Plagiarism / copycat detection

[JSONbored]

Phase 4 — Self-host-native differentiators

Context

Duplicate clustering detects overlapping scope, not code similarity or attribution (src/signals/duplicate-winner.ts). No copycat detection exists — a natural sibling of anti-slop.

Goal

Per-repo configurable plagiarism/copycat detection.

Requirements

• A deterministic containment/similarity score between a PR's added code and prior art (repo history / other PRs), with direction-by-timestamp (earlier submission = victim). Tiered response: warn → label → block → strikes, all config-driven (gate.copycat mode + thresholds). No scoring labels.
• Bounded, fail-safe, precision-first (avoid false accusations); public-safe messaging.

Deliverables

• Similarity engine + direction logic + tiered gate wiring + tests + docs.

Expected outcome

• Detects and deters copied contributions; strengthens the anti-slop moat.

Effort

XL.

[linear-code]

JSO-2049

[JSONbored]

2026-07-01 audit: folded #1409 into this issue — its containment-score + direction-by-timestamp + tiered warn/label/block/strikes is this issue's design spec.