Evidence model

Every evidence item is bound to a pinned commit, a file path, a bounded excerpt and a strength. Evidence without a source is not evidence.

Weighting

Signals are weighted by rarity. A distinctive domain term outranks a common one; a longer compound token is less likely to collide by accident. Scores are bounded and are used to rank, never shown as a probability.

Strength buckets

• HIGH — a signal that is both strong and rare.
• MEDIUM — a signal that is suggestive but has an innocent explanation.
• LOW — a signal that mostly indicates a shared ecosystem.

ForkReason has no calibration methodology, so confidence is reported as a bucket and never as a probability. A number would imply a precision that does not exist.

The manifest

The evidence manifest is the canonical, bounded set submitted to consensus. It contains the pinned repositories and commits, chronology facts, evidence items with strength, conflicting signals, every alternative explanation, and shared-upstream candidates. It is canonicalized (sorted keys, ordered collections, quantized numbers) before hashing.

manifest_hash = SHA-256(canonical_json(manifest))

Anyone with the same two commits can recompute the hash and confirm nothing was altered between analysis and record.