The first screen says what it knows — and why it can't say more.
- A persistent notice bar. "Not the lifeline status of record — leadership validates lifeline status in the EOC's own system." Every export repeats it.
- Nothing seeded. Every lifeline opens not assessed this period — a 12-hour operational period from 07:00 Pacific.
- Energy's evidence, with its age. 9,066 customers without power (unplanned) · 0.236 per 1,000 · rec 17:26Z · fetched 17:37Z · publishes every 15 min (measured).
- And why no colour. Partial coverage: 40 of 58 counties include electric territories not in the outage feed. No suggestion.
- Exposure, with denominators. 0 of 454 open acute hospitals · 4 of 2,082 operating plants · 2 of 931 CAL FIRE stations inside PG&E outage areas.
- The feed bar. 16 of 16 feeds listed · 11 bound · 5 gated or not browser-readable — each gap with its citation.
Energy · Power Grid, against a baseline — and a coverage test.
≥ 25 Significant · ≥ 5 Moderate · < 5 Minimal — only if the four reporting utilities cover the county
Calaveras 134 ÷ 44,650 × 1,000 = 3.001 · coverage full → suggested Green — Minimal
Fresno 0.541 per 1,000 · PWRPA covers 14.32 % outside the feed → no suggestion
California 0.028 per 1,000 · always partial → no suggestion
The thresholds are this app's, not FEMA's, and every screen and export says so. Coverage is an equal-area share with a 0.5 % floor, built from the utility territories: 18 of 58 counties are fully covered — the only places Green can ever be suggested.
Measured, then corrected by the build: LADWP covers 11.54 % of Los Angeles County — not the probe's 11.83 %, which counted its hole rings as covered ground. Solano / City of Pittsburg sits at 0.498 %, within 0.002 pt of the floor, so near-floor shares are re-measured on a finer grid. Worked figures: data of 2026-10-21 12:56Z.
A feed can only suggest. The branch recommends.
- Agreement is printed on the row. agrees · overrides · no suggestion possible — with who and when.
- An override needs a reason. Without one it is refused; with one it is saved as overrides.
- The app never changes the person's colour. If the feed moves later the row reads suggestion changed since recommended — review.
- The estimated time is the person's. Utility restore times are printed as the utility's estimate — or restore time passed, no estimate published — never as the toolkit's ETA.
- It leaves as an export. A lifeline status briefing in the Senior Leadership Brief's Tier I / Tier II grid, and a status JSON for the EOC's system to ingest.
"Live" means polling at a measured cadence.
data.ca.gov says the outage feed updates every 15 minutes. A two-read probe in the morning said 30 — reads at 11:09 and 11:30 could not see an 11:11 publish. Consecutive changes in the afternoon settled it:
every response carries Cache-Control: max-age=300 5-min CDN cache
withdraw at 40 min = 2 × 15 + 5 cache + 5 slack
A failed refresh keeps the last good data and degrades only the freshness. During the build one geometry answer arrived as HTTP 200 with ~9.4 MB of truncated JSON — now treated as a failed refresh, never as data.
Stated honestly: the suite's own cadence check (CADENCE=1, two reads ≥ 16 min apart) was not run for this build — the 15-minute figure rests on the probe reads above.
Click an outage. Read a cascade that says it is a guess.
Typed by CISA's dependency kinds. Hop 1: assets inside the outage area, and water systems whose service area it overlaps — tagged large ≥ 25 % · partial 5–25 % · touches < 5 %. Hop 2: hospitals inside a large or partial system, inferred.
So touches hospitals are counted, not listed, and no population is ever summed — the naive sum across that morning's overlapping systems was 2,837,834. One system can arrive as several polygons, so systems are grouped by their number, never counted by polygon.