Research group lead
Can I trust my own AI's output?
Buys: Governance Audit · $5,000
Pietarien Innov8 · The Governance Instrument
The rule registry, the evidence classification, the gate log and the coverage statement for an autonomous AI estate — delivered in five working days, with the method published, including where it is thin.
Organisations that run autonomous mathematical discovery accumulate governance rules faster than they can audit them. We turned one estate's constitution into a measured register — and published the measurement, including where it is thin.
98 numbered laws inside 165 records, each carrying the file and line it came from. Six of them instrumented. The gap between 6 and 98 is stated on the first page, not buried.
Rules are written after failures, cited by later rules, and never counted. When a reviewer asks how many govern the system, the answer is a round number nobody can produce.
That sentence is the paper's, and it is the reason for the register. If you cannot count your rules, you cannot know which of them are evidenced, which are preference, and which have quietly stopped applying.
This is for a research group or a deep-tech team that has to answer one question: “how do we know we can trust AI-generated results?” — and for a funder or an acquirer who needs someone else's estate audited, and wants to read the register rather than commission it.
Research group lead
Can I trust my own AI's output?
Buys: Governance Audit · $5,000
Deep-tech founder
Will an acquirer trust my work?
Buys: Governance Audit + the 90-day re-audit subscription
Funder · program officer
Is this grant proposal's AI governance real?
Buys: a register requirement — the grantee commissions the audit, you read the coverage statement
Regulator · compliance officer
Which of our rules are evidenced — and where are the gaps?
Buys: Governance Audit + crosswalk to NIST RMF and ISO 42001
Acquirer
Is the target's governance performative?
Buys: a diligence requirement — the target commissions the audit, you read the register
Board · audit committee
Can we produce the count?
Buys: Governance Audit — the one-page board summary is the deliverable
Seven artefacts, delivered in five working days. Not a phone call — a register.
D1
Every rule in your estate on one row, each carrying the file and line it came from. On our own estate the register ran to 165 records across 98 laws.
D2
Each rule labelled evidenced or preference, with the basis stated row by row. On ours: 39 evidenced, 59 preference — published, not rounded.
D3
Where your rules contradict or have drifted apart, each conflict marked live or stale at time of use.
D4
Which rules are actually instrumented, and what each gate checks — the difference between governed and written-down.
D5
The number you should be afraid of: how many of your rules are actually enforced. Stated plainly — including, especially, if it is worse than ours.
D6
The artefact your board, your funder or your audit committee files. The count, the gaps, and what they mean — one page.
D7
The audit output redacted for sharing — safe to put in front of a board, a funder, a regulator or an acquirer.
The register is yours, exported to JSONL. The summary PDF is yours to share. Nothing is held back behind a follow-up engagement.
What the five days look like.
Day 1 — Harvest: read the estate's constitution and configuration, produce the register.
Day 2 — Evidence classification: label each rule evidenced or preference.
Day 3 — Conflict audit and gate log: re-measure recorded conflicts against disk; determine which rules are actually instrumented.
Day 4 — Coverage statement and board summary: assemble the numbers into the one-page summary.
Day 5 — Delivery: export to JSONL, deliver PDF, walk the client through the artefact.
The paper's credibility comes from stating what it does not cover. So does the audit.
The audit measures. Fixing what it finds is a separate engagement, scoped separately.
This is not an ISO 42001 certification and issues no stamp. It produces the register and the coverage statement that a certification audit will ask you to produce anyway.
The 90-day re-audit is a separate engagement, because rules decay. The register is built to be re-run.
The audit produces counts and classifications, not maturity scores — and if the count cannot be produced, the coverage statement says so instead of rounding it.
$5,000 sits on the price of an ISO 42001 certification audit — while delivering no certificate. The actual comparable is a $50,000 readiness engagement.
| What you get | Price |
|---|---|
| ISO 42001 certification audit | $5,000–$20,000 · certificate included |
| Big Four readiness engagement | $50,000+ |
| The Governance Instrument auditTHIS AUDIT Register, evidence classification, conflict table, gate log, coverage statement, board summary — no certificate. | $5,000 · 5 working days |
Vendor-reported ranges, 2024–2026. Sources: Compyl, Scytale, Vanta, Elevate (ISO 42001); Alice Labs (Big Four engagements).
This is not a conformity assessment under Article 43 of the EU AI Act. It produces the register a notified body will ask you for — not the certificate. We are not an accredited or notified body, and nothing here is a declaration of conformity.
Bar widths are illustrative; the Big Four bar does not fit on the page. It usually doesn't.
Article 50 is already in force. Annex III is twelve months away. If you cannot produce the count of which of your rules are evidenced, the answer to a regulator is “we do not know.”
In force now. Systems interacting with people, or generating synthetic content, carry disclosure obligations — and disclosure is a claim you have to be able to evidence.
The date most providers are working toward. Twelve months to produce the record of which rules govern the system and which of them anyone has ever tested.
High-risk systems inside regulated products. The longest runway, and the one where the evidence has to survive the life of the product, not the audit cycle.
Dates as published by the European Commission's AI Act implementation timeline. A register that cannot say which of its rules are evidenced fails first on exactly this question — not on the philosophy of the rule, but on whether anyone can produce the count.
39 carry a birth event or an evidence line; 59 carry neither and are recorded as preference (95% CI 50.3–69.3%). 33 of 94 transferable laws fired in a second charter (35%, 95% CI 26.2–45.2%), each anchored to a line. Eight conflicts were recorded; two were stale at time of use and six remain open.
Falsifier F6 — Coverage floor: if fewer than 15 of the 98 prove gateable at all, the paper reports a six-gate demonstration and drops every sentence implying the corpus is enforced.
Quoting 3,500:1 would divide the 98-law analyst cost by the 6-gate run time, which is a coverage error, not a speed-up.
On the laws actually instrumented the ratio is about 211:1: 720 seconds of comparable analyst reading against 3.4 seconds of compute. The 3,500:1 figure is labelled in the paper as a projection — a target, not a result — and F6 decides whether it is reachable at all.
Six named falsifiers, each with a corpus, an instrument and a threshold written down before the result. This is the section that makes the rest of the paper worth reading.
The 39/59 split, and therefore the headline that 59 are preference.
The 35% fire rate, and whether the map is reproducible at all.
"The audit finds live drift."
"Every gate can fail."
Cross-system transfer — whether the laws are architecture or a habit of one hand.
That this is an instrument rather than a demonstration.
Six uses. Five of them are someone else's deadline or someone else's diligence. The sixth is the one that keeps the other five true.
Before an ISO 42001 audit, know which of your rules are evidenced. The audit is the pre-flight.
A funder receiving an AI-generated results paper asks how the methods can be checked. The grantee commissions the audit; the funder reads the coverage statement. We audit the party who hired us, never the party across the table.
An acquirer buying a deep-tech team needs to know whether the governance is real. The target commissions the audit as part of diligence; the register goes in the data room.
An Annex III high-risk provider has to answer the Art. 72 post-market monitoring question. The audit produces the register and the coverage statement that answer it.
A board asks: are we governed? The audit produces the one-page board summary. That is the artefact they file.
Rules decay. The 90-day re-audit is the recurring product — and the reason the register is built to be re-run.
doi.org/10.5281/zenodo.24054537
The argument, the measurement, the six falsifiers, and the cost comparison stated on matched denominators.
doi.org/10.5281/zenodo.24055163
Every law with its record, its category, its evidence status and the file and line it came from. Released CC0 so the register can be checked without asking us anything.
Free
CC0
98 laws, 165 records, every row carrying the file and line it came from. Check it without asking us anything.
Open the register →Free
CC-BY-4.0
The measurement, the method, and the six falsifiers that would break it.
Read the paper →Engagement
$5,000 · one pipeline · 5 working days
You get a registry of your rules, an evidence classification, a conflict table, a gate log, and a coverage statement — how many of your rules are actually instrumented, including if the number is worse than ours.
Pilot
$2,500 · one portfolio · 3 working days
A ranked shortlist of your technology portfolio, with the reason for each position written down, plus the fault list for what ranked low.
A ranking, never a score. Calibration +0.54 at n=24; pilot-only until n≥60.
Larger estates are scoped on a call: no figure is quoted and no tier is named until a scope document exists for it. The two prices above are the two we can stand behind today.
Excludes remediation. Re-audit at 90 days is priced separately, because rules decay. What we won't do →
1
Twenty minutes, free. You tell us your estate; we tell you what a $5,000 audit will and will not cover. Notes are yours. We say no on the call rather than after an invoice.
FREE · 20 MIN · BY REQUEST
2
The scope document is signed, and half is invoiced up front.
50% · PAYONEER / PAYPAL / CAPITEC
3
Five working days. Register, conflict table, gate log, coverage statement, board summary — all exported, all yours. The second half is invoiced on delivery.
50% ON DELIVERY · 5 WORKING DAYS
You will not be asked to sign an NDA to receive the scope document. The audit output is yours, exported to JSONL and PDF. The corpus is deleted on request. No data is used for training.
Pietarien Innov8 is a South African innovation studio. We publish under CC-BY and CC0. The engine behind the audit is Pieta, which has produced 67+ attorney-ready patent packages and 791 EPO patents in its grounding corpus. The register and the paper are our own — a method that could not survive being published with its coverage gap stated would not be sold as an instrument.
Coverage is stated rather than implied throughout — including ours.
If you cannot produce the count, the answer is “we do not know.”
Produce it. Five working days, $5,000, everything exported and yours to keep.