Reliability

How well each endpoint actually works

Success rates over the last 14 days, taken from terminal job outcomes. Every number carries the sample it came from, because a rate without a denominator is a decoration.

What these numbers are

Rates are terminal job outcomes over the window, split by caller. Internal traffic is one first-party workload and its mix moves independently of service quality; the reliability corpus in the repository is the mix-independent measurement.

Rates are withheld below 50 requests. “100% success over three requests” is not a reliability claim, and publishing it as one would make this page least trustworthy exactly where it matters most.

Terminal outcomes over 14 days, generated Fri, 20 Nov 2026 15:43:49 GMT
EndpointTierTargetCustomer trafficSuccessKeeping itFirst-party trafficSuccessp50p95
hn_postga99.0%0—not enough traffic0———
hn_searchga99.0%0—not enough traffic0———
infraga99.0%21,21596.1%no0—3.3 s37.1 s
linkedin_employeesga99.0%3,513100.0%yes0—22.5 s133.0 s
seoga99.0%44,23598.0%no0—9.0 s35.3 s
mentionsbeta75.0%0—not enough traffic0———
researchbeta75.0%7—not enough traffic0———
rssbeta75.0%0—not enough traffic0———
searchbeta75.0%8,87194.4%yes0—1.5 s2.7 s
sitemapbeta75.0%0—not enough traffic0———
urlbeta75.0%39,38191.3%yes0—1.2 s4.0 s
adsexperimental—125,04196.4%no target0—1.4 s39.3 s
brandexperimental—4—no target0———
channelexperimental—78,76079.0%no target0—2.3 s23.1 s
companyexperimental—37,18096.6%no target0—56.0 s130.8 s
linkedin_person_postsexperimental—0—no target0———
linkedin_postexperimental—0—no target0———
linkedin_profileexperimental—0—no target0———
producthuntexperimental—4,1312.6%no target0—10.1 s10.1 s
reddit_postexperimental—0—no target0———
reddit_profileexperimental—0—no target0———
siteexperimental—0—no target0———
social_engagementexperimental—0—no target0———
social_mentionsexperimental—0—no target0———
x_postexperimental—52100.0%no target0—1.3 s2.1 s
x_profileexperimental—413,67795.1%no target0—3.1 s63.2 s

What the tiers mean

ga
Succeeds at least 99% of the time over a fortnight. Safe to build on without a fallback path.
beta
Succeeds at least 75% of the time over a fortnight. Usable, but handle failures — and infrastructure-class failures are refunded.
experimental
No reliability commitment. Available, not recommended for anything load-bearing, and absent from the SDKs and the MCP tool list.

gateVerdict compares external traffic against the tier's published SLO. It reports whether a label is currently being kept; it does not promote or demote anything. Tier changes are decided on the versioned corpus, because a traffic-based rate follows the caller's mix as much as the service.