# Control coverage: observed statuses, evidence lineage and reviewer judgment — observed output evidence

**Provenance and scope:** Two real historical JSON exports from different development rounds. Private originals remain withheld because they contain client identifiers and accounting details.

Original excerpts below are reproduced verbatim (original Portuguese). Public references use anonymized source IDs; raw original documents are retained privately and are **not distributed**. SHA-256 identifies each retained private byte sequence but does not authenticate accounting facts.

## SC-A — Historical Core 0.3.2 — Web Lite JSON
- **Original field/page:** `round.summary`
- **Original SHA-256:** `1e8893e0437b83edfd014ab92e2d7689cb0174dc7069099cc35e955c7c44f2b9`
- **Exact archived statement:**
> Reconciliação mensal sem diferença; três alertas em acompanhamento; Enhanced não executado por falta de evidências.

## SC-B — Historical Pilot 0.3.8 — Web Lite JSON
- **Original field/page:** `round.summary`
- **Original SHA-256:** `b8be3bce5d7ec9e972105eb513b36567b9eb61f8da4d86e3f9041e2ece0ff378`
- **Exact archived statement:**
> Reconciliados Razão/Balancete e novos exames de natureza e alterações de conta. Controles com alcance limitado não são certificação econômica nem fechamento Core completo.

## Evaluation provenance boundary

- These are real archived development outputs, not invented A/B sample completions.
- The public package does not include the private financial records, raw provider completions, original user prompt or inference parameters.
- A and B must not be interpreted as two controlled responses to the same prompt, and are not assigned a model-performance score.

- Numerical claim-level subscores shown on the webpage are **simulated expert adjudications** based on the quoted evidence, not measured LLM performance. E03 provides the scoring checkpoints and explicitly unverified items.