133 lines
7.8 KiB
Markdown
133 lines
7.8 KiB
Markdown
# Step 7 native fix: Implement AGP only from real format evidence
|
|
|
|
## Current executable implementation amendment
|
|
|
|
Read the latest native-format continuation at the top of `plan.md`. Real local
|
|
evidence HAS BEEN FOUND. Implement the actual native Riffer writers now, rather
|
|
than another research-only or unavailable-capability report.
|
|
|
|
The installed target versions are observed from VST3 ProductVersion metadata:
|
|
`C:\Program Files\Common Files\VST3\AME.vst3` = 4.0.1 and
|
|
`C:\Program Files\Common Files\VST3\AMR.vst3` = 4.1.0.
|
|
Read-only references:
|
|
|
|
- `C:\Users\g3\Documents\Ample Sound\Riff4\Guitar4\AME Presets\*.griff`
|
|
- `C:\Users\g3\Documents\Ample Sound\Riff4\Bass4\AMR Preset\*.briff`
|
|
- User-authored examples may also exist directly under Guitar4/Bass4.
|
|
|
|
Do NOT upload these files or copy factory musical sequences into the repository.
|
|
Only schema/numeric facts may become documentation and generated original tests.
|
|
The real suffixes are `.griff` for lead and `.briff` for bass. `agp` remains a
|
|
logical request alias; never write an incorrectly suffixed native artifact.
|
|
|
|
### Observed schema facts, to verify before implementation
|
|
|
|
- XML `Riff` root, `Ver="2.1"`, `PPQBase="480"`, numStr 6/5.
|
|
- Root fields Author/Category/Measures/Time-Signature/Quantize/Tempo/Type/Rating/
|
|
Key/loopMode/loopStart/loopEnd; guitar Inst="AME"; bass may omit Inst but
|
|
user-authored AMR example includes it.
|
|
- `String ID="6"` down to 1 (guitar), 5 down to 1 (bass).
|
|
- Child `Note`: NoteID, Fret, Begin, End, Velocity, Off-Velocity, Articulation,
|
|
Legato, Muted. MIDI pitch is NoteID; Begin/End are integer ticks at PPQBase.
|
|
- Observed labels: Sus/PM (both), guitar PH/Tap/Mute/SI, bass Acc/DN/PMP/STA.
|
|
Support only evidence-backed semantics; normal and palm mute are sufficient
|
|
initial target techniques. No guessed special legato or effects.
|
|
- Default guitar open-string notes by ID 6..1: 36,43,48,53,57,62; one factory
|
|
variant uses 38 on string 6, so tuning must be explicit rather than inferred
|
|
indiscriminately. Default bass 5..1: 35,40,45,50,55; observed drop variants
|
|
use 33/38 on lower strings, not a universal default.
|
|
- Root `ccName-1="1.Mod"` / `ccVisibility-1="1"` and `cc-1.Mod` child with `p`
|
|
points tick/value/curveType. Values are MIDI controller units (observed
|
|
0..120), not normalized 0..1. `curveType="stair"` encodes discrete CC actions.
|
|
Optional curvePointTickProp/curvePointValueProp exist for curved interpolation.
|
|
|
|
### Owned implementation
|
|
|
|
1. Replace the unavailable AGP boundary with real Riffer capability/writer/
|
|
saved-file validator APIs for supported version/target profiles. Keep legacy
|
|
check/require names compatible; native metadata alone cannot invent a codec.
|
|
2. Add strict native Riffer mapping/settings to InstrumentProfile as needed:
|
|
layout version, explicit tuning/fret/range policy, actual extension and
|
|
supported articulation/CC representations. No arbitrary template scripting.
|
|
3. Add evidence-backed Eclipse 4.0.1 and Ray5 4.1.0 profile JSON files and
|
|
provenance. Use public manuals plus safe local evidence IDs, not private
|
|
absolute paths in distributed metadata. Keep existing standard profiles
|
|
explicit fallbacks and synthetic profiles excluded.
|
|
4. Public/native artifact contracts must represent actual griff/briff/sd3p
|
|
suffixes, while retaining agp/sdp request aliases. Make the smallest coherent
|
|
schema additions and update relevant validation/tests. Do not relabel actual
|
|
extensions as agp/sdp in result metadata.
|
|
5. Keep new target profiles NONDEFAULT until the later integration fix wires
|
|
all native writers. Preserve operational default MIDI service meanwhile.
|
|
6. Map verified normal/palm_mute techniques and velocity behavior. Public
|
|
manuals establish guitar C0 normal/D0 palm mute; numeric C0=24 is consistent
|
|
with the vendor's public exported Eclipse MIDI and actual native note ranges.
|
|
Bass normal C0 has explicit velocity bands: 64..126 sustain, 32..63 palm mute;
|
|
avoid 127 accent and <32 dead-note when generating those named techniques.
|
|
Describe generator timing/routing policy separately from instrument facts.
|
|
Confirm facts against local/manual evidence; do not mark unsupported modes.
|
|
7. Add CC1 modulation/vibrato support only if its exact behavior is established;
|
|
the native renderer must at least preserve supported compiled CC1 events.
|
|
No hard-coded fabricated CC reset or vendor layer ranges.
|
|
8. Render generated ORIGINAL score notes with correct string/fret assignment,
|
|
no overlapping notes on a physical string, exact rational timing converted
|
|
to PPQ 480, on/off velocities and native articulation. Preserve or explicitly
|
|
reject unrepresentable pre-roll/tails rather than silently misalign MIDI.
|
|
9. Write exclusively into the supplied staged path, reject wrong suffix/version/
|
|
target/unsupported content, read saved XML back and independently compare
|
|
actual notes/CC/timing to the requested score/event plan. No fake success.
|
|
10. Update native/profile/model/contract tests and docs, including truthful
|
|
corrected suffixes. Add generated original minimal XML fixtures only.
|
|
Existing factory files are read-only local evidence, not redistributed tests.
|
|
|
|
Run focused new native/profile/schema tests, the full current suite, Ruff
|
|
lint/format, strict mypy and package/resource checks as affected. Fix actual
|
|
regressions; do not weaken the tests or claim all existing profiles native.
|
|
Do not wire service native output yet, implement SDP, commit, delegate or create
|
|
later prompts. Produce actual generated .griff/.briff read-back evidence.
|
|
|
|
Update `prompts/report_7_fix.md` with a current implementation section and retain
|
|
the earlier research as clearly labeled historical evidence inside the same
|
|
single plain triple-backtick report fence. Report implementation/tests/actual
|
|
native files/limitations and next-step readiness, not a research-only stop.
|
|
|
|
---
|
|
|
|
Historical evidence gate (superseded where the amendment supplies facts):
|
|
|
|
Status: native acceptance BLOCKED; capability software reviewed and verified.
|
|
Read `report_7.md` and `docs/formats/agp.md`. The current boundary intentionally
|
|
returns native unavailable and performs no filesystem IO.
|
|
|
|
## Objective and required inputs
|
|
|
|
Implement real Eclips/Ray musical-content AGP export after lawful specifications
|
|
or controlled real exports establish the layout, supported versions and content
|
|
semantics. Required: empty/one-note/multi-note files, isolated timing/tempo/
|
|
velocity/articulation changes, source/permission records, and exact native
|
|
load/save workflow. Do not substitute preset-only settings or renamed MIDI.
|
|
|
|
## Files and constraints
|
|
|
|
Modify `formats/agp.py`, only necessary shared capability/profile definitions,
|
|
AGP unit/integration tests, `docs/formats/agp.md` and compatibility documentation.
|
|
Add redistributable native fixtures only with permission; otherwise document
|
|
local-only tests using safe evidence IDs. Preserve MIDI fallback and strict
|
|
errors. Native availability requires an actual codec, not profile metadata.
|
|
No scope expansion, commits, delegation or later prompts.
|
|
|
|
## Tests and acceptance
|
|
|
|
Unit-test verified field serialization, version/invalid input handling,
|
|
length/checksum/compression rules, and separate guitar/bass behavior as required.
|
|
Integration-test actual fixture round trips and generated musical content;
|
|
record real Eclips/Ray load/save/reload acceptance separately. Run related
|
|
fallback/MIDI regressions, Ruff lint/format and strict types.
|
|
If evidence is unavailable, keep native acceptance blocked; do not invent it.
|
|
The orchestrator must review this report before claiming native completion.
|
|
|
|
When complete, write your end-of-step report to ./prompts/report_7_fix.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: exactly one plain triple-backtick fence. Include exact commands,
|
|
source/fixture versions, application acceptance, skips and unresolved blockers.
|