renato97/ableton-mcp-ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 1 Wiki Activity
Files
6d080d43b3c869e644c709f0b13bead55fe66559
ableton-mcp-ai/docs/SPRINT_v0.1.39_NEXT_GLM_PROJECT_EDITING_AND_COHERENCE.md

13 KiB
Raw Blame History

SPRINT v0.1.39 - NEXT FOR GLM

Open-Project Editing, Less Symmetry, Less Silence, More Coherent Freedom

Owner: GLM via OpenCode
Reviewer: Codex
Fecha: 2026-04-03
Mode: Edit existing open project, not generate from scratch

Primary benchmark:

  • C:\Users\ren\Desktop\song Project\song.als

Reference docs:

  • C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\docs\PROJECT_AUDIT_song_2026-04-03.md
  • C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\docs\SPRINT_v0.1.38_VALIDATION_REPORT.md
  • C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\docs\SPRINT_v0.1.37_VALIDATION_REPORT.md

1. Reviewer Summary

v0.1.38 was real progress, but it did not close the product gap.

The current project looks better in Arrangement than older versions, but it still has three musical failures:

  1. too much symmetry
  2. too many silence islands
  3. too little source freedom inside coherent boundaries

The screenshot and live MCP inspection point to the same root pattern:

  • repeated sources on the same grid
  • large empty windows between anchors
  • harmonic MIDI still not carrying continuity in Arrangement

This sprint is not about adding more random generation heuristics. It is about making the open-project editing workflow capable of repairing those failures on a real .als.

2. Multi-Agent Review Findings

Codex reviewed this with parallel code review and live MCP inspection.

2.1 What was true in the report

  • MCP inspection tools are genuinely useful now
  • project editing surface is broader than before
  • audit_project_coherence() exposes real repetition and gap problems

2.2 What was still wrong

  • the benchmark still has HARMONY_* MIDI support only in Session, not Arrangement
  • the project still repeats the same source on multiple tracks with near-mirrored boundaries
  • the current coherence audit was still too lenient for the musical result
  • the rhythmic backbone was being over-consolidated into sparse anchors

2.3 Concrete bugs Codex already fixed after the report

  • arrangement clip creation and session-to-arrangement duplication now fall back more safely if direct track.create_clip(...) fails
  • _send_command_to_ableton(...) compatibility helper was restored so legacy validation tools stop crashing
  • one-shot backbone roles (kick, snare, clap, hat) are no longer candidates for destructive position consolidation
  • section-aware consolidation for loop roles is less grid-locked
  • audit_project_coherence() now:
    • detects harmonic MIDI tracks by is_midi_track, not only fake type == midi
    • ignores non-audio clips when counting repeated audio sources
    • exposes same_source_dominance_tracks
    • scores the current type of project more honestly

These fixes are baseline. Do not re-break them.

3. Product Direction

We are now optimizing for this:

  • coherent song identity
  • less mirrored arrangement geometry
  • fewer dead spaces
  • more musical freedom inside bounded families/packs
  • editing of already-open projects as a first-class workflow

We are not optimizing for:

  • max clip-count reduction
  • perfect geometric cleanliness in Arrangement
  • “one sample per role for the whole song”
  • random variation that breaks identity

The target is:

  • structured, but not robotic
  • coherent, but not sterile
  • freer sound choice, but still musically believable

4. Non-Negotiable Musical Rules

4.1 No mirrored 32-beat arrangement as the default

If the same source starts at the same relative place across repeated 32-beat windows on most tracks, that is failure.

The current anti-pattern is exactly this:

  • AUDIO KICK
  • AUDIO CLAP
  • AUDIO HAT
  • AUDIO PERC MAIN
  • AUDIO PERC ALT
  • AUDIO TOP LOOP
  • AUDIO SYNTH LOOP

all behaving like the same grid template with different names.

That must stop.

4.2 No silence islands created by consolidation

If a rhythmic or harmonic role originally had dense useful placements, you cannot “fix fragmentation” by deleting 80 percent of the actual musical continuity.

A cleaner Arrangement that sounds emptier is a regression.

4.3 Harmonic MIDI must exist in Arrangement

The user may say piano roll.

Interpret that as:

  • harmonic MIDI lane
  • note backbone
  • continuity support

Do not interpret it as:

  • add piano timbre everywhere
  • turn the production piano-forward

The benchmark needs harmonic MIDI in Arrangement to support continuity.

4.4 Freedom is allowed, randomness is not

Sound choice should become more flexible, but only within coherence constraints.

Allowed:

  • rotate between 2-3 compatible sources in a role across sections
  • section-specific substitutions inside the same family/pack neighborhood
  • softer/harder variants per section

Not allowed:

  • same sample dominating every section of a role
  • completely unrelated packs across adjacent sections
  • “freedom” that just turns into incoherence

5. Live Truth You Must Start From

Before any coding validation:

  1. restart Ableton
  2. restart OpenCode
  3. reconnect MCP

Then prove live truth with:

  • get_session_info()
  • get_tracks()
  • audit_project_coherence()
  • get_track_info(track_index=15)

You must show the actual live values.

Do not reuse stale values from older reports.

6. P0 Tasks

P0.1 Re-validate the fixes Codex just made

These are mandatory validation points:

A. Runtime editing

  • get_device_parameters(track_index=15, device_index=0) must stop failing on non-quantized parameters
  • validate_key_conflicts() must stop failing with _send_command_to_ableton undefined
  • create_arrangement_clip(...) and duplicate_clip_to_arrangement(...) must be re-tested after restart using the updated fallback behavior

B. Coherence audit truth

audit_project_coherence() must now detect:

  • empty harmonic MIDI support in Arrangement
  • sample dominance across tracks
  • poor continuity more honestly than before

If the project is still clearly repetitive and sparse, the audit must not score it as GOOD.

Required evidence:

  • exact raw JSON output
  • short interpretation against the live project

P0.2 Add first-class arrangement inspection for open projects

The MCP still lacks a clean public way to inspect Arrangement clips directly.

Implement one of these:

  • get_arrangement_clip_info(track_index, start_time, track_type="track")
  • or extend get_clip_info(...) with view="session" | "arrangement" and start_time

Minimum fields:

  • track name
  • clip name
  • start time
  • length
  • audio vs MIDI
  • loop state when available
  • source/sample path when available

This is needed so project editing stops depending on guesswork and screenshots.

P0.3 Add coherence-repair tools for open projects

Stop adding only primitives. Add bounded repair tools.

Minimum public tools to add:

  • repair_harmonic_gaps(track_index, start_time=None, end_time=None, strategy="midi_backbone")
  • extend_track_continuity(track_index, max_gap_beats=12.0, respect_sections=True)
  • reduce_repeated_clip_overuse(track_index, max_same_source_ratio=0.65, preserve_family=True)

Each tool must return:

  • actions taken
  • clips touched
  • tracks touched
  • before/after counts
  • whether it changed Session, Arrangement, or both

Do not make these black boxes.

P0.4 Fix the symmetry failure directly

You must address the actual musical regression visible in the benchmark:

  • too many tracks start clips at the same repeated 32-beat anchors
  • too many tracks repeat the same source without section-level variation
  • too many gaps are “clean” but musically empty

Required engineering changes:

A. Add an explicit symmetry metric

Inside project coherence auditing, add at least:

  • grid_locked_repetition_tracks
  • or mirrored_section_layout_tracks

It must identify tracks where:

  • one source dominates
  • clip boundaries repeat on near-identical grid offsets
  • variation is cosmetic rather than structural

B. Use section-aware freedom, not whole-song locking

For editing/refinement tools:

  • allow source rotation per section
  • but constrain rotation by family and pack neighborhood
  • do not let a single source own every drop/build by default

C. Preserve rhythmic backbone density

Backbone one-shot roles must not be “cleaned” into sparse anchor hits.

If a role is:

  • kick
  • snare
  • clap
  • hat

then continuity matters more than clip-count aesthetics.

P0.5 Put harmonic MIDI into Arrangement and use it to fill holes

This is a core acceptance item.

The open benchmark project must gain real harmonic MIDI support in Arrangement.

Requirements:

  • use the existing harmonic MIDI lane, do not create random piano-forward production
  • place note content across the song, not just one isolated clip
  • use it specifically to reduce dead harmonic windows
  • respect the key and identity already present in the audio material

Success criteria:

  • get_tracks() shows arrangement clips on the harmonic MIDI track
  • audit_project_coherence() shows reduced harmonic emptiness
  • the edit is confirmed by re-inspection

7. P1 Tasks

P1.1 Make sound-choice freedom measurable

Add a metric to the open-project coherence audit for source diversity that does not reward chaos.

Good examples:

  • source_rotation_by_role
  • dominant_source_ratio_by_track
  • section_source_diversity

The metric should distinguish between:

  • coherent variation
  • frozen repetition
  • incoherent scatter

Acceptance rule

A track using exactly one source in all major sections should usually be flagged unless that role is intentionally static.

P1.2 Make repair tools section-aware

Repair tools must understand section boundaries.

Bad behavior:

  • inserting content uniformly every N beats
  • extending clips blindly across build/drop transitions

Good behavior:

  • keep intros lighter
  • make drops fuller
  • keep breaks supported but not overloaded
  • avoid mirrored section edits unless justified

8. Code Review Checklist

When you code, explicitly review these areas:

Specific review questions:

  1. Is this change reducing silence, or just reducing clip count?
  2. Is this change increasing real variation, or only moving the same source around?
  3. Does this change preserve identity across sections?
  4. Does this change help the currently open .als, not just future generations?

If you cannot answer those clearly, the change is incomplete.

9. Required Validation

Minimum local:

python -m py_compile "C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\AbletonMCP_AI\AbletonMCP_AI\MCP_Server\server.py" "C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\abletonmcp_init.py" "C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\AbletonMCP_AI\abletonmcp_runtime.py"
python "C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\AbletonMCP_AI\AbletonMCP_AI\MCP_Server\tests\test_runtime_truth.py"
python "C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\AbletonMCP_AI\AbletonMCP_AI\MCP_Server\tests\test_selection_coherence.py"

Required live validation:

  • inspect current benchmark project
  • run audit_project_coherence()
  • perform at least one arrangement edit
  • perform at least one continuity repair
  • re-inspect the exact affected track
  • prove before/after on:
    • harmonic gap
    • repeated-source hotspot
    • arrangement clip presence on harmonic MIDI support

10. Deliverable

Write:

  • C:\ProgramData\Ableton\Live 12 Suite\Resources\MIDI Remote Scripts\docs\SPRINT_v0.1.39_VALIDATION_REPORT.md

Must include:

A. Runtime truth

  • current MCP connection proof
  • current benchmark truth
  • exact live track state for the harmonic MIDI lane

B. Bug validation

  • device parameter inspection on a non-quantized device
  • key conflict validation proof
  • arrangement clip creation/duplication re-test

C. Coherence findings

  • exact audit_project_coherence() output
  • explain whether the project is still symmetrical and sparse

D. Repair proof

  • exact tool call
  • exact output
  • exact re-inspection output

E. Reviewer conclusion

Only one of:

  • COMPLETED
  • PARTIAL
  • BLOCKED

Do not mark COMPLETED if:

  • harmonic MIDI is still absent from Arrangement
  • large silence islands remain
  • sample dominance remains extreme on multiple tracks
  • the arrangement is still visibly mirrored section-to-section

11. Final Rule

Do not optimize for prettier screenshots.

Optimize for:

  • less emptiness
  • less mirror symmetry
  • more coherent freedom of sound choice
  • stronger harmonic continuity
  • real editable truth in the open project
Reference in New Issue View Git Blame Copy Permalink