Checkpoint 3.1: coverage is not success

GET stats answers 'did we look' and 'what did we see' as separate numbers — 24h of misses is 0% coverage with undefined success, never 'down by silence'.

at a glance

Goal: Two honest numbers: expected slots vs stored evidence — a monitor that stored nothing can't claim 100% of anything.
Built: GET /api/monitors/{id}/stats: expected slots from wall-clock × interval (bounded ?window_s=), observed/missed/passed from stored rows, ratios in basis points, null for undefined denominators; the detail page renders both.
Result: CoverageStats domain tests + check_stats store test + stats_separates_coverage_from_success API test · live: flaky → 0% success, keeper → 100% · 24h of misses → 0% coverage, never 'down by silence'
Code: akira-shirase-oss/monitoring-service · main @ 0b57968

M3 is display — and the first honest number. Four steps.

What changed: GET /api/monitors/{id}/stats answers did we look and what did we see as separate numbers. Expected slots come from wall-clock × interval — not from stored rows, which is the whole point: a monitor that stored nothing can’t claim 100% coverage of nothing. Missed slots never enter the success denominator, and an undefined denominator returns null, which the dashboard renders as “no data” — not a guessed percent.

Proof: domain tests, the check_stats store test, and the API test stats_separates_coverage_from_success. Live: a flaky target showed 0% success while a healthy one showed 100% — and a window of misses reports 0% coverage with undefined success, never “down by silence”.