# Faultprint v14 — External Validation Pass 002 Prompt

You are acting as an independent external validator for Faultprint. Your job is not to praise the system. Your job is to find weaknesses, mislabels, overclaims, bad gates, unsupported anchors, and any row that should not be client-visible.

## Files To Review

Primary sample:
- FAULTPRINT_EXTERNAL_VERIFICATION_PACKET_V14_SAMPLE_001.csv

Supporting files:
- FAULTPRINT_EXTERNAL_VERIFICATION_PACKET_V14_README_001.md
- FAULTPRINT_EXTERNAL_VERIFICATION_PACKET_V14_SOURCE_CHECK_001.md
- FAULTPRINT_EXTERNAL_VALIDATION_PACKET_V14_SOURCE_LIST_001.csv
- FAULTPRINT_V14_FULL_BASELINE_CALIBRATION_SUMMARY_001.md
- FAULTPRINT_V14_FULL_BASELINE_CONSISTENCY_AUDIT_001.md
- FAULTPRINT_OPERATING_POSTMORTEM_CROSSWALK_014.csv
- FAULTPRINT_CLIENT_OUTPUT_ELIGIBILITY_012.csv

## Context

Faultprint is an institutional failure intelligence engine. It does not claim that an institution has failed. It identifies structural conditions in operating documents and compares those conditions against postmortem failure libraries.

The current v14 baseline has already gone through internal source-stress and consistency checks. Your job is to independently test whether that internal hardening actually holds.

## Review Unit

Review each row as a complete unit:

condition_name + evidence_anchor + source_url_or_path + primary_pathology + resemblance_band + client_output_status + allowed_claim + falsification_evidence_required + boundary + source_verification_auto_verdict

## Decision Labels

Use exactly one decision label for every reviewed row:

- pass
- revise_label
- revise_band
- revise_gate
- hold_for_evidence
- remove

## Required Checks

For each row, answer these questions:

1. Does the evidence anchor appear supported by the named source/document?
2. Is the evidence anchor source content, not analyst inference?
3. Is the primary pathology the narrowest defensible label?
4. Is the resemblance band justified by the stated evidence and support?
5. Is the client-output status appropriate?
6. Is the allowed claim bounded, non-predictive, and falsifiable?
7. Does the boundary state what the row does not prove?
8. Is the falsification evidence concrete enough to test the row?
9. If the row is null/absence/process-only, is it held internally rather than client-visible?
10. If the source is fetch-limited, token-limited, secondary, or partial, is the row caveated or held appropriately?

## Strictness Rules

Do not pass a row if:

- the anchor is an analyst proposition rather than source-supported content;
- the anchor cannot be found or reasonably confirmed in the cited source;
- the boundary widens the claim instead of limiting it;
- the row is client_ready but source support is partial, blocked, secondary, token-limited, or fetch-limited;
- the row predicts failure, asserts illegality, gives recommendations, or claims institutional wrongdoing;
- a null/absence row is client-visible;
- the pathology label is directionally wrong;
- the resemblance band is stronger than the evidence can carry.

## Confidence Rules

Use confidence labels:

- H = high confidence
- M = medium confidence
- L = low confidence

If you do not fetch the source, cap confidence at M.
If the source is fetched and the anchor is not found, use hold_for_evidence unless the row is already internal and clearly marked as limited.
If the source is dynamically rendered, blocked, or partially inaccessible, state that clearly and do not overstate the defect.

## Output Format

Return a report with these sections:

1. Coverage Disclosure
   - Sources fetched
   - Rows directly checked against source
   - Rows not fetched
   - Any confidence caps used

2. Decision Summary
   - Counts by decision label
   - Rows by decision label

3. Systemic Findings
   - Gate issues
   - Label issues
   - Band issues
   - Anchor/source issues
   - Boundary issues
   - Null-row issues
   - Falsifiability issues

4. Row-by-Row Verdict Table

Use this table schema:

| library_row_id | decision_label | what_you_checked | issue_if_any | recommended_change | confidence |
|---|---|---|---|---|---|

5. Bottom Line
   - Is v14 materially stronger than v7/v11?
   - What must be fixed before client use?
   - Which issues are blockers versus calibration improvements?

## Important Boundary

This validation checks Faultprint engine output only. It should not make findings about any named institution beyond whether the row is source-supported, bounded, and properly gated.