Every claim, and every struck claim.
This site is built fail-closed: a factual sentence appears only if its claim is on the register with a passing verifier. A claim that fails, or cannot yet be verified, is struck or held, and the sentence does not ship.
- commit
- 9865101c
- built
- 2026-09-04
- last full run
- 2026-09-01
- verified
- 27 claims
- struck
- 7 claims
- held
- 2 claims
Read from the verification report the build generates. Repository-fact checks re-run at every build.
How the build decides
Three outcomes for a claim. Only the first one reaches a page.
Verified
Verified
27 claims on the register, each with a named verifier: a test run, a repository fact, or a primary source. Demonstrated means a visitor can watch the behaviour run.
Struck
Struck
7 candidate claims that failed verification, or cannot yet be verified. They appear nowhere on this site; the sentence does not ship.
Held
Held
2 awaiting an external confirmation before they may appear. A sentence we cannot verify does not stay up while we look into it.
The register
Claims on the register. 27 sentences, each with the receipt beside it.
Nothing here is written by hand. The table is the claims register as it stands at this commit, so the page cannot disagree with the sentences the build let through. A claim struck from the register is struck from this page and from every page that stated it, in the same build.
Effective 2026-09-04 · regenerated with each build · generated 2026-09-04, commit 9865101c
| ID | Claim | Class | Verifier | Verified |
|---|---|---|---|---|
| CLM-0001 | Demo Company Limited is registered in England and Wales, company number 00000000. | Verified | primary source — Companies House register, demo entry (name, number and status asserted) | 2026-09-01 |
| CLM-0100 | Every action the platform governs is checked before it executes; an action that fails its checks does not execute. | Demonstrated | test run — kernel adjudication suite — an in-bounds action verified with a signed record; over-limit and authority-breach actions refused | 2026-09-01 |
| CLM-0101 | Same input, same verdict, every time. | Verified | test run — kernel determinism and property suite, green | 2026-09-01 |
| CLM-0102 | When the governance layer cannot decide, the action is refused, not waved through. | Verified | test run — fail-closed suite, green — an empty verdict set persists as a refusal, never a pass | 2026-09-01 |
| CLM-0103 | A refused action is recorded with the reason it was refused. | Verified | test run — refusal records carry the failing checks by name (governed-refusal test, green) | 2026-09-01 |
| CLM-0104 | Where the platform gates a dispatch, the refusal is enforced at the database itself, on every write path. | Verified | test run — dispatch-gate integration suite, green — the database trigger refuses an ungoverned assignment on every write path | 2026-09-01 |
| CLM-0105 | The kernel applies deterministic mathematical governance; if its own checks cannot be discharged at start-up, it does not serve. | Verified | test run — kernel mathematical property checks and boot self-check suite, green — the kernel refuses to serve if its checks cannot be discharged at start-up | 2026-09-01 |
| CLM-0106 | A complete governed workflow — dispatch refusal, human approval gates, closure control, and an offline-verified evidence pack — was re-executed end-to-end over HTTP on 1 September 2026. | Demonstrated | test run — governed-workflow integration suite re-executed end-to-end over HTTP, green, including the refusal cases and the offline-verified evidence pack | 2026-09-01 |
| CLM-0200 | Every governed action is sealed in a signed, tamper-evident record. | Verified | test run — evidence suite, green — signed records issued; a tampered record stops verifying | 2026-09-01 |
| CLM-0201 | Records can be verified offline, without access to our systems. | Verified | test run — the shipped offline verifier, built and exercised in the same run: pass on the true record, fail on the altered copy | 2026-09-01 |
| CLM-0203 | The evidence a period produces can be handed to a third party and verified with Admissible absent. | Verified | test run — a period's evidence pack built and verified by the shipped verifier in a clean process | 2026-09-01 |
| CLM-0301 | An industry is a configuration file, not a codebase. | Verified | repository fact — the industry pack configuration files load through one loader; engines contain no industry-specific code paths | 2026-09-01 |
| CLM-0302 | The industry pack is load-bearing: at boot the governance layer recomputes the pack's rule digest and refuses to serve if it does not match. | Verified | test run — pack-digest binding suite, green — boot refuses to serve on a rule-digest mismatch | 2026-09-01 |
| CLM-0303 | Agents recall prior governed context on every run. | Verified | repository fact — recall runs inside the live agent execution path, with a real-database test | 2026-09-01 |
| CLM-0400 | Agents run under quotas and spend caps, with human approval gates on the actions that matter. | Verified | test run — quota suite green (atomic deduction, no overspend under concurrency) and approvals suite green (fail-closed gate, hold and resume) | 2026-09-01 |
| CLM-0401 | An unsafe agent recipe is blocked at compile time, before it can run. | Verified | test run — the recipe compile gate blocks on critical behavioural findings (compile-gate tests, green) | 2026-09-01 |
| CLM-0402 | An action held for approval resumes by replaying only what was approved — nothing else. | Verified | test run — approvals resume test, green — the resumed run executes only the stored approved action | 2026-09-01 |
| CLM-0500 | Tenant isolation is enforced in the database itself and attacked by our own adversarial suite in CI. | Verified | repository fact — row-level isolation enforced by the database with real-database tests; an adversarial suite attacks the same primitives in CI | 2026-09-01 |
| CLM-0601 | Insurance OS includes broker submission intake by API and email, feeding an underwriting queue. | Verified | repository fact — broker submission intake by API and email exists in the Insurance OS codebase | 2026-09-01 |
| CLM-0602 | Insurance OS includes a governed tool interface through which AI assistants quote and hand off binds, with dispatch recording built in. | Verified | repository fact — an assistant-platform tool interface with recorded dispatch exists in the Insurance OS codebase | 2026-09-01 |
| CLM-0603 | Insurance OS covers underwriting, claims, reinsurance, bordereaux and regulatory reporting workflows. | Verified | repository fact — underwriting, claims, reinsurance, bordereaux, regulatory-returns and finance modules present in the Insurance OS codebase | 2026-09-01 |
| CLM-0700 | EU AI Act facts as recorded in the evidence pack, including the amended high-risk timetable. | Verified | primary source — EUR-Lex: Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, fetched 2026-09-01 | 2026-09-01 |
| CLM-0701 | Standards navigator entries as recorded in the evidence pack. | Verified | primary source — primary sources fetched and recorded for DORA, Consumer Duty, SM&CR, SS1/23, NIST AI RMF, ISO/IEC 42001 | 2026-09-01 |
| CLM-0702 | Air Canada ledger entry. | Verified | primary source — Moffatt v. Air Canada, 2024 BCCRT 149 — pinpoints cross-verified across convergent legal sources, 2026-09-01 | 2026-09-01 |
| CLM-0703 | July 2025 coding-agent database deletion ledger entry. | Verified | primary source — contemporaneous reporting 21–22 July 2025 (incident account and vendor confirmation) | 2026-09-01 |
| CLM-0704 | Runtime assurance lineage facts. | Verified | primary source — engineering lineage sources (Simplex 2001; ASTM F3269; DARPA Assured Autonomy) fetched and recorded | 2026-09-01 |
| CLM-0706 | DEWR assurance-review ledger entry (report corrected February 2026). | Verified | primary source — DEWR assurance-review page (replacement history, fetched directly); contested detail attributed in-text to contemporaneous reporting | 2026-09-01 |
Not on a page
Struck, and held. The sentences we will not write until they verify.
struck · 7
Failed verification, or cannot yet be verified
| ID | Why it does not appear |
|---|---|
| CLM-0600 | End-to-end cascade runs were mixed on the local instance on 2026-09-01; present-tense cascade claims withheld until a canonical instance runs green. |
| CLM-0900 | A signing-capability claim is withheld until proven end-to-end on the exact surface described. |
| CLM-0901 | Latency figures: no benchmark verifier in v1. |
| CLM-0902 | No claim about live commercial deployments is made; the internal gate for that claim is explicitly not met. |
| CLM-0903 | Not on the live path; no claim is made. |
| CLM-0904 | Operational cadences (weekly/monthly services): not operating. |
| CLM-0905 | Internal metrics are not used as public claims. |
held · 2
Awaiting an external confirmation
| ID | Status |
|---|---|
| CLM-0604 | Held pending an external confirmation. |
| CLM-0802 | Held pending an external confirmation. |
Generated from commit 9865101c on 2026-09-04. Last full run 2026-09-01.
The honest page
The honest page
Before you spend an hour with us: what is demonstrated, what is verified, and the sentences we will not write.