Cubase-MCP-GAJ3T.MusicGen.R.../prompts/step_12.md

152 lines
8.6 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Step 12: Exact beat-scheduled chugs, triplets, pinch harmonic and cross-bar hold
## Objective and reviewed prerequisite
Step 11 is reviewed/accepted: complete JSON TextContent now reaches text-only
clients, errors identify formats/profile recovery, and structured Em works.
Read report_11.md and current description/composition/schema/profile/compiler/
native writer APIs. Implement the user's actual musical request accurately,
not a generic loop or an error-message-only workaround.
Original user description (from the shared LM Studio screenshot):
Please produce a 4/4 riff in Em. With Chugging palm muting 8th notes from 1 to 3&;
and triplets on 4&A with the A as a pinch harmonic that carries over a whole note
in the 2nd measure.
No commits, delegation, later prompts, audio analysis/hosting or DAW changes.
Do not read/edit the user’s LM Studio conversation/configuration.
## Musical interpretation and exact acceptance
Interpret 4 & A as the three eighth-note-triplet subdivisions of beat 4;
"A" refers to the final count, not an A-pitch instruction. The last note sustains
through ALL of measure 2 without retrigger. Return these assumptions explicitly.
Tempo was unspecified: retain the documented 120 BPM default and disclose it.
Use a documented low-tonic chug register (E2/MIDI40 for the default target)
unless an actual supported pitch instruction says otherwise; do not invent a
melody or silently transpose the PH to an unrelated key.
At 960 PPQ, for the original request:
- Six palm-muted eighth notes at 0,480,960,1440,1920,2400, duration 480.
- First two beat-4 PM triplets at 2880 and 3200, duration 320 each.
- Final triplet PH starts 3520 and ends 7680, one continuous note.
- Musical score duration: 2 bars / 8 quarter beats / 7680 MIDI ticks.
- Exactly 9 musical note-ons: 8 PM + 1 PH; no musical onset at bar-2 tick 3840.
- MIDI keyswitch PM=D0/26; PH=G0/31; current zero-lead policy and positive
switch gates remain explicit generator policy.
- Native Riffer PPQ480: musical starts
0,240,480,720,960,1200,1440,1600,1760; PH end=3840;
PM and PH are the actual native note Articulation values.
Do not truncate the hold, retrigger at the bar, quantize triplets to eighths,
or count keyswitches as musical notes.
## Evidence-backed native PH mapping
- Actual AME ProductVersion 4.0.1; target factory folder:
C:\Users\g3\Documents\Ample Sound\Riff4\Guitar4\AME Presets
- A read-only aggregate found 17 genuine native PH notes, label "PH",
observed velocities 80..100 and pitches 49..73. These observations establish
encoding, not hard limits of the articulation. Copy no factory sequence.
- Official Main_Panel_Manual-AME.pdf at https://www.amplesound.net/en/
documents PH at G0 and range C1-C5; existing numeric octave interpretation
establishes C0=24, so G0=31. Verify these technical facts before declaring
support. No copyrighted manual/factory assets in source/tests or uploads.
- Extend only the evidenced guitar profile/native PH mapping. Do not invent
Ray/SD3 PH support or loosen all native mappings to arbitrary labels.
## Owned implementation
- schemas.py: the smallest reusable typed internal note/phrase timing plan needed
to represent mixed subdivisions and per-note articulation; reuse rational
beat models. Avoid a prompt-specific flag or untyped dicts.
- description.py: bounded natural-language beat-range/triplet/PH-hold parsing,
shorthand key recognition, and explicit assumptions/unsupported diagnostics.
- compose.py: generic application of the typed timing plan, low tonic selection,
per-note articulation/velocity and range/budget validation.
- profiles.py + the Eclipse profile JSON: strict evidenced PH support; update
native articulation types/guards and profile metadata/version facts coherently.
- errors.py/tool/schema descriptions and docs for the actual expanded grammar.
- Unit/integration/stdio/installed/native tests; generated original fixtures only.
- prompts/report_12.md.
Do not add a new public score tool, model-provider API, unrelated file format or
generic plugin framework. The original description should work directly.
## Parser and model requirements
1. Implement reusable parsed rational timing, not an exact whole-string match
returning a canned riff. Parse numeric beat positions/ranges and generate
the corresponding subdivisions; test a nearby range variant too.
2. Support the original wording/spelling/spacing, eighth/8th, palm mute/
palm-muted/palm muting, 4&A/4 & A, 2nd/second measure, and Em/full E minor.
Recognized beat-range from/to clauses must not trip the old blanket
changes/sequences guard. Unrelated tempo changes/unsupported instructions
still fail; do not globally remove that protection.
3. A first-measure phrase tied through measure 2 implies two bars when length
is absent. Validate explicit meter/length overrides against the actual plan.
Reject impossible overlaps, out-of-range beat labels and too-short scores
rather than silently dropping material. Preserve explicit/default semantics.
4. New precise grammar must remain bounded/documented. Do not advertise arbitrary
melodies, song sections or every free-form instruction as supported.
New model-generated embellishment outside that grammar must remain actionable,
not be ignored into an approximate success.
5. Existing simple patterns/defaults produce unchanged musical output.
Every newly parsed timing/technique instruction is represented in the plan
and survives composition/compilation; no profile-based accidental fallback.
## Composer and native requirements
1. Use exact Fraction/rational arithmetic; triplet duration 1/3 and tied final
duration 13/3 quarter notes remain exact through MIDI960/native480.
2. Respect selected profile ranges/tuning and per-note articulation. PH must
switch using the actual profile data and render native PH, not softer PM.
3. Extend strict Riffer guards only for the supported AME PH mapping, leaving
Ray/standard/SD3 unsupported-technique errors honest.
4. A continuous PH note crosses the bar boundary; no duplicate/split retrigger.
Existing pairing/string assignment/shared duration safeguards still apply.
5. Enforce the same event budget, finite values, stable ordering and file
transaction safety. Preserve explicit CC requests; do not invent CC91/effects.
6. Update supported-text/tool/profile descriptions and old tests which explicitly
classified THIS now-supported phrase/PH as unsupported. Preserve rejection
tests for genuinely unsupported modes/instructions/invalid maps.
## Required unit tests
- Parse exact user wording plus spacing/wording/key aliases and a changed
beat-range variant; plan math independently asserted.
- Parsed per-note articulation, inferred 2 bars/default tempo/register/count-A
assumptions, invalid overlaps/meter/too-short length, unsupported prose.
- Compose exact timing/range/pitch/intensity plan, one cross-bar PH, no second-bar
attack, deterministic seed, budget and invalid-profile rejection.
- Verified PH keyswitch/native label; no bass/drum PH promotion.
- Existing ordinary pattern/default byte/semantic regressions remain intact.
## Required integration tests
- Real source stdio AND noneditable installed wheel receive the original
description verbatim with only safe basename/optional fallback=false.
- Text-only consumer sees full corrected arguments/results; automatic formats
are mid+griff for lead, no bogus drum project/profile/CC.
- Independently parse actual MIDI: 9 musical notes, 8 PM/1 PH, selector26 then31,
exact note-on/off times above, common EOT7680, E-minor low tonic and legal
velocities; no CC91 or unspecified effects.
- Independently parse actual .griff: PPQ480, two measures, correct PM/PH,
exact Begin/End/fret/pitch/velocities and final End3840.
- Native validator/independent comparison must match the same requested score;
actual saved files, not just file existence or serializer self-equality.
- Preserve simple native/MIDI/CC/atomic/source-template/package regressions.
Run focused tests, complete current suite, Ruff lint/format, strict mypy and pip.
Generate/retain only ORIGINAL acceptance evidence; no real user output creation
yet outside test artifacts (the orchestrator will make the final requested riff
after reviewing this report). No app/UI manipulation.
When complete, write your end-of-step report to ./prompts/report_12.md including: what was implemented, test results (pass/fail with details), any issues encountered, and whether you're ready for the next step
Entire report in exactly one plain triple-backtick fence. Include exact observed
MIDI/native timings/PH proof, unsupported limits, assumptions, before/after
grammar changes, exact test counts and next-step readiness.