``` Step 5 fix report — explicitly labeled Standard MIDI fallback Date: 2026-10-19 CURRENT EXECUTABLE OBJECTIVE: IMPLEMENTED AND VERIFIED. STEP 6 READINESS: YES, using Standard MIDI fallback only. VERIFIED VENDOR MAPS: NONE; Eclips/Ray/SD3 mapping gates remain open. NATIVE/APPLICATION ACCEPTANCE: UNVERIFIED; no native support claimed. SCOPE AND PERSISTENT CHANGES Implemented only the current objective at the top of prompts\step_5_fix.md, following the amended continuation decision in prompts\plan.md and reviewing report_5.md, existing contracts, profile/compiler code and tests. Historical vendor-only stop wording was not treated as a blocker for standard fallback. Created: - src\riff_groove_builder\data\profiles\standard-midi-lead-v1.json - src\riff_groove_builder\data\profiles\standard-midi-bass-v1.json - src\riff_groove_builder\data\profiles\general-midi-percussion-v1.json - tests\unit\test_standard_profiles.py - prompts\report_5_fix.md Updated: - src\riff_groove_builder\schemas.py - src\riff_groove_builder\profiles.py - src\riff_groove_builder\errors.py - src\riff_groove_builder\data\profiles\__init__.py - tests\unit\test_profiles.py - tests\unit\test_schemas.py - tests\integration\test_contracts.py - tests\integration\test_articulation_midi.py - README.md - docs\architecture.md - docs\articulation-profiles.md - docs\compatibility.md - docs\formats\agp.md No delegation, commits, later prompts, parser/composer, native codecs, service wiring, dependency-manifest changes or dependency restoration. Existing server, CLI, generic compiler, MIDI writer, synthetic fixtures and previous reports were not edited. The Step 3 server still lists no profiles and returns its unfinished generation error until Step 9; no service completion is claimed. PUBLIC CONTRACT AND PROVENANCE ProfileSummary.mapping_kind is precisely typed as instrument | standard_midi. Its backward-compatible default is instrument. Existing fixtures/models retain that default; serialized summaries expose the discriminator. JSON Schema and model validation agree on the new enum and standard normal-only/native-false conditions. Existing strict/extra-forbid/numeric validation is preserved. Provenance adds status=standard_midi, distinct from verified and test_only. Standard mapping kind and provenance must agree. Standard profiles require public standard identity/range-or-voice facts and separately referenced generator channel/action/state/timing/velocity policy. New standard_public and generator_policy source kinds cannot satisfy verified vendor provenance. Existing verified profiles still require complete official/lawful evidence. Test-only identities remain excluded from production, with no fixture promotion. Standard profiles are constrained to canonical normal, alias sustain, latched/action-free behavior, explicit Standard/General MIDI product labels, and no native capability. Unsupported techniques are explicit named errors. Error wording no longer calls an arbitrary selected standard profile verified; error codes and envelope shapes were not changed. THREE PRODUCTION PROFILES, NOT VERIFIED INSTRUMENT MAPS ID Standard/product version Mapping version Channel standard.lead.v1 MIDI 1.0 standard-fallback-1 0 standard.bass.v1 MIDI 1.0 standard-fallback-1 1 standard.gm-drums.v1 General MIDI Level 1 standard-fallback-1 9 Exactly these three JSON resources are declared and packaged. Each is the unique default for its part; explicit-ID selection remains available and defensive. Pitched profiles accept inclusive legal MIDI note numbers 0–127, not asserted Eclips/Ray playable ranges. The drum profile explicitly declares all 47 GM Level 1 sounds, note numbers 35–81. It is NOT the SD3 default map. Channels 0/1 are generator routing policy; channel 9 explicitly selects GM's human-numbered percussion channel 10. Intended instrument routing fields remain eclips/ray/sd3, disambiguated by mapping_kind, product names and public limitations. No installed vendor build or preset is asserted: target_build/mapping_preset explicitly identify standards/policy and the absence of verified vendor setup. All three profiles use normal/sustain only, no keyswitches, CC actions, pedal, program/bank changes or GM initialization. Sustain is a normal-technique alias, not CC64. Every undeclared requested technique fails rather than being simulated. Single-interval dynamics use round(1 + 126 * intensity), ties to even: 0 -> 1, .25 -> 32, .5 -> 64, .75 -> 96, 1 -> 127. These are positive MIDI velocities and generator policy, not vendor layers. Exact score starts/durations are retained; no control pre-roll/tails are added. Receiver configuration and drum choke behavior remain external. EVIDENCE AND SOURCE IDS — CHECKED 2026-10-19 MIDI1-MESSAGES: https://midi.org/summary-of-midi-1-0-messages Official MIDI Association table exposes seven-bit note/velocity fields and zero-based channels 0–15. Supports legal pitched-note numbers 0–127 and MIDI message identity, not vendor playable ranges or routing defaults. Used by standard.lead.v1 and standard.bass.v1 as standard_public provenance. GM1-OVERVIEW: https://midi.org/general-midi-level-1 Official MIDI Association overview establishes GM Level 1 and standardized sound names, not acoustic timbre. It alone is not a numerical percussion map. Used by standard.gm-drums.v1 for standard identity. GM1-PERCUSSION-TABLE: https://computermusicresource.com/GM.Percussion.KeyMap.html Accessible Computer Music Resource teaching table identifies Vanderbilt University's Blair School of Music, explicitly gives GM channel 10 and all 47 number-to-sound pairs 35–81. Independently checked core entries: 35 acoustic bass drum; 36 bass drum 1; 38 acoustic snare; 42 closed hi-hat; 44 pedal hi-hat; 46 open hi-hat; 49 crash cymbal 1; 51 ride cymbal 1; 81 open triangle. Tests independently transcribe/assert all 47 pairs. Octave labels were not used to derive numeric notes. Used by standard.gm-drums.v1 for percussion facts, never SD3 verification. GENERATOR-POLICY-V1: docs/articulation-profiles.md#standard-fallback-policy-v1 Separately documents routing, aliases, legal pitched acceptance, action-free normal/sustain, timing and intensity policy. All three profiles reference it. External facts are not misrepresented as evidence for generator-specific policy. CORRECTED AMPLE DOCUMENTATION The raw HTML of https://www.amplesound.net/en/manual.asp was checked in memory. It exposes these exact relative public links under https://www.amplesound.net/en/: - Main_Panel_Manual-AME.pdf - Main_Panel_Manual-AMR.pdf - Guitar_Riffer.pdf - Guitar_Strummer.pdf - Settings_and_CPCv2.pdf Compatibility, profile, architecture/README status and AGP evidence documentation were corrected. The earlier simplified extraction missed links; it did not prove that the manual page contained only a domain label. Discovering public manuals does not verify installed versions/maps, resolve octave-label interpretation or establish native layouts. No manual or factory MIDI song was copied into the repository, and no authenticated vendor access was attempted. ACTUAL VALIDATION COMMANDS AND RESULTS Environment: existing repository .venv, Windows AMD64, CPython 3.13.13. No other Python/OS/plugin configurations are certified by these runs. Commands below ran from the repository root. Repository-local scratch setup: $scratch = New-Item -ItemType Directory -Force '.test-artifacts\scratch' $env:TEMP = $scratch.FullName $env:TMP = $scratch.FullName $env:TMPDIR = $scratch.FullName $env:PIP_NO_INDEX = '1' Focused schema/profile/compiler/MIDI/wheel validation: & '.\.venv\Scripts\python.exe' -m pytest tests\unit\test_profiles.py tests\unit\test_standard_profiles.py tests\unit\test_schemas.py tests\integration\test_contracts.py tests\integration\test_articulation_midi.py -q -s --tb=short PASS: 453 passed in 4.24s; 0 failed, 0 skipped. Includes installed and ZIP-wheel profile evidence, independent standard MIDI events/read-back, defaults, all GM numbers, provenance and schema regressions. Final full current suite, after all code corrections: & '.\.venv\Scripts\python.exe' -m pytest -q --tb=short PASS: 700 passed in 31.00s; 0 failed, 0 skipped; no warnings reported. This is 48 additional cases over the previous 652-test suite. Existing generic switch/CC/state/timing, contract, errors, writer, distribution and real stdio skeleton regressions remain passing. Unsupported standard techniques/voices fail honestly; no synthetic/vendor maps appear in the production registry. & '.\.venv\Scripts\python.exe' -m ruff check . PASS: All checks passed! & '.\.venv\Scripts\python.exe' -m ruff format --check . PASS: 41 files already formatted. & '.\.venv\Scripts\python.exe' -m mypy PASS: Success: no issues found in 23 source files. & '.\.venv\Scripts\python.exe' -m pip check PASS: No broken requirements found. git --no-pager diff --check PASS: no tracked-diff whitespace errors. Existing implementation files are untracked in this worktree; Ruff checks cover the actual Python files too. Additional final wheel-only verification: & '.\.venv\Scripts\python.exe' -m pytest tests\integration\test_articulation_midi.py -k built_wheel_standard_registry -q -s --tb=short PASS: 1 passed, 8 deselected in 2.73s; 0 failed, 0 skipped. The test builds through an actual sdist/wheel using build --no-isolation, installs the package offline/no-deps into its isolated target, and probes both the installed target and wheel ZIP with Python -I -B from ROOT.parent, outside the checkout. Both actual module locations were printed and validated; both reported isolated=1 and exactly: standard.bass.v1, standard.gm-drums.v1, standard.lead.v1 Probes check the exact JSON manifest, standard-only metadata, no synthetic identity or verified vendor status, normal/sustain, default channels 0/1/9 and core drum numbers. Full-suite distribution tests additionally build both artifacts and import from a real isolated wheel venv outside the checkout. These are packaging/MIDI tests, not target-application acceptance. Editor problems checks reported no errors in all changed Python modules/tests. Editor runTests found no discovered tests for the supplied absolute paths; the actual repository-venv pytest runs above provide execution evidence instead. ISSUES RESOLVED AND CLEANUP - Initial strict typing identified an unannotated heterogeneous test-update collection and a test assuming channel lived on EventTrack. Added a precise list annotation and asserted actual per-event channels; no suppressions. - Ruff found one 102-column error-message line; wrapped it without changing its value. Final full suite and all static checks pass on the corrected code. - Two candidate CMU table URLs returned 404; the accessible Vanderbilt teaching table supplied the numerical evidence. Missing vendor maps were not treated as a standard-fallback implementation blocker. - Final owned articulation/distribution/MIDI/stdio workspace count: 0. Root build directory: absent. Unique wheel/test workspaces clean themselves. Existing venv, caches, dist products and repository scratch were preserved. No background services were started. REMAINING GATES AND FINAL DECISION Standard MIDI fallback registry, selection, compilation, real SMF read-back and wheel resources: VERIFIED and persistent. Verified Eclips/Ray/SD3 maps: STILL OPEN. Exact installed products/builds, libraries/presets/active maps, numeric playable ranges/controls, switching and reset timing, vendor layers and target behavior require separate evidence. Public manual links and GM facts do not verify those configurations. Native AGP/SDP: STILL OPEN. Musical-content semantics, actual supported layouts, lawful controlled exports/permissions and plugin/Cubase load/save/reload acceptance remain unverified. No native files were serialized. Actual application acceptance: NOT RUN, not counted as a passing test or skip. Future generation results must expose standard-only identity and warnings; the unchanged server remains development-only until Step 9 service wiring. Ready for Step 6: YES, STANDARD FALLBACK ONLY, as permitted by the amended plan. No later prompt or implementation was started. This is not completion of the vendor mapping, native, application, service or full-project gates. Stop here. ```