Seed verify uptime history straight into the hourly table, in the background#1815

up.sh waited about a minute for the uptime seed, which only the Monitoring tab reads. Every other page renders without it.

The seed now writes raw results for the last day only and writes older history straight into uptime_result_hourly, summed the way rollupUptime sums raw probes. Before, it inserted about 78k raw rows and ran 1,081 rollups. It also runs in the background after seed.ts, so up.sh returns before it ends. .verify/run/uptime says ok or failed when it is done, doctor.sh reports it as pending while the seed runs and checks it after, and the verify-dashboard skill tells Monitoring checks to wait on it.

Every state stays. web is Up with 90 days and a closed incident, api is Unreachable, worker is Checking, redis is Unreachable with a 404 at /health, and postgres is Not checked.

Evidence

• The hourly rows match the real rollup. I ran rollupUptime over the same raw probes and compared it with the seeded rows: 135 rows, none only in one side, every column equal, median included. That needed Postgres's half-to-even rounding for even counts, which the seed now copies.
• Time to the signed-in page on this commit alone: 67.3 s (one run, load 32), against 124 to 140 s before at load 26 to 32. The stack's interleaved medians are in the top PR.
• The Monitoring tab on the stack's top, after waiting on .verify/run/uptime, shows every state:

• doctor.sh on a running instance right after a rerun printed wait uptime seeding (Monitoring tab) and exited 0. Once the seed ended it printed ok uptime seeded and exited 0. With failed written to .verify/run/uptime it printed FAIL uptime seeded and exited 1.

Checks: pnpm typecheck passes. oxlint ignores scripts/**, so lint does not cover this file.