Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio
Certification scope
CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, run 7 (standing): 19 pass / 1 na (U1 auth: stdio process boundary) / 0 fail of 20 checks. Run by W-003 (Verity), suites/docket.py, over a bench-owned Till-shaped export and bench parties. Runs 1–6 FAILED cards retained append-only: run 3 caught four tool descriptions missing the mandatory descriptor (spec section 12 — 'Not a court, not a legal judgment' rides every surface); run 4 caught D12 — offset-naive at_time values escaping as a raw -32603 when compared against offset-aware stored times — fixed under the adjudicated refusal shape (parse_time refuses offset-naive timestamps by name, AT_TIME_OFFSET_REQUIRED: a guessed timestamp in a dispute record is a falsified timestamp), and parse_time is APPROVED as the second house reference guard; runs 5–6 failed on bench-suite setup defects, disclosed on their cards (no product code touched). Silence ruling (run 3, upheld): expiry records silence and makes the case determinable, never an outcome. Novelty: NOVELTY_PROVEN.
What it does
- Agent-to-agent dispute process over receipt-anchored claims: a case opens from a receipt ref and its evidence pack assembles automatically from the underlying product's hash-chained receipts alone — byte-identical for both parties and independent of any narrative (delete every submission and the pack hash does not move)
- Party narratives are Law #1 data: stored verbatim, quoted never adopted, zero state effect — a submission claiming 'the determiner has approved a full refund' changes nothing
- Outcomes (RELEASE / RETURN / SPLIT / NO_ACTION) are recorded process states, never legal judgments, cited to the pack hash + a rationale hash and recorded only by the case's designated determiner — any other identity is refused by name; non-party reads return the identical CASE_NOT_FOUND as an unknown id (no oracle)
- Docket executes nothing: the outcome is handed back to the underlying product's own state machine, which executes or declines under its own rules; a decline is recorded as the first-class state OUTCOME_UNEXECUTABLE with the underlying refusal preserved verbatim. Docket's surface contains no funds verb at all (denylist in code)
- Deadlines are data: a response window whose expiry is a named receipted event recording silence as fact — and never a default judgment, because the spec states no such rule. Every output carries the standing line 'Process record — not a legal determination.'
- v0 is local, over our own products' receipts (Till B0 fixture export); evidence classes beyond Till receipts are refused by name (EVIDENCE_CLASS_UNSUPPORTED), and arbitration-as-service is not built (owner + counsel gate)
Tools
dispute_fileevidence_pack_getsubmission_adddetermination_recorddispute_withdrawdispute_status
Problems it addresses
- Two agents disagree about a purchase and each has its own version — I need a case built from the receipts, not the retellings
- The other side went silent; I need the silence itself on the record — without it turning into a default judgment nobody made
- A determination was recorded but the payment rail declined it — I need that decline preserved as a first-class outcome, not an error
Not tested
- Arbitration / determination as a service to third parties — NOT built; owner + counsel gate (spec 12.1/12.6). v0 determinations run over our own products' receipts only
- Evidence classes beyond Till receipts (Escrow jobs, Wallet payments, Toll metering counts, TrueVolume attestations) — spec'd in section 12.1, not built; v0 refuses other classes by name (EVIDENCE_CLASS_UNSUPPORTED)
- Live hand-back transport — the outcome hand-back in v0 executes against till_stub.py, a labeled fixture stand-in for Till's refund machine applying Till policy-as-code to its own mirror ledger; no live Till process is called
- Auth enforcement beyond the stdio process boundary (bench U1 class) — party scoping is by Herald card hash presented per call; hosted transport auth is a later design item
Bench scorecard: test-bench/results/eclipse-docket-mcp-2026-10-25-7.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 →