Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio (HTTP transport on roadmap)
Certification scope
CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, 18 pass, 1 na of 19 checks (U1 auth: stdio process boundary). Run by W-003 (Verity), suites/codex.py. Run 1 FAILED retained (genuine product defect U5: clock_status on an unknown instance id returned a raw -32603; fixed class-wide via require_str_id plus a transport schema refusal, then re-benched clean). Run 2 FAILED retained (bench-suite assertion bound to the pre-fix refusal shape). Non-failing observation: a whitespace-only redaction reason is currently accepted (queued as a bench note).
What it does
- Versioned Regulation Library: provision records are never overwritten, resolution is as of a date, and proposed (NPRM) text never resolves as a current obligation; Federal-Register-style change events surface as data, not prose conclusions
- Obligations Mapper (provision -> obligation -> control -> evidence); the CDD Rule duty (31 CFR 1010.230) and CTA BOI reporting (31 CFR 1010.380) are modeled as separate obligations the merge guard refuses to combine — and same-citation duplicates still merge
- Deadline Engine that distinguishes rule clocks from guidance cadences: SAR 30d/60-max, CTR 15d, Reg E windows can BREACH; the ~90-day continuing-activity cadence can be overdue and can never render as a breach
- Local sanctions screening against a synthetic SDN-shaped fixture list: MATCH_CANDIDATE / NO_MATCH with near-matches flagged, never silent; no auto-clear exists — dispositions are human acts, recorded verbatim with an officer reference required
- Evidence Packager with a MANDATORY gap section: gap suppression is refused and receipted as a violation; redactions require reason + actor, are receipted, and stay visible marked REDACTED on both surfaces
- Posture is product law: not a bank, not counsel, not a compliance department — the suite files nothing, determines nothing, and a banned-output lint quarantines any compliance-verdict language
Tools
library_searchlibrary_provisionlibrary_provision_addlibrary_changesobligation_mapobligation_coverageobligation_mergeclock_rulesdeadline_computeclock_openclock_statusclock_escalationsclock_exceptionsscreen_list_infoscreen_lookupscreen_disposition_recordevidence_pack_buildevidence_pack_redactevidence_pack_verifyreceipt_listreceipt_verify
Problems it addresses
- Exam evidence lives in five folders and nobody can prove which version of a rule applied on a given date
- On our tracker a guidance cadence and a filing deadline look identical — only one of them can actually breach
- A sanctions near-match was silently cleared and nobody can show me who decided, or what they wrote
Not tested
- Real provision corpus — R0 library carries public-domain-shaped FIXTURE records with real CFR citations; the live eCFR / Federal Register corpus and change-feed mechanics are R1 scope
- Real OFAC SDN / Consolidated list data — R0 screens against a SYNTHETIC fixture list shaped like the SDN list; no real list was downloaded
- A real institution — R0 runs against the synthetic Meridian Fictional Bank fixture only; first real use is the owner-gated R1 design partner, inside that institution's boundary, after counsel review (spec 22)
- Deadline values flagged [verify] in the spec (OFAC reporting clocks, per-charter SAR variants, CTR extension mechanics) — carried as fixture values with the flag visible in the rule record
Bench scorecard: test-bench/results/eclipse-codex-mcp-2026-10-25-3.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 →