# Claim Audit - CAUM Sample Fix Receipts

Generated for the public `/audit/` page.

## What is asserted

- The comparison validation run did not produce a hard alert.
- The controlled retry-loop validation run produced a hard alert and reached T5 critical structural evidence.
- The controlled run had one exact cycle interval, exact-cycle coverage of 1.0, 12 core steps, 1,356 observed tokens, and $0.018 observed cost.
- The cost values are observed counters and reviewable exposure, not guaranteed savings.

## What is not asserted

- CAUM does not judge whether the agent output was true or false.
- CAUM does not predict future outcomes.
- CAUM does not block or control agents.
- CAUM does not require prompts, source code, customer messages, or private business payloads.
- These receipts do not publish a universal waste rate.

## Metric classification

- Public: hard alert status, T5 critical structural evidence, exact-cycle coverage, cycle interval count, observed tokens, observed cost, reviewable cost exposure.
- Internal/restricted: per-agent tier distribution as a prevalence or market-wide waste rate.

## Evidence

- `AGENT_VALIDATION_20260501_164309/summary.json`
- `AGENT_VALIDATION_20260501_164309/website_smoke_agent.json`
- `AGENT_VALIDATION_20260501_164309/controlled_retry_loop_agent.json`
- `AGENT_VALIDATION_20260501_164309/public_readout/claim_audit.json`

## Risk

The main risk is readers confusing reviewable cost exposure with guaranteed savings or treating the comparison run's T3 review-only signal as confirmed waste. The page wording keeps those claims separate.

## Before / after validation addendum

Generated: 2026-05-02.

What is asserted:

- A controlled retry-loop validation run reached T5 with hard_alert=true, exact-cycle coverage 1.0, 1,356 observed tokens, and $0.01545944 reviewable cost exposure.
- A guarded remediation-pattern validation run reached T2 with hard_alert=false, exact-cycle coverage 0.0, 281 observed tokens, and $0.00039349 reviewable cost exposure.
- The after run used a retry ceiling, route-existence precheck, fallback route, and checkpoint reset pattern.

What is not asserted:

- This is not a universal savings guarantee.
- This is not a market-wide waste rate.
- This does not prove answer truth, future task success, or semantic correctness.
- This does not mean CAUM blocked or controlled the agent. CAUM observed and reported only.

Evidence:

- `before-after-summary.json`
- `before-controlled-retry-loop-20260502.json`
- `after-retry-guard-20260502.json`