cross-source-fact-verification

Original：🇨🇳 Chinese

Translated

Use this method when fact-checking drafts that include dates, quantities, or causal claims by cross-referencing multiple independent sources.

1installs

Sourceshimo4228/claude-code-learned-skills

Added on2026-02-14

NPX Install

npx skill4agent add shimo4228/claude-code-learned-skills cross-source-fact-verification

SKILL.md Content (Chinese)

View Translation Comparison →

Cross-Source Fact Verification

Extracted: 2026-02-14 Context: A pattern for verifying factual claims in pre-writing drafts using multiple independent sources

Problem

Drafts and context files are prone to containing memory-based inaccuracies.

Date errors (e.g., misremembering January 31st as February 8th)
Missing counts (e.g., recording 3 inventory checks as 1)
Inaccurate figures (e.g., overcounting or undercounting the number of skills)

Errors that cannot be detected by checking a single source can be identified through cross-referencing multiple sources.

Method

Step 1: Identify Claims to Verify

Extract verifiable factual claims from the draft:

Dates: "Did X on month X day Y"
Quantities: "N skills...", "Conducted N times"
Sequence: "Did B after A"
Causality: "B happened because of A"

Step 2: Map Evidence Sources

Identify independent sources that can be used to verify each claim:

Source Type	Location	Specialized Verification
Debug Logs	`~/.claude/projects/*/`	Exact timestamps, tool calls, errors
MEMORY.md	`~/.claude/projects/*/memory/`	Inventory results, decision-making, numerical values
Git History	`git log --oneline`	Commit timestamps, change details
File Timestamps	`ls -la` , `stat`	Creation/modification timestamps
Published Articles	`articles/*.md`	Consistency with previously published content
Research Drafts	`drafts/*.md`	Raw data from research phase
Skill Files	`~/.claude/skills/learned/`	Extracted dates, content

Step 3: Parallel Search

Simultaneously search independent sources using parallel sub-agents:

Launch in parallel:
  Agent 1: Extract dates and operations from debug logs
  Agent 2: Collect records from MEMORY.md (all projects)
  Agent 3: Investigate git history and file timestamps
  Agent 4: Verify consistency with published articles and drafts

Step 4: Cross-Reference

Cross-reference results from multiple sources:

Consistent → Accept as fact
Conflicting → Prioritize more reliable sources (Debug Logs > Memory-based MEMORY > Drafts)
Single source only → Conduct additional research for supporting evidence; if none exists, accept with a note

Reliability Priority Order:

Debug Logs & Git History (machine-recorded, hard to tamper with)
File Timestamps (OS-level records)
MEMORY.md (recorded at session end, may have memory bias)
Published Articles (published but may have writing bias)
Drafts & Oral Accounts (highest risk of memory bias)

Step 5: Build Verified Timeline

Reconstruct a timeline using only confirmed facts, and clearly indicate differences from the original draft.

When to Use

When fact-checking context files before writing articles
When drafts contain numerical claims such as "when, how many times, how many"
When organizing events across multiple projects in chronological order
When there may be discrepancies between a user's memory and actual records

Anti-Patterns

Single Source Dependency: Relying solely on MEMORY.md for fact-checking
Accepting Memory at Face Value: Adopting a user's oral account without verification
Sequential Search: Checking independent sources one by one (should be parallelized)
Ignoring Conflicts: Leaving discrepancies between sources unresolved