +N more overflow line.
FireDigestScheduler no longer accepts or uses llm_backend. Updated the
pipeline __init__.py call site accordingly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fire events are already change-detected by the wfigs handler so the
grouper coalescing window adds no value and causes commit callbacks
to be lost when events are replaced. Setting severity to immediate
unconditionally bypasses the grouper entirely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Apply .title() to all town name returns in _location_anchor so
anchors render with proper capitalisation regardless of source
casing. Remove the redundant Near: prefix from the location line
in _render — the anchor text is self-describing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove hamlet, suburb, and locality from _TOWN_OSM_VALUES so the
nearest_town Photon lookup only returns meaningful population centers,
avoiding misleading anchors from tiny named places.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Acres: prepend IncidentSize to _WFIGS_ACRES_RAW_KEYS so the normalizer
picks up the primary size field before falling back to DiscoveryAcres
and FinalAcres.
Location anchor: query the curated town_anchors table before falling
back to the Photon geocoder nearest_town call, giving consistent
anchor names for Idaho fires.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Normalizer: add fire_cause, agency, personnel, unique_fire_id from
WFIGS raw payload to the normalized incident dict.
Renderer: replace single-line wire format with structured multi-line
output — header, size/contained with bold deltas on updates, location
anchor, cause/discovered date, and unique fire ID. Update call sites
pass last_bcast_acres and last_bcast_contained for case-(iii) updates
to enable delta calculation and selective bolding.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
event.data can contain callables and internal _on_ callbacks that
cause json.dumps to fail with TypeError. Filter these out before
serializing to SQLite.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
WFIGS API returns both "WF" and "wildfire" as IncidentTypeCategory
values. The previous check only accepted "WF", silently dropping
wildfire-typed incidents.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When the non-WF filter returns None, the caller must check before
assigning to n[_kind]. Fixes TypeError on non-wildfire incidents.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Fire events now carry _severity_override in the data dict:
- New fires (case i/ii): priority by default
- Update fires (case iii): priority by default
- All fires: immediate if acres > 1000 OR contained_pct == 0
consumer.py checks for _severity_override before falling back to
map_severity(inner.severity). This ensures fire broadcasts are
prioritized appropriately in the dispatch queue.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add 'Central' as first tab in FAMILIES array (before Weather)
- Import Server icon from lucide-react
- Remove Central Connection card from header area
- Render Central config in its own tab panel (no adapter sub-tabs)
The Central tab now shows URL, Durable, and Region fields with the
enabled toggle, matching the previous inline card behavior.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rework the NIFC Fire Perimeters tab to show all fire config in one place:
- Native polling settings (tick_seconds, state) shown only when native
- Broadcast Settings section (always shown):
- Broadcast on acres increase toggle
- Broadcast on containment increase toggle
- Update cooldown (hours, converted from cooldown_seconds)
- Freshness window (hours, 0 = disabled)
- Daily Digest section (always shown):
- Digest enabled toggle
- Schedule times (HH:MM list)
Changes persist to adapter_config table via PUT /api/adapter-config/{adapter}/{key}.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Fire events are always relevant regardless of age (a wildfire burning
for 9 hours is not stale — it's ongoing). The staleness filter was
designed for incidents with time_validity semantics, not persistent
fire state.
- defaults.py: add wfigs.freshness_seconds = 0 (disabled)
- dispatcher.py: for fire toggle family, read from adapter_config
instead of toggle; skip staleness check when freshness_s == 0
Fixes Blue Ridge fire being dropped after LAST_PER_SUBJECT replay.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- consumer.py: change DeliverPolicy from NEW to LAST_PER_SUBJECT to
get latest state per subject on reconnect instead of replaying backlog
- central_normalizer.py: drop RX and non-wildfire (non-WF) incident
types early in _parse_wfigs_incidents before they reach the handler
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
All three approved tiers in one commit. Reference.tsx is the deep docs
hub (8 new sections); the 10 other pages get short helper text +
tooltips that cross-reference back into Reference; 3 components get
operational-context tooltips. No new features land here -- this is the
copy that catches the GUI up to v0.6 + v0.7 system behavior.
Decisions applied per Matt's call:
- Keep both bang commands AND the LLM DM path (bangs are short on a
mesh-constrained interface; LLM is the anything-else path). Cross-
references between the two land in Reference -> Commands and
Reference -> LLM DM Queries.
- Rename "wire-string rendering" to "broadcast text" in user-facing
copy on TownAnchors.tsx, GaugeSites.tsx, and the Curation section of
Reference.tsx.
- Keep the "AND-model anti-pattern" tooltip as-is on Environment.tsx +
GaugeSites.tsx (specificity is the value for advanced users); the
OR-not-AND Reference section is its home definition that other
tooltips can link to.
Ham terminology preserved:
- Reference.tsx solar/Kp section retains "Quiet sun" / "Quiet HF
conditions" language (SFI/Kp vocabulary, not the deleted Quiet Hours
feature -- confirmed via direct grep before writing).
Tier 1: Reference.tsx (the depth doc) -- 8 new sections, ordered for
readability:
- "Fire Tracker (Fusion)": Phases 1-4 unified. Six fire-family alert
categories with example wire strings (wildfire_declared,
wildfire_growth, wildfire_halted, wildfire_spotting,
unattributed_hotspot_cluster, wildfire_incident). Attribution
mechanics (spread_radius_mi default, centroid as 24h median).
Movement mechanics (pass_id bucketing, per-pass centroid, 8-way
bearing, mi/h drift). Spotting mechanics (convex-hull perimeter +
vertex-distance approximation + per-fire cooldown). Daily LLM digest
(twice-daily summary broadcaster). The 10 fires.* adapter_config
knobs with defaults.
- "Broadcast Types": the three prefix categories -- New: (first sight),
Update: (material change), Active: (clock-driven reminder).
- "Reminder System": cadences per adapter (WFIGS 8h, SWPC 8h, ITD 511
per-zone). The tombstone (fires.tombstoned_at) termination. The
per-adapter reminder_enabled flag.
- "LLM DM (Natural-Language Queries)": all 7 env_reporter adapter
blocks (build_fires_detail / build_alerts_detail / build_quakes_detail
/ build_traffic_detail / build_gauges_detail / build_swpc_detail /
build_drop_audit) with example questions that hit each one. The
grounding clause behavior ("No active X right now" when an adapter
block is empty -- the v0.7-fire-tracker-4-final clamp). The
include_in_llm_context per-adapter toggle.
- "OR-not-AND Architecture": the per-adapter Central vs native
contract. Mutually exclusive. The AND-mode anti-pattern definition
(referenced by the Environment + GaugeSites tooltips). The Spokane
fix context.
- "Adapter Config & the CODE Rule": the GUI knob hub. The CONFIG-vs-
CODE split (thresholds in CONFIG, sentence templates / emoji /
translation maps in CODE). Restart-required vs live keys. The
include_in_llm_context toggle.
- "Curation: Gauges & Towns": Gauge Sites (NWS-AHPS thresholds, USGS
lookup, Action/Minor/Moderate/Major). Town Anchors (broadcast text
suffix lookup chain: Photon -> this table -> landclass -> county
-> coords). Example output "3 mi N of Almo".
- "Schema Migrations": light touch. v11-v16 schema additions tagged
with the phase they shipped under.
Tier 2: per-page tooltips and cross-references (10 pages):
- AdapterConfig.tsx: header paragraph extended with the CODE rule
pointer + LLM context toggle explanation.
- Alerts.tsx: !subscribe blurb extended with the three broadcast types
and links to Reference -> Broadcast Types + Reminder System.
- Config.tsx: environmental section description updated to point at
Environment.tsx for adapter knobs + Reference -> OR-not-AND for the
architecture.
- Dashboard.tsx: RF Propagation title carries SWPC R/S/G + Kp legend
tooltip; LOCAL badge defines what counts as local.
- Environment.tsx: Central region-token helper now references the
OR-not-AND section; tick_seconds defined inline as the native-mode
poll interval.
- GaugeSites.tsx: page description rewritten -- replaces "envelope
time" jargon with operational language, explains USGS lookup
mechanics, points at Reference -> OR-not-AND for the central-feed
disable.
- Mesh.tsx: Topology + Geographic buttons get tooltips defining the
rendering model.
- Notifications.tsx: band-conditions block extended with the daily
fire digest pointer + Reference -> Fire Tracker + Broadcast Types
cross-refs.
- TownAnchors.tsx: page description rewritten -- "wire-string
rendering" -> "broadcast text", chain fallback explained ("Photon
-> this table -> landclass -> county/state -> coords"), example
output included.
Tier 3: component tooltip polish (3 components):
- NodeTable.tsx: Battery + Last Heard column headers get title-bearing
spans with the voltage chart + offline-threshold legend.
- NodeDetail.tsx: SNR quality bands documented as a comment in the
neighbor render block (the legend lives next to where the colored
quality dots are computed).
- RestartBanner.tsx: banner copy extended with the restart-required
catalog (Config -> environmental, LLM backend swap, dispatcher
cold-start grace) so operators know what touched it.
Build verification:
- tsc + vite build green (one warning about chunk size > 500kB --
pre-existing).
- All 8 new TOPICS ids resolve in the served bundle:
adapter-config, broadcast-types, curation, fire-tracker,
llm-dm, or-not-and, reminders, schema.
- Distinctive new strings present in the bundle ("3 mi N of Almo",
"Photon nearest-town", "AND-mode anti-pattern", "R (Radio Blackouts").
- "Quiet sun" preserved (the ham SFI/Kp vocabulary in the Solar
section, not the deleted Quiet Hours feature).
- Container Up healthy, 0 tracebacks in 2 min post-rebuild.
Changelog: v0.7-docs-rewrite.md (per-page strip / rewrite / add table).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Follow-up to v0.7-fire-tracker-4-revised. 7-path verify identified 4
paths failing for 2 root causes: (A) _ENV_KEYWORDS_TO_SUBTYPE missing
"traffic" + drop-audit phrases so build_traffic_detail and
build_drop_audit never ran; (B) LLM hallucinated specific numbers when
env blocks were empty (fabricated 144 earthquakes from a blank
quake_events table). This commit widens the keyword catch (phrase-match
for multi-word triggers to avoid false positives) and adds a positive-
framed anti-hallucination instruction to the env-context system prompt.
Re-verified all 7 paths against real Gemini in the prod container;
verdicts in v0.7-firetracker-phase4.md.
Class A -- routing miss fix:
- _ENV_KEYWORDS_TO_SUBTYPE gains "traffic"/"commute"/"highway"
mapped to the existing "traffic" subtype.
- New _ENV_PHRASES_TO_SUBTYPE dict for multi-word triggers, matched
as whole-phrase substrings (NOT single-word membership). Drop-audit
phrases: "why didn't"/"why didnt"/"why am i not"/"why am i missing"
/"what was filtered"/"drop audit"/"filtered out" all map to a new
"drop_audit" subtype. Phrase-match keeps "why" alone from
false-positing every "why is X" question.
- _detect_env_subtype now checks phrases first, then falls back to
single-word tokenized match.
Class B -- positive-framed anti-hallucination clause:
- New module-level ENV_GROUNDING_CLAUSE constant. Appended to the
system prompt whenever env scope is detected (after env_block +
drop_block injection).
- Per Matt's mitigation guidance: positive ("answer from the blocks")
not negative ("do not hallucinate"). Wording:
"ENVIRONMENTAL CONTEXT GROUNDING:
Answer only from the environmental context blocks above. If a
block is empty or missing for an adapter the user asked about
(e.g. no NWS alerts in the block), say something like 'No active
<category> right now' -- never invent specific numbers, place
names, or counts. If you do not have a relevant block for the
question, say so briefly."
7-path verification, post-fix (real Gemini, prod container):
| # | query | method | verdict |
|---|-------------------------------------------------|----------------------|---------|
| 1 | "are there any fires near me?" | build_fires_detail | PASS |
| 2 | "any weather alerts?" | build_alerts_detail | PASS |
| 3 | "any earthquakes nearby?" | build_quakes_detail | PASS |
| 4 | "how's traffic on I-84?" | build_traffic_detail | PASS |
| 5 | "what's the snake river level?" | build_gauges_detail | PASS |
| 6 | "what are the band conditions?" | build_swpc_detail | PASS |
| 7 | "why didn't I hear about anything today?" | build_drop_audit | PASS |
Hallucination evidence (pre vs post on the quakes path):
pre-fix: "There have been 144 earthquakes of magnitude 1.5 or
greater in the past 24 hours worldwide. Some of the most
recent earthquakes reported include: A magnitude 2.1
earthquake in Pahala, Hawaii..." (fabricated)
post-fix: "I haven't observed any information about earthquakes
in the mesh data." (grounded)
Routing-miss evidence (pre vs post on the traffic path):
pre-fix: _detect_env_subtype("how's traffic on I-84?") -> None
-> env scope NOT triggered, build_traffic_detail never
called, LLM fabricated I-84 conditions in OR
post-fix: _detect_env_subtype("how's traffic on I-84?") -> "traffic"
-> env scope triggers, build_traffic_detail returns
185 chars of real Ada-county incident data, LLM
grounds on it: "I haven't observed any active
traffic incidents on I-84 within the last two hours.
The current active incidents are on North 9th Street
/ South 9th Street and SH-21, both in Ada."
Tests:
- 7 phase4 tests pass (no new tests needed; verification is the LLM
DM path itself).
- Full suite: 56 passed in 3.80s across phase1+phase2+phase3+phase4
+or-arch+include-roundtrip.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Matt review caught a scope error: ?status was a hypothetical sketch
in the design doc ("a node could ping ?status cache peak") treated as
authorization without asking. Ripping the structured-command path
entirely. The LLM DM path with env_reporter injection is the natural-
language interface; ?status was redundant infrastructure parallel to
the path the design depends on.
What landed:
- router.py: _maybe_rewrite_status_query + _lookup_fire_fuzzy +
_build_fire_status_context removed. route() restored to:
bang -> IGNORE-empty -> LLM with verbatim query.
- tests/test_fire_tracker_phase4.py: 5 ?status tests removed; replaced
with two regression guards:
test_natural_language_fire_question_routes_to_llm -- "how's the
cache peak fire?" returns RouteType.LLM with the verbatim query
(no in-router rewriting).
test_status_helpers_removed_from_router -- hard-block on
_maybe_rewrite_status_query / _lookup_fire_fuzzy / "?status"
appearing anywhere in router.py source. If anyone adds a
structured-command path for fires, this test fails and the
author has to talk to Matt first.
- 56 passed in 3.80s across phase1+phase2+phase3+phase4+or-arch+
include-roundtrip.
What stays (NOT ripped):
- Daily fire digest -- scheduled broadcaster, not a command. Its 4
adapter_config rows (fires.digest_enabled / digest_schedule /
digest_timezone / digest_max_chars) stay GUI-editable.
- Bug A fix (UnboundLocalError at router.py:745) -- independent of
?status. Confirmed still in effect.
LLM DM 7-path verification result -- 3 of 7 pass, INCOMPLETE:
| # | query | env_reporter | verdict |
|---|-----------------------------------------------|----------------------|---------|
| 1 | "are there any fires near me?" | build_fires_detail | PASS |
| 2 | "any weather alerts?" | build_alerts_detail | FAIL |
| 3 | "any earthquakes nearby?" | build_quakes_detail | FAIL |
| 4 | "how's traffic on I-84?" | build_traffic_detail | FAIL |
| 5 | "what's the snake river level?" | build_gauges_detail | PASS |
| 6 | "what are the band conditions?" | build_swpc_detail | PASS |
| 7 | "why didn't I hear about anything today?" | build_drop_audit | FAIL |
Two distinct failure classes:
Class A -- routing miss (#4 traffic, #7 drop):
_ENV_KEYWORDS_TO_SUBTYPE lacks "traffic" (only road/jam/crash/
closure/511/incident map to "traffic"), so a query literally
mentioning "traffic" never triggers env scope -> build_traffic_detail
never runs even though traffic_events has 9 rows on disk. The LLM
fell back to training data and hallucinated I-84 conditions.
build_drop_audit has no natural-language trigger phrase at all;
"why didn't I hear about anything today?" has no env keyword.
Class B -- empty data + LLM hallucination (#2 alerts, #3 quakes):
Env scope IS detected, build_alerts_detail and build_quakes_detail
DO run, but return empty because nws_alerts has 0 rows and
quake_events 24h-window has 0 rows (legitimate empty state). The
LLM has no env block to ground on and hallucinated "144 earthquakes
worldwide" -- sounds authoritative, is fabricated.
Not fixed in this commit -- needs Matt's call on:
(a) keyword additions to _ENV_KEYWORDS_TO_SUBTYPE for traffic +
drop_audit triggers (risk: false-positive env-scope triggers
for unrelated phrases).
(b) anti-hallucination prompt clamp: "If a topic's env block is
missing/empty, say you don't have live data instead of
answering from general knowledge." (risk: bot apologizes
every other message.)
Per the "STOP if any path fails" instruction, this commit does NOT
claim verification done; the report at
v0.7-firetracker-phase4.md has the full table + per-row mesh-receiver
wire + per-failure root cause analysis.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 4 of FIRMS+WFIGS fusion. Foundation: every direct LLM DM
mentioning a fire/weather/quake/avalanche/flood/etc. keyword was
failing silently in prod with UnboundLocalError because router.py
referenced scope_type before assigning it. With that path restored,
two new features land: a twice-daily fire-digest scheduled broadcast
(LLM-rendered) and a ?status <fire_name> on-demand mesh-DM intent.
BUG-FIX ROOT CAUSE (Job Zero):
router.py:745 ("if should_inject_mesh and scope_type == 'env'") read
`scope_type` -- a local variable bound only at line 761 inside an
unrelated `if self.source_manager and self.mesh_reporter` block.
Python's lexical scoping made scope_type a local of the whole
generate_llm_response function, so reading it before the assignment
raised UnboundLocalError on every env-keyword DM. The exception
propagated to main.py's outer except, no response went out, bot
appeared dead on fire/weather/quake/avalanche/flood queries.
Evidence (synthetic in-process trace against the live container's
config + GoogleBackend):
"are there any fires near me?" -> UnboundLocalError (pre-fix)
-> real LLM answer (post-fix)
"Yes, there are a few active
fires reported in the region.
Salmon River: 4,200 acres, 78%
contained. Cache Peak: 1,847
acres, 23% contained. ..."
"what's the weather?" -> UnboundLocalError (pre-fix)
-> "I do not have current weather
information. I can tell you
about active fires, stream gauge
levels, space weather, or band
conditions if you'd like." (post-fix)
"hi there" -> normal LLM answer in both cases
Fix: hoist `scope_type, scope_value = self._detect_mesh_scope(query)`
to right after `should_inject_mesh` is computed; remove the
now-duplicate detection inside the source_manager block.
Secondary mitigation: tightened the "do not invent commands" prompt
with an explicit "if no list appears above, you have NO commands"
clause. The prior prompt told the LLM "answer based on the command
list provided below" without always providing one, so the LLM
hallucinated plausible-sounding !commands (the "use ! commands"
canned-looking response Matt was seeing on non-env queries).
PHASE 4 FEATURES:
1. Fire-digest scheduler (meshai/notifications/scheduled/fire_digest.py).
Modeled after BandConditionsScheduler. Runs in the pipeline's
start_pipeline coroutine alongside band_conditions + reminders.
On each slot (default 06:00 + 18:00 America/Boise):
- Queries active fires (tombstoned_at IS NULL) + last 24h passes.
- Builds a prompt asking for a single mesh-wire summary <= 200
chars.
- Calls the LLM (Google/Anthropic/OpenAI per config).
- Falls back to a terse "Fires today (N): Cache Peak 1847 ac;
Twin Peaks 320 ac; +N more" line when the LLM is unavailable.
- Dispatches via dispatcher.dispatch_scheduled_broadcast (same
path band_conditions uses).
Idempotency: v16.sql adds fire_digest_broadcasts(slot_epoch PK,
sent_at, summary, source). INSERT OR IGNORE pattern blocks the same
slot firing twice (matters when container restarts mid-day).
2. ?status <fire_name> on-demand intent (router.py).
Before falling through to the LLM, route() now checks for a leading
"?status" / "status:" sigil or natural-language triggers like
"how is X fire?". On match:
- _lookup_fire_fuzzy walks fires by exact -> startswith ->
contains -> word-overlap (skipping a trailing " fire" word so
"cache peak fire" matches "Cache Peak"). Active fires rank
above tombstoned ones.
- _build_fire_status_context composes a small context block
(name, acres, containment, county/state, last 3 passes with
drift).
- The query is REWRITTEN into an LLM prompt with that context
inlined; the rest of the normal LLM path (chunking, history,
summary persistence) runs unchanged.
Live verification: "?status Cache Peak" -> "The Cache Peak fire is
1,847 acres and 23% contained. It's located in Probe / ID.";
"?status Salmon" -> word-overlap matches "Salmon River" ->
"The Salmon River fire is 4,200 acres and 78% contained, located
in Probe / ID."
3. adapter_config rows (GUI-editable per CONFIG-vs-CODE rule):
fires.digest_enabled = true (master toggle)
fires.digest_schedule = ["06:00", "18:00"]
fires.digest_timezone = "America/Boise"
fires.digest_max_chars = 200
Schema (v16.sql):
- fire_digest_broadcasts(slot_epoch INTEGER PK, sent_at, summary,
source) with source in {'llm', 'fallback_terse', 'skipped_no_fires'}.
- Index on sent_at for ops queries.
Tests (tests/test_fire_tracker_phase4.py, 10 cases all green):
- Regression guard: scope_type appears as an assignment BEFORE the
env_reporter check (prevents the UnboundLocalError from coming back).
- adapter_config seeds all 4 digest keys with expected defaults.
- render_digest returns ('', 'no_fires') when no active fires.
- render_digest falls back to terse line when LLM is None; wire fits cap.
- render_digest with a stub LLM returns ('<llm text>', 'llm').
- _lookup_fire_fuzzy: exact, "X fire" trim, word-overlap, no-match.
- _maybe_rewrite_status_query: builds context-bearing prompt; returns
None on non-status queries.
Combined suite: 60 passed in 3.81s across phase1+phase2+phase3+phase4
+or-arch+include-roundtrip.
Live verification on CT108 after rebuild:
- v16 migration applied (schema_meta=16, no Traceback in 3 min).
- FireDigestScheduler started: enabled=True schedule=['06:00','18:00']
tz=America/Boise.
- LLM DM probe (real Gemini) returns real answers on env queries
(Bug A fixed end-to-end).
- ?status Cache Peak + ?status Salmon return fire-specific summaries.
- render_digest with real LLM returns source=llm + non-empty wire.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
