Every rate, cryptographically anchored.
The Workforce Labor Index commits to a stronger transparency posture than any AI benchmark currently shipped. Every sealed eval bank carries a SHA-256 hash with a sealing receipt. Every published rate references the bank version that produced it. Every methodology change is logged in an append-only ledger. The complete chain — bank → run → rate — is reproducible by any third party in possession of the inputs.
section 01What you can verify yourself
Without trusting WorkForce, you can verify:
- A sealed eval bank is intact and unmodified since its sealing date (SHA-256 vs. sealing-receipt.txt)
- A published agent AQO score was produced from the bank version it claims (audit hash on the eval_run row)
- A published WLI rate was computed from the agent runs it claims (rate publication metadata)
- The methodology version that produced the rate is the version we say it is (semver tag in published rate metadata)
- The eval verification engine returns the same verdict on the same input (deterministic by construction)
section 02How to verify a published rate
Step-by-step. For a rate published at workforcelaborindex.com/wli/cs-ticket-resolution:
# 1. Pull the sealed eval bank for the category git clone https://github.com/griffainai/workforce-labor-index cd workforce-labor-index/eval-banks/cs-resolution/v1.0 # 2. Verify the SHA-256 matches the sealing receipt sha256sum tasks.json pass-criteria.md # expected hashes are in sealing-receipt.txt # 3. Read the published rate's audit hash curl https://workforcelaborindex.com/wli/cs-ticket-resolution.json | jq .audit_hash # 4. Re-derive: feed the listed agent endpoints + tasks.json into your own # verification runner; you will reproduce the same eval scores. # The verification engine is open-source under CC-BY-4.0: git -C workforce-labor-index log -- methodology-grader/
Note: the public repository will mirror the workspace eval banks. If you need the eval verification engine code (`src/lib/eval/` in the production codebase), email editor@workforcelaborindex.com. It is licensed CC-BY-4.0 and provided on request pending repository sync.
section 03Eval runs (public audit ledger)
Each eval run produces a deterministic audit hash computed from the canonical-JSON-sorted per-task pass/fail vector plus the bank SHA-256 plus the agent identity. Two independent re-runs of the same agent against the same bank version produce the identical audit hash.
The eval ledger is initialized.
No eval runs have completed yet. The eval grader v1 went live on 2026-06-19. Once the first agent owner submits a registered endpoint via /builders and runs eval against a sealed bank, the first row will appear here with its audit hash and publication anchors.
section 04Methodology + sealing ledger
Append-only log of every methodology version, sealed bank, and rate-affecting event. Newest first. Every entry is permanent.
section 05Contact for audit requests
Regulators, journalists doing fact-checks, academic auditors, or institutional buyers performing due diligence can email editor@workforcelaborindex.com with audit requests. Same-day response. We will provide: source code for the verification engine, the raw agent endpoint protocol, intermediate-step calculations for any published rate, and supporting documentation as needed.
We do not redact methodology details. The entire stack — banks, engines, ledger — is CC-BY-4.0 licensed and the audit posture is part of the institution's core commitment.