feat(engine): port slop-scoring to gittensory-engine for local pre-submission prediction

[JSONbored]

Found during a 2026-07-08 miner-roadmap audit: the single highest-value unextracted module for the miner's self-review goal.

predicted-gate.ts documents this exact gap in its own note text (PREDICTED_GATE_NOTE_SLOP): "The slop score is NOT evaluated pre-submission (it needs the diff content)." That's a correct limitation for the hosted metadata-only API — it never has the diff on hand pre-submission. It is NOT an inherent limitation for a local miner CLI, which has the full diff on disk before it ever submits. Without this, a miner can predict most of what the live gate will do, but not the one signal most directly tied to the roadmap's own #1 strategic risk ("slop-at-scale paradox" — an autonomous fleet maximizing score mass-produces gate-passing-but-low-value PRs).

Dependency audit of src/signals/slop.ts (~25KB): imports from ./engine (2 named symbols only — GENERIC_COMMIT_PATTERN, hasClearNoIssueRationale, out of the much larger un-extracted src/signals/engine.ts), ./local-branch (symbols already portable), ./test-evidence (already extracted), ./focus-manifest (already shimmed), and ./path-matchers (extraction filed separately, see the sibling engine-extraction issues). It's closer to extractable than its size suggests — genuinely blocked on 2 named symbols plus the path-matchers prerequisite.

Flagged maintainer-only specifically for the final port, not the mechanical prerequisites: a subtly-wrong port could make a miner predict "clean" when the live gate would score high and auto-close — this is scoring-integrity-sensitive in a way the pure extraction prerequisites aren't.

Deliverables

• Prerequisite: path-matchers.ts extraction lands first (separate issue)
• Lift or port GENERIC_COMMIT_PATTERN and hasClearNoIssueRationale from src/signals/engine.ts to gittensory-engine (narrowly — not the rest of engine.ts, see the separate 7-builder extraction tracking issue for that)
• Port slop.ts's pure assessment logic to packages/gittensory-engine/src/slop.ts, exported from the package's public entrypoint
• src/signals/slop.ts becomes a thin re-export shim, matching the existing pattern (src/scoring/preview.ts, src/rules/predicted-gate.ts)
• Golden-fixture regression tests proving the ported version produces byte-identical scores to the pre-port src/ version across a representative sample of real historical PRs
• Update predicted-gate.ts's PREDICTED_GATE_NOTE_SLOP note text once a miner can actually evaluate this pre-submission

References

• src/signals/slop.ts (source to port)
• src/rules/predicted-gate.ts (PREDICTED_GATE_NOTE_SLOP — the gap this closes)
• packages/gittensory-engine/src/signals/test-evidence.ts, src/scoring/preview.ts (existing shim precedent to follow)
• Roadmap's "feat(docs): add install site and mcp diagnostics #1 strategic risk — slop-at-scale paradox" (project memory, Autonomous Miner System roadmap)

[linear-code]

JSO-3237

[JSONbored]

Closing as already delivered — every deliverable on this checklist was completed by #5133 (merged as e7d95a904 / #5140, "build a real production runSlopAssessment implementation"), which landed before this issue was ever referenced in any commit:

• Prerequisite path-matchers.ts extraction: done — packages/gittensory-engine/src/signals/path-matchers.ts is the real implementation; src/signals/path-matchers.ts is a thin shim.
• GENERIC_COMMIT_PATTERN / hasClearNoIssueRationale lifted to the engine package: both are in packages/gittensory-engine/src/signals/slop.ts's export set.
• slop.ts's pure assessment logic ported to the engine package (at packages/gittensory-engine/src/signals/slop.ts, not literally .../src/slop.ts as this issue's text guessed — same outcome, different path) and exported from the package's public entrypoint (packages/gittensory-engine/src/index.ts:373).
• src/signals/slop.ts is a thin re-export shim, matching the test-evidence.ts/predicted-gate.ts precedent.
• Byte-parity verified: the Build a real production runSlopAssessment implementation #5133 commit message documents "all 63 pre-existing tests pass unchanged, and every extracted function diffed identical against the original," and test/unit/miner-slop-assessment.test.ts asserts runSlopAssessment(input) (the miner's binding) equals buildSlopAssessment(input) (the engine export) as a standing CI regression check — not a literal historical-PR-corpus fixture, but the shim architecture makes drift structurally impossible (src/signals/slop.ts has no implementation left to diverge).
• Wired into the miner's real pre-submission path: packages/gittensory-miner/lib/slop-assessment.js wraps buildSlopAssessment, injected into attempt-cli.js's deps and consumed by the self-review pass inside runIterateLoop (create → score → self-review → decide), strictly before any open_pr write — this is real diff-content slop prediction before submission, not the metadata-only kind.

The one item left technically unchecked — predicted-gate.ts's PREDICTED_GATE_NOTE_SLOP note text — is correct as-is and shouldn't change: that note describes the metadata-only, pre-PR-exists predictor's own structural limitation (no diff content available at that stage), which is unrelated to and unaffected by the miner's separate, later, real-diff self-review pass having slop scoring now. Nothing to update there.

No PR needed since the work is already on main.