[Idea] Palantir Phase 2: Risk heatmap — incident and anomaly density over the session

[phassle]

This was generated by AI during triage.

Summary

A toggleable heat overlay weighting map cells by accumulated anomalies and incident-minutes over the session — rendering trouble, not traffic.

Motivation

The map shows where vehicles are; the inbox shows discrete incidents. Neither answers where is today's session concentrating trouble? A corridor with many small resolved incidents becomes invisible as each row resolves. Risk heatmap is explicitly out of scope in PRD #84; this picks it up.

Proposed behaviour

• Heat layer over grid-bucketed cells: weight = anomaly count + incident-minutes, with exponential time decay so resolved incidents still contribute but fade. Cell size and half-life are named constants calibrated against the live feed (PRD: Transit Command Center — Incident Inbox, anomaly detection, nearby webcams, replay (MVP) #84 threshold pattern).
• Feed-outage incidents (operator subject, no geometry per PRD: Transit Command Center — Incident Inbox, anomaly detection, nearby webcams, replay (MVP) #84 story 12) are excluded from ground heat; they tint the operator's bounding box from the operator registry instead.
• Clicking a hot cell lists the incidents/anomalies that heated it — heat is auditable evidence, not decoration.
• Pure service (anomalies, incidents, now) → weightedCells[] over the PRD: Transit Command Center — Incident Inbox, anomaly detection, nearby webcams, replay (MVP) #84 buffer; Leaflet.heat or hand-rolled canvas (preferCanvas already on). Zero new feed calls. Requires PRD: Transit Command Center — Incident Inbox, anomaly detection, nearby webcams, replay (MVP) #84.

Area

Map / markers

What is out of scope (optional)

• Cross-session/historical risk (no persistence, ADR 0003).
• Predictive risk (separate idea: ETA degradation).
• Severity weighting beyond count and duration.