Skip to main content
< All Topics
Print

Claims Evidence Registry

name: claims-evidence-registry

description: Build and maintain a claims evidence registry that maps every outward-facing quantitative or capability claim to its supporting evidence. Use when creating a CLAIMS-EVIDENCE.md for a project, during monthly staleness checks, or when a new quantitative claim is added to any ITI document.

Claims Evidence Registry

Instructions

Create and maintain a structured registry that maps every quantitative, capability, or comparative claim in ITI’s outward-facing materials to its supporting evidence. The registry is the single source of truth for “can we defend this claim?”

Registry Format — CLAIMS-EVIDENCE.md

Create this file at the project or portfolio level:


# Claims Evidence Registry — [Scope]
**Last Updated**: YYYY-MM-DD
**Next Review**: YYYY-MM-DD (90-day cycle)
**Owner**: [Name or agent ID]

## Active Claims

| ID | Claim | Documents | Evidence Type | Evidence Location | Last Verified | Status |
|----|-------|-----------|---------------|-------------------|---------------|--------|
| C-001 | [Exact claim text] | [File paths where it appears] | Commit / Test / Benchmark / Estimate / Self-reported | [Path to evidence] | YYYY-MM-DD | Current / Stale / Disputed |

## Retired Claims
[Claims removed from active materials but kept for audit trail]

| ID | Claim | Retired Date | Reason | Replaced By |
|----|-------|-------------|--------|-------------|

Evidence Types (in descending order of strength)

Type Description Strength Example
Third-party verified Independent test, audit, or certification Highest Security audit report, analyst benchmark
Automated test Reproducible test with logged results High CI test suite pass rates, performance benchmarks
Commit/artifact Verifiable in version control High Git log, file count, dependency list
Measurement Tracked metric with methodology documented Medium-High Token count comparison, timing logs
Estimate Informed projection without formal measurement Medium “~3-4x faster based on developer experience”
Self-reported Assertion without independent verification Low “20+ production AI systems”
Historical/anecdotal Based on memory or single instance Lowest “We built X in Y days”

Creating a New Registry

  1. Inventory claims: Run the claims-integrity-audit skill on all documents in scope
  2. Extract quantitative claims: Pull every number, percentage, count, comparison, and timeline
  3. Map to evidence: For each claim, identify the strongest available evidence
  4. Assign status: Current (verified within 90 days), Stale (not verified in 90+ days), Disputed (conflicting evidence exists)
  5. Set review date: 90 days from creation

Monthly Staleness Check Protocol

Run monthly on every active Claims Evidence Registry:

  1. Pull all claims with Last Verified > 90 days — these are stale
  2. For each stale claim:
  • Can the evidence be re-verified now? If yes, verify and update date
  • Has the underlying fact changed? If yes, update claim text in all documents
  • Is the claim still in active documents? If no, move to Retired
  1. Check for orphaned claims: Claims in documents but not in the registry
  2. Check for phantom claims: Claims in the registry but removed from documents
  3. Produce staleness report:

## Staleness Report — [Date]
- Total active claims: [N]
- Current (< 90 days): [N]
- Stale (90+ days): [N]
- Newly disputed: [N]
- Retired this cycle: [N]

### Stale Claims Requiring Action
| ID | Claim | Last Verified | Action Needed |
|----|-------|---------------|---------------|

Claim Lifecycle


New claim in document
    ↓
Extract and register (assign ID, evidence, status)
    ↓
Verified → Active (90-day review cycle)
    ↓
Stale? → Re-verify or update or retire
    ↓
Disputed? → Escalate to Claims Ombudsman for resolution
    ↓
Retired → Move to retired section with reason

ITI-Specific Evidence Locations

Claim Type Where to Find Evidence
Agent counts Agents/AGENTS-INDEX.md (indexed); agents-and-skills.md Section 7 (total unique)
Skill counts Skills/SKILLS-INDEX.md (ITI + platform breakdown)
Product counts coding-playbook.md product tables (with status column)
Speed claims Git commit history (before/after timelines); session logs
Token savings Token count logs from ITI_Claude_API usage tracking
Architecture claims Actual file structure (ls -R); wc -l for LOC claims
Version claims package.json, Cargo.toml, plugin headers
“Production” status Deployment evidence: live URLs, App Store listings, plugin directory listings

Outputs: CLAIMS-EVIDENCE.md registry file, monthly staleness report, claim lifecycle tracking, orphan/phantom detection.

Table of Contents