# Faultprint Core Calibration Audit 005

## Purpose

This audit returns to the core engine. It tests whether v7 pathology labels, resemblance bands, and client-output eligibility are coherent enough to support a defensible product.

This is not a report-packaging step. It is an engine-quality step.

## Inputs

- Operating pathology mapping v7: 680 rows
- Operating/postmortem crosswalk v7: 680 rows
- Client output eligibility table: 680 rows
- Pathology support table v3: 12 rows

## Audit Sample

Rows reviewed by rule-based calibration audit: 65

Sampling rule: all high-resemblance rows plus up to three elevated, moderate, and low rows per primary pathology.

## Review Outcomes

- pass: 65

## Issue Register

No issues detected in this audit sample.

## Honest Core Status

The core is working as a prototype, but it is not yet a finished commercial scoring engine.

What is working:

- The pipeline can ingest operating findings, map them to pathologies, compare them to postmortem support, gate client output, and generate reports.
- The output language is bounded, falsifiable, and generally avoids predictions or recommendations.
- The v3 calibration is materially better than v1 and v2 because it reduced inflated high-resemblance rows.

What still needs hardening:

- Some pathology categories have operating evidence but weak postmortem support.
- Some resemblance bands still need reviewer-confirmed downgrade rules.
- The client-output gate needs to be stricter where postmortem support is thin.
- The scoring system is deterministic and needs a human-reviewed calibration set before it can be marketed as robust.

## Next Engine Step

The next engine step is to fix the issue register, especially band/support problems, then rerun v4 mapping/crosswalk/eligibility. Only after that should we scale polished reports across the full report pack.