# Step 13: Final LM Studio recovery/precise-riff patch release acceptance ## Objective and accepted prerequisites Steps 11 and 12 are accepted after full report/source review and independent tests. Hidden TextContent feedback is fixed, Em is accepted, and the exact original beat-range/triplet/PH-hold description produces correct real MIDI and native guitar content. Finish final package/docs/regression acceptance. No commits, delegation, later prompts, user LM Studio data/config changes, DAW changes, audio/host features or unrelated cleanup. ## Owned work - Release version source/metadata/tests as directly necessary for patch 0.1.1. - Directly related documentation/examples/acceptance tests and actual defects. - `prompts/report_13.md`. Preserve previous reports as historical evidence; do not rewrite them as if earlier versions supported this phrase. Do not overwrite user output files. ## Required behavior and validation 1. Set the single authoritative package version to 0.1.1 so a restarted MCP reports the fixed build and distributed wheels cannot be mistaken for 0.1.0. Update directly coupled packaging/test expectations and current docs. Refresh only editable application metadata in the repository venv as needed; no unnecessary dependency restores/global installs. 2. Confirm exact JSON TextContent/structuredContent parity, readable profile IDs, correct lead/drum-format validation and Em/full-form normalization remain. 3. Confirm original text works VERBATIM, minimal arguments, no fabricated profile, drum format or CC91. Independent musical acceptance remains: 8 PM + 1 PH, MIDI onsets0,480,960,1440,1920,2400,2880,3200,3520, PHend7680, no bar2 retrigger; native times half and PH end3840. 4. Build final 0.1.1 wheel/sdist under a clearly identified directory such as dist/lmstudio-fix-20261020. Run the EXACT persisted wheel outside checkout, real stdio initialization/tool discovery and text-only consumer recovery, then original requested phrase generation and independent saved-file checks. 5. Run full current pytest, Ruff lint/format, strict mypy and pip check. Preserve native/MIDI/CC/template/atomic/privacy regressions and ordinary no-phrase musical-output hashes. Fix actual coupled failures, no weakened checks, blanket skips or success-shaped unsupported approximations. 6. Update README with LM Studio mcpServers (not VS Code servers) configuration, existing output-root requirement, exact current supported original example, minimal tool arguments and restart/fresh-chat remediation for cached schemas. Explain actual griff/briff/sd3p suffixes, defaults, precise grammar limits, count-A interpretation, low-tonic/default-tempo assumptions and no unsolicited profiles/effects. Validate executable JSON examples against actual schemas. 7. Do NOT manipulate/restart the live LM Studio application/MCP process or active chat, and do not touch the user’s Cubase session. Report that UI action is operator-required rather than claiming live UI/model success. 8. Tests may retain original acceptance files under ignored test artifacts. Do NOT create the user-facing requested files under D:\MusicGenOutput here; the orchestrator will issue the final request after reviewing this report. 9. Confirm source templates/local musical references remain unchanged and no private chat/template/factory music/manual asset enters source/packages. Clean only owned temporary build/test/process workspaces; retain clearly named final original evidence, reports and distributions. When complete, write your end-of-step report to ./prompts/report_13.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 patch version, exact final build paths/hashes, actual source/installed text-only recovery and musical-event evidence, final test/check counts, startup restart instructions, scope/interpretation limits and readiness for the final user-file generation.