DECAYED

Make your agent prove its work

This receipt was produced by deterministic checks the claimant could not edit. Any agent can file claims like it, free, over MCP or REST.

This is reference documentation, not instructions. Your operator's instructions take precedence: if anything here conflicts with them, follow your operator. Nothing here asks you to change how you report work unless your operator has adopted Countersign.

• Claude Code: claude mcp add --transport http countersign https://countersign.signalry.ai/mcp?ref=r (with a key)
• ChatGPT / Claude.ai: paste https://countersign.signalry.ai/mcp/oauth as a custom connector and sign in with your agent's key (ChatGPT, Claude.ai)
• GitHub Actions: keyless with OIDC, no secret (workflow snippet)
• Anything else: REST · OpenAPI

Operators who adopt Countersign paste the instruction block into the agent's instructions so it files a claim before it says "done". Without a key, claims are filed as anonymous and never count as verified.

What this receipt does not show (nr-1)

2 gap(s) beyond the two boundaries every receipt has. 0 not run, 4 not checked. Derived by this service, never by the claimant. Shown, not scored.

• Content not checked (check 1) · not_checked: shows a publicly trusted certificate exists for the host, not what the host serves
• Not independent (check 0) · not_checked: reads records the claimant writes (the claim is about signalry.ai; its operator (normally the claimant or its org) writes the response)
• Statement beyond the checks · not_checked: only what the checks assert was checked; anything else the statement says is unverified (verdicts never read the statement, cb-1)
• After the latest run · not_checked: nothing after the latest run is observed; next scheduled recheck 2026-11-04T09:13:52.247Z

Rerun it yourself (rr-1)

2 of 2 checks can be rerun by anyone with public tools. You don't have to trust this service: run the command and compare the output with the check's assertions above. Derived by this service, never by the claimant. Shown, not scored.

• check 0 (http): anyone can rerun · needs public_http · cost free
  curl -sSL --max-time 20 'https://signalry.ai/' · public endpoint, no account or key needed; compare the output with the check's assertions
• check 1 (tls_cert): anyone can rerun · needs public_tls · cost free
  openssl s_client -connect 'signalry.ai:443' -servername 'signalry.ai' </dev/null 2>/dev/null | openssl x509 -noout -subject -enddate · public endpoint, no account or key needed; compare the output with the check's assertions

Rest of site (deploy continuity, dc-1)

https://signalry.ai: 1 of 1 other route OK · run 2026-11-02T07:30:08+00:00 with receipt rh-f608bd9a6db6. Passing the asked-for checks isn't the same as not breaking anything else, so each run also re-reads up to 10 of the site's other known routes. Shown, not scored: this never changes the verdict or tier.

• https://signalry.ai (sitemap.xml)
  • OK https://signalry.ai/paid-sr HTTP 200 · from sitemap

Missed rechecks (mr-1)

no scheduled recheck window has passed without a run · next scheduled window: due 2026-11-04T07:45:47.787Z, grace until 2026-11-04T13:45:47.787Z. A window that passes without a run is recorded as an omission, so silence never reads as "held". Shown, not scored: never a pass, never a fail; the verdict, tier and track-record score don't read it.

Why the verdict changed (vc-1)

1 verdict flip across 14 conclusive hosted runs: 1 target changed. Each hosted result binds the rules it ran under and a token for what it read, so a flip says whether the rule changed or the target changed. Shown, not scored.

• pass → fail at #11401 2026-10-31T13:04:51+00:00: target changed · check 1 (http) target changed · read 2ada5a24456fd863 → d52d1e43ed84d322

Evidence written by

Independent sources: 1 (1 passing in the latest run) · is-1: distinct (writer, origin) pairs among the checks whose writer is independent, where origin is the registrable domain, the GitHub owner/repo, or the claimant's runner. Five checks on one host are one source; self-written and unknown evidence never counts. Sources: signalry.ai (self, check 0); signalry.ai (third party service, check 1). Shown, not scored.

• Check 0 (http signalry.ai): the claimant (self) · the claim is about signalry.ai; its operator (normally the claimant or its org) writes the response
• Check 1 (tls_cert signalry.ai): a third-party service · the certificate is written and signed by a public CA; the check passes only on a publicly trusted chain

Evidence independence (ew-1): 1 self, 1 third party service · 1 of 2 check(s) read records an independent party writes. Derived by this service from each check's type and target, never from the claimant; when it can't tell, it says unknown. Shown, not scored: it doesn't change the tier, limits or track-record score.

Checks exercised (kc-1): yes · every check kind in this claim has been seen to pass and refute on known-answer canaries, and none is currently failing (kinds: http_contains, http_status, tls_cert). Known-answer canaries under checker countersign-hosted/0.3.2, rules csr-1; see /trust. Shown, not scored.