# The Eleven Readings, Audited Against Their Transcripts

**version** v0.33.82
**date** 31 August 2026
**from** The site agent (Fable)
**to** Strategy, the RiskMandate team, the registry site

**type** Audit brief — no memo behind it

*The project lead asked for a review of the eleven insurance readings against the raw transcripts, with changes made where the interpretation did not hold. **This brief has no transcript above it**, because there is no memo behind it: it is the site agent auditing the site agent, and it says so at the top so nobody reads it as the project lead's position. The six defects are recorded in the documents they were found in rather than only here, because a correction that lives only in an audit is a correction nobody reading the document will see.*

---

## What This Is

**Every one of the eleven insurance readings was re-read against the transcript it came from**, and the four load-bearing corpus quotations were re-verified out of the pre-pivot files they cite. **The method held. The arithmetic and the housekeeping did not.** Six defects were found: three stale counts, one claim identified with the wrong quantity, one over-claim, and one dropped question. All six are fixed, and a build gate now refuses the class of defect that produced half of them.

## What Was Checked, And How

| | |
|---|---|
| **Transcripts** | All 11, read whole from the briefs, before re-reading any of the readings |
| **Readings** | All 11 briefs and all 11 doctrine documents |
| **Corpus quotations** | 4, re-read out of the files they name, with their dates checked |
| **Decisions** | GM-D35 to GM-D80 checked against the memo each cites |

## The Verdict First

**The interpretation is sound and I agree with it.** The things that would have been most damaging to get wrong were got right:

- **The two-stage split** — memo 1 does decouple the rating from the money, in those words, and everything downstream depends on that being a real separation rather than a convenient one.
- **The three brokers** ([doctrine 05](../insurance/not-in-line.html) §3). Memo 5's *"our job is to be the broker, right? Not the broker"* is the **insurance** broker. Memo 6's brokers are **access and execution** brokers — technology products, a different referent for the same word one memo later. Conflating them would have produced a doctrine that contradicted itself and looked consistent. It was caught.
- **The Fibonacci repair** ([doctrine 08](../insurance/the-world-model.html) §3). *"a finibash in a finibashy sequence"* is **Fibonacci**, and the reading's split — *what the bands are called* is settled at 1–5, *what the bands represent* is open — saves a settled decision from a good instinct that appeared to contradict it.
- **The 2D-first sequencing** disagrees with the memo, which explicitly asks for 3D. It is labelled as the site agent's engineering advice and carried into `does_not_prove`. That is the disagreement handled correctly.
- **All four corpus quotations verify verbatim**, at the dates claimed: *a signature proves possession* (19 Aug), *a verification is not a use* (20 Aug), the revocation-latency line (20 Aug), *trust is a path and revocation is that path no longer existing* (27 Aug).

## The Six Defects

### 1 · Three stale counts, and the pattern behind them

| Where | Said | True |
|---|---|---|
| [Doctrine 00](../insurance/what-this-is.html) | *"Eight memos are being recorded"* | Ten, and complete |
| Doctrine 01–08 footers | *"memo N of 8"* | of 10 |
| [Doctrine 08](../insurance/the-world-model.html), twice | memo 8 *"the last of the series"* | Memos 9 and 10 followed |

**This is the third time this class of defect has appeared in this folder**, after the hub counter reading *8 of 8* and `llms.txt` printing *seventy-one decisions* against a log of eighty. The common cause is the same each time: **a number typed into prose by a writer who knew it, in a folder whose own rule is that a claim must be checkable.**

So the fix is not the three edits. **`gen_insurance.py` now gates it** (GM-D81): a doctrine document's *memo N of M* is checked against the manifest, and only the highest-numbered memo may be called the last one.

> **The gate caught real text on its first run** — the correction note in doctrine 08, which recorded the false claim by restating it. The note was reworded rather than the gate loosened.

### 2 · The tier was identified with the wrong quantity

**The most consequential finding.** [Doctrine 04](../insurance/why-insurance-and-what-broke-it.html) §3 took memo 4's observation — an incident response team reduces the *impact* of incidents, not their number — and answered it with the enforcement tier, claiming the estate *"has been computing that quantity since v0.1.28 without naming it."*

**It has not.** Insurance splits a loss into **how often** and **how bad**. The incident-response team is squarely the second: the incident happened, and the team makes it smaller. The tier is **neither** — attempts are unchanged, but fewer of them succeed. That is closer to frequency than to severity.

> **The tier ranks reachability. That is blast-radius reduction, not severity reduction** (GM-D82, narrowing GM-D52).

Still the useful half, because for an agent placement the blast radius *is* most of the severity. **But it is a proxy, and the document now says so** rather than claiming the memo's quantity under a different name. A rating built on the original wording would have credited a boundary control for post-incident recovery capability it has nothing to do with.

### 3 · An over-claim of immunity, contradicted three documents later

[Doctrine 02](../insurance/the-ecosystem-and-the-gate.html) §5 said stage 1 is **structurally immune** to moral hazard, because there is no payout to offload into.

**Doctrine 09 contradicts it without noticing.** Its guard-rail — *make your agents insurable must not become make your agents look insurable* — is moral hazard with no carrier anywhere near it. A level is a badge, and *we are a level 2* is available as a reason to stop looking.

> **Removing the payout removes one channel and opens another** (GM-D84). The rating's channel is cheaper to police, because a derivation anybody can recompute is a badge anybody can dispute — which is why the rule in doctrine 00 is not optional.

### 4 · A question dropped between brief and doctrine

Memo 3 asks: *"is Claude that actually owns that risk — is Claude going to actually have a policy on its own to cover any mistakes?"* [Brief v0.33.74](v0.33.74__strategy-brief__who-pays-for-the-delta-nobody-chose-and-the-rating-that-moves-overnight.md) §9 answered it. **Doctrine 03 dropped it entirely.**

That is a real loss, because [doctrine 06's](../insurance/the-broker-market.html) central claim — a broker asserting *we reduce your level* must carry its own policy or the claim is unbacked marketing — **is the same argument, and memo 3 made it three memos earlier about a different party.** The doctrine read as though memo 6 invented it.

Restored as [doctrine 03](../insurance/who-pays-and-the-moving-rating.html) §8, with the generalisation the two memos jointly support:

> **Any party claiming to reduce somebody else's exposure should carry cover against being wrong about it** (GM-D83). Broker, platform, or model vendor — the argument does not care which.

Flagged as speculation about the vendors, who disclaim liability today. **The structural point is not speculation**: a vendor offering such cover would be making the first falsifiable quality claim in the category, because a benchmark is not a number anybody pays when they are wrong.

### 5 · A forward reference that never landed

[Brief v0.33.73](v0.33.73__strategy-brief__the-ecosystem-without-the-money-insurance-as-a-go-live-gate.md) flagged memo 2's ambiguous *"rider"* and promised: *"the endorsement reading turns out to describe something this estate has already done, in §6."* **Section 6 is about the control-to-premium loop and never mentions endorsements.**

The promise was worth keeping, because the reading is right. An endorsement amends cover after issue without rewriting the policy — which is the register's rule since v0.1.26:

> **An endorsement is an append** (GM-D85). The estate held the endorsement mechanism before it had a policy to attach one to.

[Doctrine 10](../insurance/the-schemas-and-the-clocks.html) reached the same place from memo 10 — *amending, activating, ending: appends, never edits* — three memos later and independently. **Now delivered in [doctrine 02](../insurance/the-ecosystem-and-the-gate.html) §7**, where it was promised.

### 6 · One transcription artefact never flagged

Every memo in the series had its artefacts flagged when it was read — *redempting* → indemnity, *instance response* → incident response, *finibash* → Fibonacci, *rider* discussed at length. **Memo 3's was missed.** It reads *"the risks that exists outside the main"* in a passage whose subject is what happens outside the **mandate**; the sentence before it names *"the authorised one, i.e. the mandates."*

Flagged in doctrine 03 §1, **not repaired in the transcript**, per the rule that the memo is the source.

## One Correction That Was Only Half Made

[Doctrine 06](../insurance/the-broker-market.html) opens by announcing it corrects doctrine 03 §6. **Doctrine 03 §6 carried no notice that it had been corrected**, so a reader arriving there got the superseded claim with nothing to warn them. A superseded notice now stands in both places.

> **A correction only the correcting document knows about is not a correction.**

## What This Audit Did Not Do

- **It did not touch a transcript.** Not one word of any memo was changed. The artefact in §6 is flagged in the doctrine, not repaired at source.
- **It did not re-open settled decisions.** GM-D54 is the project lead's and stands untouched.
- **It did not check the readings against the world.** Every defect found here is internal — an arithmetic error, a category error, a dropped thread. **The one thing that could establish whether any of this is right about insurance is still the survey, and it still has not been run.**
- **It did not audit the pre-pivot corpus.** Only the four quotations the insurance documents rely on were verified, not the documents they came from.

## Decisions This Implies (proposed into change control)

| # | Decision | Status |
|---|---|---|
| GM-D81 | **A printed count is a claim and is derived, never typed.** `llms.txt` reads the decision count out of the log; *memo N of M* is gated against the manifest; only the highest-numbered memo may be called the last | Proposed — **three counts were stale when the gate was written, and it caught a fourth on its first run** |
| GM-D82 | **The enforcement tier ranks reachability, not severity.** Blast radius is used as a proxy for severity and is named as one | Proposed — **narrows GM-D52** |
| GM-D83 | **Any party claiming to reduce somebody else's exposure should carry cover against being wrong about it** — broker, platform or model vendor alike | Proposed — restores memo 3's question, which GM-D58 later asked of brokers only |
| GM-D84 | **Removing the payout removes one channel of moral hazard and opens another** — a level is a badge, and GM-D72's *look insurable* hazard needs no carrier | Proposed — **corrects doctrine 02 §5** |
| GM-D85 | **An endorsement is an append** — memo 2's *rider*, read the second way at last | Proposed — brief v0.33.73 promised this connection and never made it |

## Open Questions, The Project Lead's

1. **Do the audit's five decisions stand?** GM-D82 and GM-D84 narrow or correct earlier proposals of mine rather than adding to them, and both make this folder claim slightly less than it did yesterday.
2. **Should the briefs be corrected too, or only the doctrine?** The briefs say *memo N of 8*, which was true the day each was written. I have treated them as dated records and left them, and corrected only the doctrine, which is the living layer. **Say if you would rather they were superseded.**
3. **Still open, and unchanged by this audit:** N23 the world model, N24 the market survey, N25's three deliverables — and the two questions memo 10 left, being what finishes its truncated sentence and whether a warranty breach voids, suspends or downgrades.

---

*CC BY 4.0. Sources: the eleven memo transcripts as filed in briefs v0.33.71–81; the eleven doctrine documents in `insurance/src/`; briefs v0.33.60, v0.33.63 and `packs/registry-mvp/src/11__observability.md` for the four quotation checks. **There is no voice memo behind this brief** — it is the site agent's audit of its own readings, and every finding in it is mine rather than the project lead's.*