Cyan Eclipse Wageward MCP — Cyan Eclipse Bazaar

Vendor: Cyan Eclipse Technologies · Platforms: MCP over stdio

Certification scope

CERTIFIED(local) — Eclipse Bazaar Test Bench 2026-10-25, standing card eclipse-wageward-mcp-2026-10-25-5.json (scoped re-bench -5): 20 pass, 0 na of 20 checks. CEILING — read exactly: this certifies the approval-and-evidence rail at local scope over a SYNTHETIC workforce, and nothing broader. Wageward is an approval-and-evidence rail ONLY — it is not a system of record, has no payroll-calculation or tax engine, and the submit-class tools (payroll_run_submit / filing_submit) are ABSENT from the surface by design in P0/P1: the bench verified by live stdio that calling one returns unknown-tool. Every approval/filing packet carries the mandatory statement that approval does not transfer payroll-tax liability (the employer remains liable); a packet missing it is invalid by construction. The four prior FAILED cards are retained append-only and cited here: eclipse-wageward-mcp-2026-10-25.json (FAILED), -2.json (FAILED), -3.json (FAILED), -4.json (FAILED — D2-GEN2 HIGH, separator-run identifier evasion) — the -5 scoped re-bench closes D2-GEN2 on every channel (live-stdio echo 7/7 bench shapes, persistence 2/2, approval-signing 4/4, plus fresh no-cap probes 8/8) on top of everything closed in -2/-3/-4. OPEN SPEC ITEM (not a P0 claim): receipt-TTL is an open spec item to settle pre-P1. N-CHECK: NOVELTY_PROVEN (stands — the fused fingerprint + Herald-signed approval + drift-refusal mechanism is unchanged by the fix rounds; no CRUD verb added).

What it does

Approval-and-evidence rail ONLY: agents prepare payroll runs and filing packets, humans approve the exact run — payroll_run_submit / filing_submit class tools are ABSENT from the surface by design (spec §10.2), in P0 and P1 alike
Independent canonical fingerprint of every run (employee set × amounts × deductions × accounts-as-HMAC-tokens) with a Herald-signed approval bound to that fingerprint — any post-approval drift refuses RUN_CHANGED_SINCE_APPROVAL with named diff classes
SSNs and bank numbers exist only as employer-scoped keyed-HMAC tokens (raw hashes banned); a deep identifier screen refuses separator-obfuscated SSNs/bank numbers on every channel — echo, persistence, and approval-signing
Bank-detail changes verify through the employee's own channel of record; every sensitive read seals an access record; evidence packs are portable and hash-chained per tenant
Every approval/filing packet carries the mandatory liability statement — approval does not transfer payroll-tax liability, the employer remains liable — and a packet missing it is invalid by construction

Tools

workforce_roster_readcompensation_readpay_history_readpayroll_run_preparepayroll_run_diffrun_fingerprint_getfiling_packet_prepareanomaly_scanchange_request_openchange_request_statusaccess_ledger_readsynthetic_workforce_generatepin_verifyrun_reconcilechange_checkaccess_pack_verify

Problems it addresses

My agent prepares payroll, and I need a human approving the exact run — fingerprint-bound — not a per-call confirmation click that approves whatever the run became later
I need SSNs and bank details to never exist as raw values or raw hashes anywhere in the agent's reach — tokens only, with obfuscated forms refused on sight
When an auditor asks who approved this run and what exactly it contained, I need a sealed evidence pack, not a chat log

Not tested

Live payroll/HR providers — P0 is an aggregator-shaped fixture adapter over synthetic data only; no live grant custody (P2), no real employer, employee, money, or filing
Any real payroll submission or filing — submit-class tools are ABSENT by design in P0/P1; the internal simulated rail is exercised by the builder suite only, never exposed
Production Herald/vault/KMS custody — P0 uses simulated per-tenant fixture keys
Receipt-TTL behavior — open spec item, to be settled pre-P1

Bench scorecard: test-bench/results/eclipse-wageward-mcp-2026-10-25-5.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 →