Find competing answers before your AI learns the wrong one.

What it helps you do

• Inventory each source with its title, date, owner and intended audience
• Surface passages that may answer the same operational question differently
• Keep legitimate exceptions—such as region, product or customer type—separate from true contradictions
• Record owner-approved decisions without assuming the newest file automatically wins
• Flag dependent answers for revalidation when a source changes

What you receive

• Codex skill 0.1.2 and corresponding Claude skill 0.1.2
• Getting Started plus Codex and Claude install/use guides
• Five editable CSV templates for sources, conflicts, owner decisions, supported answers and evaluation
• Illustrative review with traceable passages and staged owner decisions
• Fictional six-source practice corpus and worksheet guide

The owner stays in charge

The workflow proposes potential conflicts and organizes the evidence. It does not choose policy, delete or update sources, change a live assistant, or turn a plausible inference into an approved answer.

Use approved redacted copies. Keep secrets and unnecessary personal information out of the working folder, preserve the originals, and review every proposed resolution before relying on it.

Tested scope and limits

Codex: native PASS on the exact 0.1.2 edition for up to six UTF-8 text or Markdown sources, 6,000 words total and 1 MiB per file. Tests covered staged owner decisions, changed-source revalidation, saved Markdown and CSV outputs, exact/over-limit behavior, unsupported or empty inputs, and cleanup.

Claude: included under the approved corresponding-edition release rule, but not natively tested. Shared files are aligned; no Claude PASS is claimed.

These are validated input boundaries—not guarantees of accuracy, completeness, compliance or hallucination-free AI.