docs(incorporations): forking plans for 5 audio-MIR discoveries (ASA + Harmonia)

[slittycode]

Summary

Adds incorporations/forking-plans-2026-05-14.md — five concrete forking / incorporation plans for upstreams from the 2026-05-13 / 2026-05-14 audio-MIR discovery passes. Three ASA-facing, two Harmonia-facing. Plan-only — no implementation.

Mirrors the seven-part track structure of docs/external-repo-review-2026-05-13.md (Source / What to lift / Why / Approach / Cross-check oracle / Definition of done / Risks), and extends the prior review's Track-1–3 numbering with Plans 4–8.

#	Upstream	Serves	License	Recommended approach
4	JuzzyDee/audio-analyzer-rs	ASA + Harmonia	MIT	Cross-check oracle first; fork-to-WASM gated on a real in-browser need
5	jhartquist/resonators	ASA	Dual MIT / Apache-2.0	Incorporate as a dependency (npm install resonators) — the near-drop-in
6	linuxmatters/jivetalking	ASA	GPL-3.0-only	Clean-room port of the measure-then-derive-parameters pattern into Phase 2; no source-reading
7	ptnghia-j/ChordMiniApp	Harmonia	MIT	Reference-only UX + architecture review; explicit adopt / adapt / reject per surface
8	spyroskantarelis/chordonomicon	Harmonia	Apache-2.0 scripts + CC-BY-NC-4.0 dataset	HF dataset eval harness; NonCommercial license gates product inclusion

Composition & cross-references

• Plans 4 ↔ 5 — both Rust → WASM stories on paper. Plan 5 is the cheap end state (published WASM dependency, zero toolchain); Plan 4b (fork-to-WASM) is gated on a need that doesn't exist today. Don't pre-fork.
• Plan 6 ↔ existing review's Track 1 — jivetalking's measurement inputs come from ASA's existing Phase 1 (which the EBU R128 spike confirmed correct on stereo at 44.1 / 48 kHz), not from a new measurement layer. Plan 6 does not duplicate Track 1.
• Plan 7 ↔ Plan 8 — Plan 7 sets the chord-presentation UX bar; Plan 8 provides labeled ground truth to evaluate against. Compose.

License discipline

All five upstream licenses confirmed against the source. Two carry real risk and make license confirmation item 1 of their DoD:

• jivetalking (GPL-3.0-only): full LICENSE file read; the plan explicitly forbids reading source for the port and routes everything through the README's prose specification.
• chordonomicon (HF dataset, CC-BY-NC-4.0): the dataset card itself was auth-gated to direct anonymous fetch; the license is reported via the HF dataset listing and confirming on the card is DoD item 1. The Apache-2.0 scripts repo is separately confirmed.

Environment note

The task brief referenced a separate routine-discoveries repository with discoveries/*.md source-of-truth files and an asa-2026-05-13.md template. Those files are not present in this checkout (which is scoped to ableton-sonic-analyzer). I checked with the user before proceeding; the plan was written using the task description as the spec, direct upstream research for licenses / APIs / schemas, and docs/external-repo-review-2026-05-13.md as house-style reference. See the "Environment note" paragraph in the document header.

Sequencing recommendation

1. Plan 5 (~1–2d) — lowest friction; new browser-side surface
2. Plan 4a (~1–2d) — cross-check oracle; serves PURPOSE.md Quality Invariant 1
3. Plan 6 (~3–4d) — Phase 2 measurement-cited parameter heuristics
4. Plan 7 (~2d, Harmonia-side) — UX review doc
5. Plan 8 (~3–4d, Harmonia-side, gated on commercial-posture decision) — eval harness

Test plan

This is a plan-only document — no code, no tests to run. Reviewer checks:

• Each plan's Source line cites a real upstream URL with confirmed license
• Each plan's Definition of done is concrete and testable (checkbox list with specific outcomes)
• The Adopt / Adapt / Reject tables (Plans 4 / 5 / 6 / 7 / 8) have a one-line rationale per row
• Plans 4 and 5 cross-reference each other on the Rust → WASM question
• Plan 6 composes with the prior review's Track 1 rather than duplicating measurement
• License risk is called out in the License Map and reflected as DoD item 1 for jivetalking and chordonomicon
• Dedup is intact against the three already-covered tracks (openmeters / Partiels / forever-jukebox)

https://claude.ai/code/session_017Pt2pELLM2qNQxoZrWQt4M

---

Generated by Claude Code

[slittycode]

Verdict: COMMENT

Summary

Docs-only PR — one new file (incorporations/forking-plans-2026-05-14.md, 942 lines). No code, no schema changes, no test impact. The license discipline is careful (GPL-3.0 and CC-BY-NC-4.0 risks are correctly flagged and made DoD item 1), the adopt/adapt/reject tables are well-reasoned, and the sequencing recommendation maps accurately to effort and risk. No phase boundary violations possible in a plan-only doc.

Findings

Should fix

Plan 6 DoD is missing the frontend types contract. The DoD correctly says to update apps/backend/JSON_SCHEMA.md and EXPECTED_TOP_LEVEL_KEYS in test_analyze.py when adding noise-floor estimate and spectral flux to Phase 1. It does not mention updating apps/ui/src/types/measurement.ts. CLAUDE.md tripwire #3 is explicit: renaming or adding a field on only one side of the Python/TS boundary is undetectable by either type system — the field just disappears from the UI. Whoever picks up Plan 6 will hit this if the DoD checklist is followed as written. Add apps/ui/src/types/measurement.ts to that checklist.

Worth considering

Plan 5 DoD doesn't enforce its own stated guardrails. The Approach section says "behind a feature flag" and the Risks section says "The preview is labeled as approximate." Neither appears as a DoD checkbox. Given the risk that users will interpret a resonators-derived browser preview as a Phase 1 measurement (cited in the Risks table, severity: Medium), these two items belong in the DoD as testable conditions, not prose asides.

Plan 5's "after PR #45, Phase 3 already audits heuristic WAV/MIDI" is either wrong or ambiguous. Per BACKLOG.md and CLAUDE.md, patchSmith.ts (Phase 3) is still an open backlog item, not shipped. If PR #45 lands Phase 3, that reference will be fine post-merge; if not, the sentence is describing a future state as a present fact. Either clarify that this is conditional on #45 merging, or remove the parenthetical.

Test results

N/A — plan-only document.

Phase boundary check

Clean. No code. The plans, particularly Plan 5 (browser-side resonators preview vs. server-side Phase 1 measurements), explicitly call out the boundary and the oracle pattern for validating it. The "Phase 1 stays the cited number throughout the UI" principle is correctly restated.

---

Generated by Claude Code