Portfolio trajectory
reported crash volume
Executive operating picture
Crash portfolio command center
Reported-event intelligence for territorial strategy, prevention, claims operations, and actuarial research.
→
CONFIRMED · CCRS-derived descriptive evidence
INFERRED · modeled signal, explicitly labeled
UNKNOWN · no policy exposure or paid-loss denominator
2026 year-end outlook
seasonally scaled · inferred
Spatial concentration
0.1° crash-density cells · all years
LOW ● DENSITY ● HIGH
Color = injury involvement
Color = injury involvement
Decision brief
highest-signal findings
Territorial intelligence
Risk geography
Credibility-adjusted injury involvement and crash concentration. Volume is not exposure-adjusted claim frequency.
California event surface
all reported crashes with valid coordinates
Cell size = crash volume
Color = injury involvement
Color = injury involvement
Territorial severity × volume
top 100 cities · smoothed risk
Interpretation guardrail: risk index compares injury involvement among reported crashes. It does not measure per-policy or per-mile frequency.
Territory ledger
selected analysis window
Prevention + claims operations
Loss-driver observatory
When, how, and under what conditions reported crashes become injury-involved.
Weekly risk clock
volume intensity by day × hour · all years
Collision mechanism
share + injury involvement
Primary violation codes
unique crashes · canonical CVC section
Lighting severity
injury involvement rate
Occupant protection
injured-person records
At-fault composition by driver age
descriptive · not a pricing recommendation
Vehicle age signal
at-fault share among known vehicle years
Transparent predictive research
Scene severity prediction lab
Out-of-time validation for injury-or-fatal involvement. No protected attributes. Not an approved rating model.
Scenario lab
empirical Bayesian model
Holdout calibration
trained 2016–2023 · untouched 2024 test
Model card
intended use + limitations
Target: crash has ≥1 injured or killed person.
Features: city, hour, day, weather, lighting, collision mechanism.
Method: credibility-smoothed empirical rates combined with fixed, transparent weights.
Use: portfolio segmentation research and post-event triage.
Do not use: automated adverse action, individual pricing, causal attribution, or loss-cost projection.
Features: city, hour, day, weather, lighting, collision mechanism.
Method: credibility-smoothed empirical rates combined with fixed, transparent weights.
Use: portfolio segmentation research and post-event triage.
Do not use: automated adverse action, individual pricing, causal attribution, or loss-cost projection.
Commercial data gap
what converts signal into pricing value
Required next: geocoded policy exposure, vehicle-years, earned premium, claim counts, paid/incurred loss, limits/deductibles, and weather/road exposure. Join under privacy controls, validate chronologically, then test lift against the carrier baseline.