The method
Watch follows reviewed public authorities and your saved plans, then tells you what changed, why it matched, how old the observation is, and where to confirm it. A notice is decision support—not a license, regulation, access guarantee, or animal forecast.
Every connector is a reviewed registry entry with a fixed government URL, capability, cadence, parser version, invariants, and its own fresh/stale limits. Its first valid run establishes a quiet baseline. Later canonical changes create immutable events. A failed parser never replaces last-good truth, and a stale source emits no availability notice. Contradictory authorities are shown separately rather than numerically blended.
See the live registry and each source’s last run on the Watch coverage page.
A notice stores the immutable source event, the exact rule and selectors that matched, spatial distance where relevant, formula version, and render version. Current/expired status is evaluated when you open it. Retractions and corrections are new events; history is never rewritten. Legal closures require an identifiable official order. Fire perimeters say only that an active fire was mapped nearby unless a separate legal closure authority says land is closed.
Conditions rules use the same open conditions formula as the map, with a five-point hysteresis band and a 24-hour cooldown. Daily briefs group spots instead of sending one alert per point. Quiet hours and local-time digests are evaluated at delivery. Inbox is authoritative; email and push are at-least-once external delivery rails with idempotency, rate ceilings, receipts, and circuit breakers.
History never changes the public conditions score and can never trigger a notice. It appears only after at least five schema-valid, completed Hunt Records within 1 km of the watched point. Each past hunt is compared with the current window across available dimensions; missing dimensions drop out and the remaining values are averaged. At least three comparable dimensions are required.
Each component is capped at 1. A mean distance ≤0.35 is “similar.” The notice exposes the completed count, comparable count, matched record IDs, and method version.
“Saw,” “shot,” and “killed” each count once as “saw game”; kills receive no extra weight. No population model is trained, no spot ranking is published, and no sentence claims animal behavior. The required caveat is: Small sample—personal context, not a prediction.
Open the official link, confirm current availability or the controlling order, check regulations and access, then make your own decision. For weather, review warnings and the factor breakdown rather than treating a composite score as a safety clearance.