Cubase-MCP-GAJ3T.MusicGen.R.../prompts/report_15.md

8.2 KiB

Step 15 — BLOCKED: canonical native requirement is physically incompatible
Date: 2026-10-20

OUTCOME / STEP16 READINESS
NOT READY for Step16. Stopped at a demonstrated blocker, as instructed.
No Step15 public hybrid parser is shipped. Provisional implementation and its
tests were removed after the canonical native acceptance gate failed.
Accepted Step14 source/tests/docs remain in place. This is not a completion
report, not an approval to advance, and not a claim that lane descriptions
work in the restored source or installed package.

BLOCKER
The mandatory canonical example requires two simultaneous D2 major attacks:
  MIDI pitches [38,42,45], at MIDI960 ticks960 and1440, each480 ticks.

The selected/default lead profile is ample.ame.4.0.1.riffer.v1.
Its accepted native tuning, strings6..1, is:
  MIDI [36,43,48,53,57,62], maximum22 frets on every string.

Exhaustive legal positions for the D2-major pitches:
  MIDI38 -> string6/fret2 ONLY.
  MIDI42 -> string6/fret6 ONLY.
  MIDI45 -> string6/fret9 OR string5/fret2.

MIDI38 and MIDI42 must sound together but require two different frets on the
same string at the same instant. No legal simultaneous assignment exists.
This is a physical contradiction, not a whitespace, octave-convention,
tokenizer, key-resolution, MIDI-grid or backtracking bug.

Evidence sources in the preserved code:
  src\riff_groove_builder\data\profiles\
    ample-metal-eclipse-4.0.1-riffer-v1.json:108-114
  src\riff_groove_builder\formats\agp.py:206-244
    _assign searches legal positions and rejects occupied-string overlaps.
  The independent native reader also checks string/fret/pitch and
    nonoverlapping notes on each string.

Both the provisional public-description path and, separately, the restored
Step14 ExplicitAttackPlan -> compose_score -> compile_score -> preflight_riffer
path reproduce no_physical_string_assignment with all21 pitches preserved.
No duplicate voicing engine or whole-example canned branch was used.

Changing the tuning/default profile, moving D2 to another octave, dropping
F#2, splitting a simultaneous chord into retriggers, emitting overlapping
impossible string notes, or accepting MIDI-only output would violate at least
one mandatory acceptance condition. None was used to claim native success.

ACTUAL PUBLIC SOURCE STDIO OBSERVATION DURING THE PROTOTYPE
The exact canonical description was sent to a real subprocess.
Text JSON and structuredContent were identical for both calls:

  allow_midi_fallback=false:
    isError=true
    code=NATIVE_FORMAT_UNAVAILABLE
    details={part:lead, format:griff, reason:no_physical_string_assignment}
    No destination/artifacts published.

  allow_midi_fallback=true:
    status=midi_fallback
    Exactly one .mid, explicit native-unavailability warning, no .griff.
    Independently parsed MIDI contains all21 required musical note-ons.
    Zero control_change messages; no bar2 musical retrigger.

The second result is evidence of the blocker, NOT a passing native result.
The prototype result included rational interpretation, exact resolved pitches,
technique and held status. That result surface was removed with the prototype;
it is not advertised as available in the restored product.

INDEPENDENT CANONICAL MIDI READ-BACK
MIDI960 start/end, pitches, technique:
     0 /  480   [42,46,49] PM
   480 /  960   [42,46,49] PM
   960 / 1440   [38,42,45] PM  <-- impossible native simultaneous fingering
  1440 / 1920   [38,42,45] PM  <-- same conflict
  1920 / 2400   [46,50,53] PM
  2400 / 2880   [46,50,53] PM
  2880 / 3200   [60]       PM
  3200 / 3520   [64]       PM
  3520 / 7680   [67]       PH, one continuous hold

Nine attacks expand to21 independent musical tones.
Selector-ons: (0,26), (3520,31); selector velocity100 and30-tick gate.
46 scheduled note events including musical tones and selectors.
Final G4 start=11/3, duration=13/3 quarter-note beats.
No retuning into E minor, omitted chord tone, hidden retrigger or controller.

After restoring Step14, reconstructed the exact same explicit plan through
the accepted APIs, repeated native preflight failure, and wrote MIDI bytes
identical to the provisional real-stdio fallback:
  SHA256:
  4ce995ca751590b32cd85c72c3a2d8dad7f478cdf34d3abb12616d121719c3b4

There is deliberately NO canonical native event read-back to present.
There is deliberately NO claim of installed-wheel hybrid parsing acceptance.

GRAMMAR / INTERPRETATION STATUS
The disposable parser used case-insensitive labeled rows, exact Fraction
counts and bracketed triplets; deterministic Step14 pitch-tokenization and
selected-profile resolution; o/x/^ attacks, rest/tie processing, explicit
slash/Over/bracket voicings and cross-bar holds. Spacing was lexical only.
Its76 unit cases passed, including canonical/compact nearby-pitch variants,
eighth/sixteenth/triplet counts, bare1&a ambiguity, rests, ties, voicing
conflicts, selected-profile limits, slide rejection and bounded style context.
The first real source canonical native gate then failed as detailed above.
Installed hybrid gates were not completed. The implementation was not kept
as an incomplete new public contract, and those provisional tests are absent
from the final passing suite.

Final supported grammar remains the accepted documented0.1.1 prose/template,
original beat-range/triplet-hold phrase, and internal Step14 pitch/plan APIs.
Counts/Events/Chords/Notes lanes, new Grid directives and a new interpretation
surface are NOT newly supported in this blocked result. No documentation or
tool guidance was changed to imply otherwise.

PRESERVED ORIGINAL E2 MUSIC — SOURCE AND INSTALLED WHEEL
The full accepted suite reran real source and offline built/installed-wheel
stdio tests. Both independently read nine MIDI40 musical notes and nine native
notes, matching the original phrase:
  onsets: 0,480,960,1440,1920,2400,2880,3200,3520
  ends:   480,960,1440,1920,2400,2880,3200,3520,7680
  velocities:80,80,64,80,96,80,80,80,80
  PM26 at0; PH31 at3520; no bar2 retrigger; native PM/PH at half MIDI times.

Source and installed-wheel original-byte hashes still match:
  .mid:
  c1967e698a50b8c22a22e458c3e2dd02d04407d1ad1d61f8a462d790b0810e62
  .griff:
  e8195e74c60ac93b1a4c92967a0834a64b2a2ebd74f453d6d3640ea465c4104c

FINAL VALIDATION AFTER REMOVING THE PROTOTYPE
All commands used .\.venv\Scripts\python.exe:
  -m pytest -q --tb=short:
    PASS,1661 tests,64.71s; full unit/integration/package suite.
    Includes offline sdist/wheel build, isolated wheel installation, real
    source/installed stdio, pitch/native/CC/privacy/atomic and old phrase tests.
  -m ruff check .:
    PASS, all checks passed.
  -m ruff format --check .:
    PASS,101 files already formatted.
  -m mypy --strict src tests:
    PASS, no issues in54 source files.
  -m pip check:
    PASS, no broken requirements.
  git diff --check:
    PASS.
  Earlier targeted preserved description/phrase/explicit-plan run:
    PASS,186 tests.

PRESERVATION / PERSISTENT EVIDENCE
Final git status matches the initial accepted dirty state, plus this report.
No version/dependency/release metadata, profile tuning, native/CC compiler,
public tools, application configuration or existing source behavior changed.
No commits, delegation, later prompts, user-app manipulations or new dependency
installation. Prototype files and project-local validation scratch removed.
The Step14 preservation manifest still has exactly82/85 unchanged baseline
files, with ONLY its already accepted changes:
  docs\architecture.md
  src\riff_groove_builder\compose.py
  src\riff_groove_builder\schemas.py

Retained reproducible numeric/error/transport evidence and actual MIDI:
  .test-artifacts\step15\blocker-source\evidence.json
  .test-artifacts\step15\blocker-source\
    canonical-midi-fallback\canonical-midi-fallback.mid
  .test-artifacts\step15\accepted-foundation\canonical.mid
  .test-artifacts\step15\preserved-original-evidence.json

REQUIRED RESOLUTION BEFORE STEP15 CAN BE ACCEPTED
An authorized requirement decision is needed: permit an explicitly chosen,
evidence-backed alternate tuning/profile that can voice both low tones;
change the canonical D2 voicing/register; or permit declared MIDI fallback
for this physically impossible native chord. None can be chosen silently.
Step16 remains NOT READY. Stopped here with the blocker demonstrated.