The instrument, not testimonials.
No logo walls, no invented case studies — padding gets smelled in diligence. What's below is the thing itself: the record your audit file would hold, the ledger your repo would carry, and the crosswalk your auditors would ask for. Specimens are synthetic and marked as such; the structure is the product.
The sample validation record.
One artifact, one decision, full provenance — what was checked, against which standard version, by whom, on what date. This is the deliverable: a signed record, not a live agent.
| Artifact | Requirements draft R-2031, "Payment exception handling — current state" (v3) |
|---|---|
| Standard version | Standards Ledger @ v0.41 (pinned at check time) |
| [auto] checks | Structure validation — PASS · version-to-version integrity diff — PASS (no dropped criteria, no silent numeric changes) · citation format — PASS |
| [judge] verdict | Approved with one correction: §4 qualifier restored from "should" to "must" — correction promoted to ledger rule R-044 |
| Judge | M. Rosfelder (named human; specimen) |
| Date | 2026-06-10 |
| Record integrity | SHA-256 of signed record: 9f31…c2a7 (specimen value) |
A Standards Ledger excerpt.
Every rule exists because something went wrong once, got caught by a person, and got written down — dated, attributed, versioned. Tagged [auto] when a machine can verify it, [judge] when a human must.
| Rule | Tag | Statement | Lineage |
|---|---|---|---|
| R-019 | [auto] | Acceptance criteria keep Given/When/Then order; no criterion may be removed between versions without a logged rejection. | added v0.22 · 2026-04-30 |
| R-031 | [auto] | Dollar amounts reconcile to the cited source table within rounding; mismatches halt the run. | added v0.29 · 2026-05-14 |
| R-038 | [judge] | Regulatory references are quoted verbatim from the primary source — never paraphrased, never cited from a secondary summary. | added v0.33 · 2026-05-27 |
| R-044 | [judge] | A softened obligation qualifier ("must" → "should") anywhere in a compliance-adjacent document requires explicit judge sign-off. | added v0.41 · 2026-06-10 |
The framework crosswalk.
Your auditors recognize frameworks, not philosophy. Where each WTD deliverable lands against the references they already use.
| WTD deliverable | Maps to |
|---|---|
| Signed validation record | NIST AI RMF "Valid & Reliable" (AI 100-1; GenAI Profile AI 600-1) · ISO/IEC 42001 records of AI-system decisions · model-risk documentation expectations |
| Independent validation gate | Model-risk three-lines-of-defense (independent validation function) · EU AI Act Art. 14 human oversight, including automation-bias guards (enforcement powers Aug 2, 2026) |
| Standards Ledger | ISO/IEC 42001 documented criteria & continual improvement · FINRA Notice 24-09: supervision obligations apply "irrespective of the technology" |
| Named residual-risk owner | Three-lines-of-defense accountability · NIST AI RMF "Govern" function — risk acceptance is documented and assigned |
The rework tax is already on your books — it's just invisible.
Survey-based self-report, labeled as such: 40% of desk workers received AI-generated "workslop" in the prior month, averaging just under two hours of rework per instance — roughly $186 per employee per month (Stanford Social Media Lab + BetterUp, N=1,150, Sept 2025). At a 500-person firm, that arithmetic is about $1.1M a year of polished-but-hollow output being quietly repaired downstream. Validation gates make the tax visible, then shrink it.
Want this record with your workflow's name on it?
The diagnostic produces one — whether or not the answer is "proceed."