What does SaSame evidence prove?
Observations, compatibility, registry/runtime history, owner verification, measurements and incident reports.
- Collection
- evidence
- Updated
- 2026-07-29
- min read
- 4
- Version
- v1
Evidence records what was observed, when, how and with which limitations. It does not automatically prove safety, quality, ownership or endorsement.
- Observed
- 2026-07-13 00:00 UTC
- Claim strength
- illustrative
- Dataset scope
- Describes the Evidence Record v1.0 schema (subject, observed_at, observation_method, operational_state, history, evidence_status, claim, signed_record, limitations) that every page below instantiates. This overview is not itself a specific measurement.
- Methodology
- Every claim on an Evidence page traces to an Evidence Record v1.0 instance produced by a real MCP
initialize,tools/list, or schema-validation call against a stated endpoint at a stated timestamp. A mandatory rule applies to every record: Unknown is not Negative — a no-evidence result classifies asinsufficient_recent_evidence, never asobserved_not_callable. - Limitations
- This page is illustrative of the schema itself; it does not assert that any specific server is currently callable. See Compatibility Evidence, Registry and Runtime Evidence, and Incident Reports for dated, scoped claims.
Illustrative — describes a method or schema, not a specific measured result
In this collection
3Observation
Every observation carries time, method, target identity, result and provenance, recorded as one Evidence Record v1.0 instance.
Three states cover every observed endpoint — there is no fourth "unknown means broken" state:
| State | Meaning |
|---|---|
observed_callable | The endpoint responded to a real protocol probe at the recorded timestamp. |
observed_not_callable | The endpoint was reached but the probe failed (error, timeout, schema mismatch). |
insufficient_recent_evidence | No probe has run recently enough to make a claim either way. |
Freshness has two thresholds: a record is fresh within 7 days of observed_at, recent within 30 days, and anything older is treated as insufficient_recent_evidence regardless of what the last real result was.
Compatibility
Compatibility is bounded to a stated protocol surface and test method, not a universal quality rating. See Compatibility Evidence for the exact protocol-revision matrix and dated results.
History
Longitudinal changes and incidents remain versioned so current status never erases prior evidence. A record's history carries observation_count, consecutive_failures, and a capped list of the last 10 grade/tool-count/criteria changes — old evidence is superseded, not deleted.
Claiming ownership of an endpoint never changes its measured state or grade — the claim field and the operational_state field are recorded independently and can disagree (an owner can claim a server that measures as observed_not_callable).
Examples
2- 01
A
tools/listresponse is preserved as a timestamped observation and related to later schema changes. - 02
Every Evidence Record carries mandatory standard limitations in its own payload: "Not a security verdict", "Not an endorsement", "Reflects observations at specific timestamps" — the caveat travels with the data, not just with this page.