midnight-fact-check:fact-check-reporting

Report Generation

Placeholder Reference

Placeholder	Source	Example
[run-short-name]	Run directory basename	fast-run-core-concepts-qoMk
[timestamp]	ISO 8601 timestamp from run metadata	2026-04-02T12:00:00Z
[target descriptions]	Comma-separated list of input file paths or skill names	compact-core/skills/compact-tokens/SKILL.md
[full run directory path]	Absolute path to run artifacts directory	.midnight-expert/fact-checker/04-02/fast-run-...
[claim text]	The claim field from the extracted claim object	persistentHash returns Bytes<32>
[domain]	The primary_domain from classification	compact
[evidence_summary]	The evidence summary from the verification result	Compiled and executed — hash length is 32 bytes
[file:lines]	Source file path and line range	SKILL.md:42-44
[qualifier]	Verification method used	tested, source-inspected, type-checked
claim-XXX	Sequential claim ID from the extraction order	claim-001

Truncation Rules

When truncating text for terminal or issue titles:

• Truncate at the last word boundary before the character limit
• Append ... (the limit includes the ellipsis)
• Terminal claim text: 80 characters
• Terminal evidence: 100 characters
• Issue title claim text: 60 characters

Edge Cases

• Zero refuted claims: Omit the "REFUTED CLAIMS" section from the terminal summary. The report still includes all domains.
• Zero claims in a domain: Omit that domain section from the report entirely.
• Missing evidence summary: Use "No evidence collected" as the fallback.
• Zero total claims: Report an empty summary table and note "No testable claims found in the provided content."

Domains

The six domain categories used in reports:

• Compact — Compact language syntax, types, stdlib, compiler behavior
• SDK — TypeScript SDK packages, APIs, DApp development
• ZKIR — Zero-knowledge intermediate representation, circuit structure
• Witness — TypeScript witness implementations, type mappings
• Cross-Domain — Claims tagged by multiple domain classifiers
• Unclassified — Claims no domain classifier tagged (reported as inconclusive)

Markdown Report Template (report.md)

Generate a report using this structure:

# Fact-Check Report: [run-short-name]

**Date:** [timestamp]
**Source:** [target descriptions]
**Run ID:** [full run directory path]

## Executive Summary

| Verdict | Count |
|---------|-------|
| Confirmed | N |
| Refuted | N |
| Inconclusive | N |
| **Total** | **N** |

## Refuted Claims

_List refuted claims first — these are the actionable findings._

| # | Claim | Domain | Evidence |
|---|-------|--------|----------|
| claim-XXX | [claim text] | [domain] | [evidence_summary] |

## Results by Domain

### Compact (N claims)

| Verdict | Qualifier | Claim | Source | Evidence |
|---------|-----------|-------|--------|----------|
| Confirmed | tested | [claim] | [file:lines] | [evidence] |
| Refuted | tested | [claim] | [file:lines] | [evidence] |

### SDK (N claims)

[Same table format]

### ZKIR (N claims)

[Same table format]

### Witness (N claims)

[Same table format]

### Cross-Domain (N claims)

[Same table format]

### Unclassified (N claims)

[Claims that no domain classifier tagged — listed as inconclusive]

Verdict Indicators

Use these badges in terminal output and issue titles:

• Confirmed → [CONFIRMED]
• Refuted → [REFUTED]
• Inconclusive → [INCONCLUSIVE]

Terminal Summary Format

Print this to the terminal after writing the report:

═══════════════════════════════════════════
  Fact-Check Complete: [run-short-name]
═══════════════════════════════════════════

  Confirmed:    NN
  Refuted:      NN
  Inconclusive: NN
  ─────────────
  Total:        NN

  [If refuted > 0:]
  REFUTED CLAIMS:
    - [claim-XXX] [claim text (truncated to 80 chars)]
      Evidence: [evidence_summary (truncated to 100 chars)]
    - [claim-YYY] ...

  Artifacts: [full run directory path]
  Report:    [path to report.md]
═══════════════════════════════════════════

GitHub Issue Templates

Per-Claim Issue

Title: [REFUTED] [claim text (truncated to 60 chars)]

Body:
## Refuted Claim

**Claim:** [full claim text]
**Source:** [file path, line range]
**Domain:** [domain]

## Verification Evidence

**Verdict:** Refuted ([qualifier])
**Evidence:** [full evidence_summary]

## Context

This claim was identified by the midnight-fact-check pipeline.
- **Run:** [run directory]
- **Source file:** [link to file if GitHub URL available]

---
_Generated by midnight-fact-check_

Per-File Issue

Title: Fact-check findings: [filename] ([N] refuted claims)

Body:
## Fact-Check Results for [full file path]

[N] claims were refuted in this file.

### Refuted Claims

- [ ] **[claim-XXX]:** [claim text]
  - Evidence: [evidence_summary]
  - Line(s): [line_range]

- [ ] **[claim-YYY]:** [claim text]
  - Evidence: [evidence_summary]
  - Line(s): [line_range]

## Run Details

- **Run:** [run directory]
- **Total claims checked:** [N]
- **Confirmed:** [N] | **Refuted:** [N] | **Inconclusive:** [N]

---
_Generated by midnight-fact-check_

Summary Issue

Title: Fact-check report: [N] refuted claims across [M] files

Body:
## Fact-Check Summary

| File | Refuted | Confirmed | Inconclusive |
|------|---------|-----------|--------------|
| [file1] | N | N | N |
| [file2] | N | N | N |

### All Refuted Claims

| # | File | Claim | Evidence |
|---|------|-------|----------|
| claim-XXX | [file] | [claim] | [evidence] |

## Run Details

- **Run:** [run directory]
- **Full report:** [path to report.md]

---
_Generated by midnight-fact-check_