Cyan Eclipse SEC Figure Witness — Cyan Eclipse Bazaar

Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio

Certification scope

CERTIFIED (bench, 2026-10-28) at this scope only: Eclipse SEC Figure Witness v0.1.0, 12-verb stdio surface, serving figures-as-filed bound to value-as-printed and a byte-range passage hash or XBRL fact identity, with restatement chains and offline-recomputable receipts (shipped verify.py), under sealed snapshot secfig-2026-10-28-eb66332db1 — FIXTURE_SIGNED (fixture key fixture-secfig-1; no production-signature claim) — P0 corpus scope: 48 CIKs / 905 accessions, 809,134 figures, convergence 34,206 compared / 0 divergences, seal sha256 71d0ecb2ddeebae5d1fe24f04b5788cdd9f7b0f138718fa264dac9bd97ca45e9. INCLUDED in this certification (benched 2026-10-28, BENCH-V2): EDGAR-self divergence observations (spec 14.2) and the divergence_edgar_self verb — 219,953 sealed observation rows, each comparing the sealed as-filed value against the companyfacts-served latest-vintage value in a hash-pinned capture: 90,938 MATCHES_OBSERVED (41.3%) and 129,015 DIVERGES_FROM_OBSERVED (58.7%). The split is source-reprocessing data (the derived view reflects later vintages), not an error rate. Pairing covers 56.1% of the 391,825 eligible (CIK, figure) identities; two eligible CIKs contribute no rows — one's capture file was corrupt at seal time (skipped, logged) and EDGAR's companyfacts aggregate for the other is frozen at 2019-05-24 at the source. On 11.3% of rows the capture holds several values for the latest filed date (frame duplicates); the sealed row takes the first-served entry, reproducible from the pinned capture. EXCLUDED from this certification: the vendor divergence leg (spec 14.1 — P1, not built), live ingestion, production/non-fixture signatures, and any coverage beyond the sealed snapshot. Release-leg coverage is capped: 153 of 262 release accessions extracted (41.6% not-extracted, above the spec-17 O4 40% bar) — per-figure release receipts are certified, release coverage completeness is not. Known limitations: release figures sealed with period_end = null are served via release_figures, not figure_at_moment; where a release repeats a figure label across passages, figure_at_moment returns the first sealed match without an ambiguity signal.

What it does

Seal filing bytes exactly as published by the source (figure-canon/1.0) and extract figures bound to their value as printed and to a byte-range passage hash or an XBRL fact identity
Three-moment chains: a figure as released in an Item 2.02 exhibit, as filed in the periodic report, and as restated after an Item 4.02 non-reliance event
Receipts any holder can recompute offline with the shipped verifier (verify.py)
Coverage stated plainly: universe U1 (S&P 1500 constituent list as published by the index provider of record) plus any filer with an Item 4.02 8-K or 10-K/A / 10-Q/A in the window; period floor 2019-01-01 — figures outside the named universe or sealed snapshots are named as gaps, never answered from memory, estimation, or a live fetch

Tools

figure_at_momentfigure_historyrelease_figuresfiling_figureschain_getchain_searchfigure_verifychain_verifydivergence_edgar_selfcorpus_statusmanifest_getmanifest_verify

Problems it addresses

I need to prove exactly which figure an SEC filing printed at a specific moment — as released, as filed, or as restated — not what a terminal shows today
A figure changed after an Item 4.02 non-reliance event and I need the full restatement chain, sealed and recomputable offline

Not tested

Vendor-leg divergence (spec 14.1) — a P1 boundary statement only, not built and not claimed; the certified observation leg is EDGAR-self divergence (spec 14.2) via divergence_edgar_self only.
Release-leg coverage completeness — 153 of 262 release accessions extracted (41.6% not-extracted, above the product's own spec-17 O4 40% bar); per-figure release receipts are certified, release coverage completeness is not (BENCH-V2 condition C3).
Production signatures — the sealed snapshot is FIXTURE_SIGNED (fixture key fixture-secfig-1); no production or non-fixture signature claim is made or certified.
Live ingestion and any coverage beyond the sealed snapshot secfig-2026-10-28-eb66332db1 — figures outside the named snapshot are named as gaps, never fetched (serving modules unchanged since the v1 bench's no-live-path audit).
Known limitations disclosed per C3 (v1 bench findings B1-1 LOW, B1-2 MEDIUM, both reproduced under v2): release figures sealed with period_end = null are served by accession via release_figures, not via figure_at_moment (which returns a generic -32602 for them); where a release repeats a figure label across passages, figure_at_moment returns the first sealed match without an ambiguity signal — both figures are enumerable via release_figures and each is individually recomputable.
Full-corpus per-figure recompute by the bench — the v2 bench verified recompute on its own samples (18 figures in-server, 15 receipts via the shipped offline verify.py, 80 of 80 sampled observation rows) plus a full manifest-entries audit; the seal-time batched probe record (716 of 716 files, 0 not-recomputable) is the seal's own coverage record (BENCH-V2 NOT_RUN disclosure).

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 →