# Review lineage: *The Verdict as Axiom* v1.5 final Claude Code stage

**Date:** 11 August 2026  
**Manuscript reviewed:** `The_Verdict_as_Axiom_Preprint_v1.5_Claude_Code_Review_Candidate_Markdown.md`  
**Manuscript SHA-256:** `318ca5d182f6203ec1c885da684ed3f168c2e0c423c6ef4131108b7ce71b787d`  
**Frozen review prompt SHA-256:** `29acccbad0f356f495d0440a6afe7523d9aa7a07c3456c0c8bbd939b71e1c134`

## Review objects

| Public file | Model/configuration | Deliberate context design | Epistemic status | Original review SHA-256 | Public-copy SHA-256 | Gate |
|---|---|---|---|---|---|---|
| `CLAUDE_SONNET5_REVIEW_OF_RECORD_v1.5_PUBLIC.md` | `claude-sonnet-5` | New session; no prior manuscript verdict; frozen refusal ledger disclosed by design | Review of record; fresh-session, provider-conflicted, non-blind to editorial history, not peer review | `ea8465d7f35dddfea25ab0e383f49acc91e554767799d18d76d5d621496431f0` | `f639ee3fd8a27013713a8f1de0d82d0868d789a66a31e449d09e185924e1960d` | `READY_AFTER_BOUNDED_REVISIONS` |
| `CLAUDE_FABLE5_CONTEXT_CARRYING_MODEL_SWITCH_v1.5_PUBLIC.md` | `claude-fable-5` | Intentional mid-session model switch after Sonnet; complete Sonnet review, tool results, and source-access history retained | Same-session, context-carrying, verdict-exposed, non-independent comparative probe | `79d91065f4e7da5099000d62e436b8dcaa4b42fc562ae5bc455281aa3985ad78` | `d2c150c723763fd988dce68b8d1e215bdaa374b51c3354b7865b971418cdc583` | `READY_AFTER_BOUNDED_REVISIONS` |
| `CLAUDE_OPUS5_VERDICT_EXPOSED_COMPARATIVE_REVIEW_v1.5_PUBLIC.md` | `claude-opus-5` | Intentional new session seeded with the v1.4 Red/Claude verdict before the frozen prompt | Fresh-session but verdict-exposed, provider-conflicted, non-independent comparative probe | `5d594a16fe7e8599e8d2fb7611cdeb5514ea69b7e781f5d857303e79c1c77f08` | `3a98136887505195e13589f83740b0dba7d73f2d7a281ba854408ca9c91f1029` | `READY_AFTER_BOUNDED_REVISIONS` |

## Why the Fable and Opus configurations were intentional

The Fable and Opus runs were deliberately configured to test a narrower question than independence: **would different model configurations surface materially different defects even when an earlier evaluative context was already present?**

- Fable inherited the complete Sonnet session and was then asked to re-review after a model switch.
- Opus began in a new session but was intentionally given the prior v1.4 Red verdict before receiving the frozen prompt.

The Opus report correctly recognized that its information state violated the prompt's clean-context firewall. It described the exposure as operator error because the experimental purpose was not stated inside that prompt. The final project record corrects the intent without erasing the protocol deviation: exposure was deliberate, but the pass remains verdict-exposed and cannot be counted as independent verification.

This was **not a controlled causal benchmark of model weights**. Model label, context topology, retrieval order, tool history, and stochastic sampling were not held constant. The defensible observation is narrower: inherited context did not force identical review output.

## What differed

- **Sonnet 5**, without a prior verdict, found the unrelated third-party material in the raw supplement screenshot and the Poppler-dependent derived-text hash seam.
- **Fable 5**, with the Sonnet review in context, found that the `MoonUnit97` account-creation interval had been misread as nineteen seconds rather than 5 hours, 10 minutes, and 23 seconds, while arithmetically confirming the separate 5.4-minute X-to-HN relay.
- **Opus 5**, with the earlier v1.4 verdict intentionally supplied, found that Section 5 presented editorially relabeled output as though it were verbatim script output, and that two falsifier limbs were not fully operable from the public supplement.

These divergences are preserved because they are the point of the comparison. Convergence among the three documents is **not consensus evidence** and is counted once. New source-level findings are evaluated on their own receipts and incorporated only where independently checkable.

## Protocol accounting

The frozen prompt designated the first complete review as the review of record. Sonnet therefore remains the review of record. The Fable and Opus runs are supplementary comparative probes produced by explicit operator choice. Their existence does not reopen the review cycle and does not create three votes.

No further manuscript-review or technical-audit pass is planned.