How we score.

Trust requires showing the math. Scores apply only to APIs on our probe schedule — catalogued imports show no score until the first check lands.

The health score

60%Uptime · 90d

20%Latency · vs category

20%Shape stability

health = round(
  ( uptime_90d          × 0.60   // fraction of checks OK
  + latency_factor      × 0.20   // p50 vs category median
  + shape_stability     × 0.20   // schema-drift penalty
  ) × 100
)

Checks hit real documented endpoints — never bare homepages. Failures are classified (timeout, DNS, 4xx, 5xx, bot-challenge); endpoints our infrastructure cannot verify are marked unverifiable, never “down.”

Lifecycle · grace before judgment

DEGRADED3+ consecutive failed checks

DYING7 days of elevated error rate

DEAD30 days at ~100% failure → archived in the graveyard

RESURRECTEDa dead API that came back — it happens

Etiquette

Our checker identifies itself as shipapisbot/1.0 with a contact URL, honors 429 and Retry-After, never load-tests, and any API owner can adjust metadata, dispute a score, or opt out entirely — opted-out APIs show UNMONITORED, never a fabricated number.