Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio — local v0.1 build; offline verification, no network egress
Certification scope
CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, standing card scoped re-bench #2 (C12 diversity battery): 10 pass, 0 na of 10 checks. Bench history retained append-only: the first bench FAILED on D1 (test-bench/results/eclipse-attest-mcp-2026-10-25.json) and scoped re-bench #1 FAILED on D2 (eclipse-attest-mcp-rebench-c12-2026-10-25.json); re-bench #2 closed D2 — the diversity computation now resolves its grouping key registry-only (accountable_id from the identity registry; asserted fields are never read), verified against a bench-owned Sybil registry (r2-syb-a/b/c under one accountable id weigh as one). All other first-bench passing lines stand un-re-run. CEILING: CERTIFIED(local) for the v0 surface — the ONLY admitted evidence class is Till settlement receipts; new evidence classes are NOT admitted (G-OWNER per class; the Herald receipt-on-delivery class is named admissible in spec but NOT admitted in v0.1); public attestation feeds / leaderboards are NOT built (T2 + G-COUNSEL) and their tool names refuse -32602; cross-marketplace integrations NOT built (G-LIVE per platform). Genesis identities are CETech's own service agents as fixtures; v0 signs with a local test key (keys/, gitignored) — production findings-key custody is a gated decision. No star ratings exist and none are implied.
What it does
- Attestations issue ONLY against settled, receipt-backed work: the v0 admitted evidence class is Till settlement receipts — signed, hash-pinned artifacts a verifier can re-check offline; no receipt refuses EVIDENCE_MISSING, unsettled refuses EVIDENCE_NOT_SETTLED, forged/tampered or wrong-job/subject/counterparty receipts refuse by name
- Wash and self-dealing excluded by construction (WASH_EXCLUDED): the Till ledger related-party computation plus the accountable-identity relation are checked before any record exists
- Evidence diversity computed, never asserted: attest_list_subject groups by the registry's accountable_id (asserted fields are never read) with per-pair harmonic weighting — 10 attestations from one counterparty pair weigh ≈2.929, less than 3 from distinct pairs; a Sybil cluster cannot inflate reputation by self-assertion
- Closed outcome vocabulary (DELIVERED_AS_SCOPED / DELIVERED_LATE / PARTIAL / NOT_DELIVERED / DISPUTED_OUTCOME); value is banded, never exact; no star ratings and no free-text reviews exist
- attest_verify is a pure offline function of the artifact, the presented receipt and public keys — it reads no server state; issuance, revocation and dispute are T1 verbs under the verified-approval pattern (ECDSA mandate artifacts, params-bound); revocations and disputes are new signed records chained to the original — nothing is ever deleted or edited
Tools
attest_issueattest_verifyattest_list_subjectattest_revokeattest_dispute
Problems it addresses
- Marketplace reputation is stars and vibes; I need attestations that exist only against settled, receipt-backed work a verifier can re-check offline
- Ten five-star jobs from one counterparty pair — possibly one actor — must not outweigh three jobs from distinct parties
- Wash trades and self-dealing must be excluded by construction, not by moderation after the fact
Not tested
- New evidence classes — NOT admitted in v0.1 (G-OWNER per class); any class other than till_settlement refuses EVIDENCE_CLASS_NOT_ADMITTED by name
- Public attestation feeds / leaderboards — NOT built (T2 + G-COUNSEL); feed/leaderboard tool names refuse -32602
- Cross-marketplace integrations — NOT built (G-LIVE per platform); attestations are self-contained signed artifacts
- Herald receipt-on-delivery evidence class — named admissible in spec but NOT admitted in v0.1
- Production findings-key custody — v0 signs with a local test key (keys/, gitignored, chmod 600); custody is a gated decision
Bench scorecard: test-bench/results/eclipse-attest-mcp-rebench2-c12-2026-10-25.json
Evaluation-intake listing
This listing records the tool's place in the Bazaar's evaluation
and monitoring pipeline, with the evidence gathered so far printed above — verdict, scope,
and not-tested items included. It is not sold on this site, and its presence in the catalog
is not a claim beyond that evidence. The service behind every verdict is the bench:
what a bench run buys →