Cubase-MCP-GAJ3T.MusicGen.R.../prompts/report_16.md

18 KiB

Step 16 — COMPLETE: release 0.2.0, exact original E2 hybrid verified/published
Date: 2026-10-20

OUTCOME / STOP
The accepted Step14 and corrected Step15 implementation is released as 0.2.0.
The original single-note E2 request works directly, not just the chord variant.
Final persisted wheel/sdist, real source and installed stdio, independent saved
MIDI/XML checks and one new user publication are verified. No later prompt was
opened/executed, no commit/delegation, dependency addition, application restart,
live chat/config edit, DAW operation or playback occurred. This update is ready
for the user's MCP restart and fresh chat; no next implementation step is run.

OWNED CHANGES
- Sole package version source src\riff_groove_builder\__init__.py now 0.2.0.
  Existing pyproject dynamic metadata and MCP initialization derive from it.
  Existing configured venv metadata was refreshed OFFLINE:
    .venv\Scripts\python.exe -m pip install --no-index --no-deps
      --no-build-isolation --no-cache-dir -e .
  Package import and importlib.metadata both report 0.2.0.
  No runtime/development dependencies were installed/restored/changed.
- Current README, architecture and compatibility describe the actual release,
  exact tested prompt, bounded grammar, pitch/voicing/fallback policy, reserved
  slides and restart/fresh-chat instructions. Historical reports are unchanged.
- Tests now assert 0.2.0, exact README description, minimal original arguments,
  E2 compact/all-tilde equivalence, native metadata/string6-fret4, full transport
  payload equality and final archive/source/resource/README identity/privacy.
- Installed hybrid acceptance now uses a separate real no-SD3-template process.
  Evidence writes to step16 rather than overwriting historical evidence paths.
  No parser/composer/profile-tuning/native/compiler defect or change was needed.

EXACT FINAL RELEASE FILES
Repository root R:
D:\_Repos\Code\GAJ3T\MCPs\MusicGen\GAJ3T.MusicGen.Riff-Groove-Builder

Wheel, 96021 bytes:
R\dist\hybrid-0.2.0-20261020\gaj3t_musicgen_riff_groove_builder-0.2.0-py3-none-any.whl
SHA256 9e4ffcd9e1b997ccdafc7dcc906aa1a64c90c1930d69f27f22b20d661c1ae5a0

Sdist, 115302 bytes:
R\dist\hybrid-0.2.0-20261020\gaj3t_musicgen_riff_groove_builder-0.2.0.tar.gz
SHA256 7aa757c228bafa9dc5ccd54ac610c496cedabf5de06ade2fe6989e1886e7f5dc

Built offline with .venv\Scripts\python.exe -I -B -m build --no-isolation
--outdir dist\hybrid-0.2.0-20261020 . (wheel built from the sdist).
The final full suite selected this EXACT directory with RIFF_RELEASE_DIST;
it did not substitute another disposable build for release acceptance.
Final wheel hash is also asserted unchanged across installed stdio tests.

ARCHIVE / INSTALLED IDENTITY AND PRIVACY
Wheel package .py/.json member set equals current src package exactly, and
each member is byte-identical. Six profile resources are included, no extras.
Sdist source/profile/README/pyproject bytes match current files exactly.
Wheel METADATA reports 0.2.0 and contains the current README (only standard
Windows CRLF versus LF normalization for its metadata text comparison).
Wheel-only metadata members are METADATA, WHEEL, RECORD and top_level.txt.
Sdist contains only package sources/profiles, README, pyproject, setup.cfg,
PKG-INFO and setuptools egg-info. No private chat/template, factory music,
manual binary, output song, test artifact, venv or application config shipped.

The exact wheel was installed offline/noneditably with --no-index --no-deps.
An isolated fresh-venv import and a real installed-target stdio server ran with
cwd outside the checkout:
D:\_Repos\Code\GAJ3T\MCPs\MusicGen
-I -B, no PYTHONPATH/PYTHONHOME; dev-venv dependencies reused for server tests.
All 18 package/submodule/entry paths and all six resource paths were asserted
under the wheel install target, not src. Module and resource hashes match
current source. Metadata is noneditable and 0.2.0; real initialize returns
0.2.0, protocol 2025-11-25. This is not fresh dependency certification.
Disposable wheel environments and native probes were cleaned by test fixtures.

FINAL VALIDATION
All commands use R\.venv\Scripts\python.exe:
  -m pytest -q --tb=short, RIFF_RELEASE_DIST=final persisted directory:
    PASS 1786 tests, 66.21s, no failures/skips.
    Includes full original simple/native/CC/SD3/template/atomic/privacy suites,
    pitch/voicing/hybrid suites, source and exact installed-wheel stdio.
  -m ruff check .:
    PASS, all checks passed.
  -m ruff format --check .:
    PASS, 109 files already formatted.
  -m mypy --strict src tests:
    PASS, no issues in 57 source files.
  -m pip check:
    PASS, no broken requirements found.
  git diff --check:
    PASS.
  Focused hybrid/readme/package/source transport run:
    PASS 138 tests, 2.43s.
  Editor problems on changed Python files:
    no errors.
The editor test runner discovered no tests; pytest was used directly instead.

SOURCE AND EXACT INSTALLED ORIGINAL E2 ACCEPTANCE
Both real processes received only:
  description = the exact final description copied below
  basename = hybrid-e2
No explicit formats/profiles/parts/seed/CC/template were supplied. Default seed
is 0. Both returned complete with .mid/.griff, no fallback_for. Complete decoded
JSON TextContent equals structuredContent, including nine interpretation rows.
Both public input/output schemas validate. Each no-template transport had
22 stdout response lines, empty stderr and exit0.

Independent Mido/XML numeric read-back, SOURCE = exact wheel = user output:
Attack  MIDI pitch  MIDI960 start/end  Native480 start/end  Technique  Velocity
1       40 (E2)        0 / 480            0 / 240          PM         80
2       40 (E2)      480 / 960          240 / 480          PM         80
3       40 (E2)      960 / 1440         480 / 720          PM         64
4       40 (E2)     1440 / 1920         720 / 960          PM         80
5       40 (E2)     1920 / 2400         960 / 1200         PM         96
6       40 (E2)     2400 / 2880        1200 / 1440         PM         80
7       40 (E2)     2880 / 3200        1440 / 1600         PM         80
8       40 (E2)     3200 / 3520        1600 / 1760         PM         80
9       40 (E2)     3520 / 7680        1760 / 3840         PH         80

MIDI: SMF type1, PPQ960, two tracks, tempo120, 4/4, both EOT7680.
Exactly nine musical on/off pairs plus PM26 selector0..30 and PH31
selector3520..3550, selector velocity100, channel0; 22 scheduled note events.
No CC, program changes, effects or bar2 retrigger.
Native: Riff XML2.1, Measures2, Time-Signature4/4, KeyEm, PPQ480,
loopEnd3840. Nine actual Note elements, all string6/fret4, eight PM and one PH,
No Legato; no occupied-string overlap or CC data.
Interpretation lists all nine exact MIDI40 attacks, starts/durations in reduced
quarter-beat fractions, first eight palm_mute, last pinch_harmonic held=true.
Final start11/3, duration13/3, through end of bar2; no new bar2 attack.

SOURCE/installed compact E2, explicit all-tilde bar2 and continuation-clause
bar2 preserve the exact same bytes, not an approximate rerendered pattern.
All source/wheel musical-output hashes agree across the 15 successful public
cases per transport; six negative cases per transport return visible errors
without publication or leaked private suffix. No owned .riff-stage remained.

Retained exact original copies (relative to R):
  .test-artifacts\step16\source\hybrid-e2\hybrid-e2.mid           168 bytes
  .test-artifacts\step16\source\hybrid-e2\hybrid-e2_lead.griff   1701 bytes
  .test-artifacts\step16\installed-wheel\hybrid-e2\hybrid-e2.mid 168 bytes
  .test-artifacts\step16\installed-wheel\hybrid-e2\hybrid-e2_lead.griff
                                                               1701 bytes
All MIDI copies SHA256:
c1967e698a50b8c22a22e458c3e2dd02d04407d1ad1d61f8a462d790b0810e62
All griff copies SHA256:
e8195e74c60ac93b1a4c92967a0834a64b2a2ebd74f453d6d3640ea465c4104c
These equal accepted original 0.1.1 musical hashes exactly.

CORRECTED CANONICAL D3 / HONEST EXPLICIT D2
Same Counts/Events; canonical rows:
Chords: F#2 F#2 D3 D3 Bb2 Bb2 - - -
Notes: - - - - - - C4 E4 G4
Hold final G4, not E2. Nine attacks expand to exactly21 musical tones:
  starts0/480:      [42,46,49] PM,480 ticks each
  starts960/1440:   [50,54,57] PM,480 ticks each
  starts1920/2400:  [46,50,53] PM,480 ticks each
  start2880:       [60] PM,320 ticks
  start3200:       [64] PM,320 ticks
  start3520:       [67] PH,to7680; no bar2 retrigger
Both real transports with fallback=false return native+MIDI; independent
readers verify all21 pitches, legal string/fret/timing, velocities, selectors
and no CC/programs. Native PPQ480 is exactly half MIDI timing.
Canonical and compact hashes:
  mid   6fd027fe4f74fec953be4c31fecda8eb14dec9bae0d15f76ae0c3dc481fdd181
  griff 45084055f11997d4f046f58190a08856bdb5e9de7d647a62227213a9994b6f1f

Explicit D2 remains [38,42,45] at960/1440, not hidden D3:
  fallback=false: NATIVE_FORMAT_UNAVAILABLE/no_physical_string_assignment,
                  no published destination.
  fallback=true: midi_fallback, exactly one .mid, fallback_for=[lead],
                 visible native warning, all21 exact tones, no native file.
D2 exact-MIDI SHA256:
4ce995ca751590b32cd85c72c3a2d8dad7f478cdf34d3abb12616d121719c3b4
MIDI38 and42 require different frets on string6 simultaneously; the accepted
tuning/physical validator is unchanged. No dropped tones/hidden octave rescue.

PUBLIC VOICING AND NEGATIVE ACCEPTANCE
Real source AND exact wheel exercise:
  Em4OverB2 = Em4/B2 = [47,64,67,71]
  Em4/B = Em4OverB = [59,64,67,71], default B3 disclosed
  [C3 E3 G3] = [48,52,55]
  Em4(add4) = [64,67,69,71]
  F# = [42,46,49], default root octave disclosed
These retain their pitches even in D minor context. Also actual rest/tie
read-back and D2 rejection, ambiguous grid, mismatched lane, reserved slide,
hold/rest conflict and private-suffix failure paths are tested on both stdio
transports. Full unit coverage retains case/tab/width/compact invariance,
duplicate/mismatched lanes, incompatible prose, grid/denominator boundaries,
slash/explicit voicing validation and no approximate fallback weakening.

ONE ACTUAL USER PUBLICATION
NEW official-SDK stdio MCP process, version0.2.0, configured venv interpreter,
cwd outside checkout, output root D:\MusicGenOutput, no SD3 template.
Arguments: exact final description, seed default0, unique basename
em_e2_hybrid_20261020_104001_3876e257, allow_midi_fallback=false.
Observed successful tool result: complete, isError=false; schemas valid,
TextContent JSON=structuredContent; every interpretation/pitch/tick/technique
and native/MIDI read-back passed before a late verification-harness assertion.
Both actual user files were then independently reopened again, not regenerated:

D:\MusicGenOutput\em_e2_hybrid_20261020_104001_3876e257\em_e2_hybrid_20261020_104001_3876e257.mid
168 bytes
SHA256 c1967e698a50b8c22a22e458c3e2dd02d04407d1ad1d61f8a462d790b0810e62

D:\MusicGenOutput\em_e2_hybrid_20261020_104001_3876e257\em_e2_hybrid_20261020_104001_3876e257_lead.griff
1701 bytes
SHA256 e8195e74c60ac93b1a4c92967a0834a64b2a2ebd74f453d6d3640ea465c4104c

Exactly these two new files; no sidecars. Prior user directory
em_chug_triplet_ph_20261020_030601 and both original files remain byte-identical.
No successful output was deleted, overwritten or republished.

ISSUES ENCOUNTERED / HONEST EVIDENCE BOUNDARY
1. Initial new test lint line was formatted with existing Ruff.
2. First full run:1785 passed,1 failed because wheel METADATA used Windows
   CRLF but README.read_text normalizes LF. Corrected only the test's text
   comparison; byte-exact source/resource and sdist README checks remain.
   Exact persisted wheel was NOT replaced. Final full run:1786 passed.
3. User publication succeeded, then its harness incorrectly required warnings
   to be empty. The default profile has the legitimate VERIFIED_SUBSET_LIMITS
   warning, as independently retained source/installed responses show. Native
   success and all requested musical checks had already passed. The original
   user raw tool response was not persisted before that late assertion.
   Recovery did NOT fabricate that lost response or generate another copy.
   user-publication.json records observed pre-error assertions and independent
   actual-file reopening against retained source interpretation. Full raw
   source and installed TextContent/structuredContent remain available.
   The isolated harness assertion is not a product/native/fallback failure.

PERSISTENT EVIDENCE / PRESERVATION
R\.test-artifacts\step16:
  release-files.json; build.log; metadata-refresh.log
  pytest-full-final.log (1786 pass)
  source\evidence.json; source\hybrid-transport.json
  installed-wheel\evidence.json
  installed-wheel\distribution-acceptance.json
  user-publication.json; user-reopen.log; user-stdio-stderr.log (empty)
  preservation-before.json; final-identity.json
Source and installed evidence include full arguments/TextContent/structured
results, interpretation and actual numeric musical/native records.
Their published workspace paths are historical transport locations; durable
copied bytes are under the retained source/installed-wheel directories above.

All57 preexisting files in the prompt/report + previous dist + user-output
preservation manifest remain hash-identical. Historical reports14/15/15_fix,
old release artifacts, accepted dirty work and previous user output are kept.
New validation evidence is separate from old Step13/15 evidence.
No owned generation stage or running helper is left. Test fixtures cleaned
their disposable processes/installations; final release/output/evidence persist.

Final source/doc identities (SHA256):
  README.md
  b07fa6347e29b3f20caf1f1dc8a3738553c8eb587126df81d312ab5645e16e3a
  docs\architecture.md
  b475508c6aeaa3f9d9931c69c3a427b6d57d5c96758b12506450b169891b1285
  docs\compatibility.md
  eecb1a4e3321f97ac10029ed94f152c9a7a0e657aa8627987e665c66f1b7dcea
  pyproject.toml (unchanged dependency/build declarations)
  ed2c8de067a4e0ea6ad10b21d3dab846c355d244d1c85b21f63f36f84a082d25
  src\riff_groove_builder\__init__.py
  2549a2b71d86a16bb0433641891b716bbc2f3dd055db65d2ab0a2d757aad961d
  src\riff_groove_builder\hybrid.py
  5e2f7b62dae9bcb3696391604338b4d24576342c732485a0831737f6d4fd4827
  src\riff_groove_builder\pitch.py
  b443ef587c49c26a3a9b2e2c736d89f83d8c291c14b76a0d292a0efedd345f4f
All six profile path/hashes and all18 loaded module hashes are retained in
installed-wheel\distribution-acceptance.json and checked against current src.

LIMITS / PRECISE MEANING
Implemented Events: o=normal/open articulation, x=PM, ^=AME lead PH,
-=rest, ~=adjacent continuation without retrigger. Open is not fingering.
Events / and backslash slides are RESERVED, NOT IMPLEMENTED; actionable
origin/destination/timing codec errors are returned. Chord / and Over bass
are a separate implemented operation, not generated slides.
C4=MIDI60; a chord digit is the root octave, not a seventh/extension.
Omitted slash bass picks the nearest matching pitch strictly below all
upper tones and discloses it; explicit bass must already be lower.
Key/style never secretly retune, transpose, revoice or drop explicit tones.
Raw explicit voicings can be physically unplayable: exact MIDI fallback is
truthful, while fallback=false returns an error without fabricated native.
Counts use meter-denominator beats with exact rational grids, not visual
spacing or compound grouping. Triplet count A is not pitch A.
Prose is bounded/local deterministic, not arbitrary model comprehension:
4000-character limit, documented grids/clauses, strict plan/event budgets,
one selected pitched hybrid part, static authored private-safe errors.
No new CC/effects/program claims, audio synthesis or application certification.
Live LM Studio/model behavior, plugin native load/save/reload/playback and
Cubase routing/import acceptance remain UNRUN.

USER ACTION — AFTER THIS VERIFIED UPDATE
Restart the riff-groove-builder MCP in LM Studio NOW, THEN open a NEW chat to
refresh cached schemas/tool descriptions. The existing configured interpreter
R\.venv\Scripts\python.exe now resolves 0.2.0. No config edit, config reinstall
or dependency reinstall is required. No live restart was done on your behalf.

FINAL COPY-READY FRESH-CHAT PROMPT (identical to tested README prompt)
Call generate_music once. Set basename to "em_e2_hybrid_20261020_fresh_chat_01"
and allow_midi_fallback to false. Set description verbatim to the text between
BEGIN DESCRIPTION and END DESCRIPTION below. Omit all other arguments; do not
invent formats, profiles, control changes, programs, effects or an SD3 template.
Return the actual tool result and its output paths, not a claimed playback.
BEGIN DESCRIPTION
Lead guitar in E minor, 4/4, 120 BPM, 2 bars.
Bar 1:
Counts: 1 & 2 & 3 & [4 trip let]
Events: x x x x x x [x x ^]
Notes: E2 E2 E2 E2 E2 E2 E2 E2 E2
Hold the final E2 through the end of bar 2 without retriggering.
END DESCRIPTION

Choose a new unique basename for each later run; existing files never overwrite.
The markers/instructions are not part of the tool description. Minimal
description+basename alone was also verified through source and exact wheel.

ORCHESTRATOR INDEPENDENT ACCEPTANCE AFTER REPORT REVIEW
Read the full current report, reviewed native/pitch/hybrid parser and service
interpretation paths, and independently reran the complete final suite against
RIFF_RELEASE_DIST=dist\hybrid-0.2.0-20261020:
  1786 passed in66.48s, zero failures/skips.
Ruff source/test lint and format, strict mypy (57 source files) and pip check
independently passed. Source/imported metadata both report0.2.0.
Final wheel/sdist SHA256 independently match this report.

Independently reopened both ACTUAL user-facing files using Mido/ElementTree,
without the production parser/composer/native readers. Verified exact nineE2
onsets/offs,8PM+1PH, selector26/31 ordering, no second-bar attack or CC, native
PPQ480/Measures2/KeyEm, both original musical hashes and exact two-file coverage.
No incomplete .riff-stage directory remains under the user's output root.
Prior files were not overwritten, and LM Studio/Cubase were not manipulated.
STOP: verified release/report complete; no later prompt.