livepilot 1.10.8 → 1.12.2

package/CHANGELOG.md

@@ -1,5 +1,378 @@
 # Changelog
 
+## 1.12.2 — Post-release audit reliability fixes (April 18 2026)
+
+Six issues surfaced by an immediate post-v1.12.0 deep audit (parallel
+code-reviewer subagents + manual verification). All fixed TDD-style —
+every bug now has a named regression guard in the test suite.
+
+**No tool count change** (still 398). **+11 regression tests**
+(2195 → 2206 passing, 1 skipped, 0 failed).
+
+### Critical fixes (reliability in hot paths)
+
+- **`send_capture` no longer blocks `send_command`** (BUG-audit-C1).
+  The M4L bridge shared `_cmd_lock` between `send_capture` and
+  `send_command`, so any concurrent MCP tool invocation during a
+  recording was blocked for the full capture duration (up to 35s).
+  The two operations use independent receiver state
+  (`_capture_future` vs `_response_callback`) and now use
+  independent locks (`_capture_lock` + `_cmd_lock`).
+  [`mcp_server/m4l_bridge.py:780-790, 912-913`](mcp_server/m4l_bridge.py).
+- **`_parse_osc` no longer crashes on malformed packets**
+  (BUG-audit-C2). `data.index(b'\x00')` raised `ValueError` when a
+  packet had no null byte — on UDP port 9880 collision with
+  non-OSC traffic, every incoming packet logged a noisy stack
+  trace. Replaced with `data.find(...)` + bounds checks on every
+  offset; malformed packets drop silently.
+  [`mcp_server/m4l_bridge.py:513-565`](mcp_server/m4l_bridge.py).
+- **`classify_simpler_slices` returns structured error on bad WAV**
+  (BUG-audit-C3). `sf.read()` was unguarded — corrupt or missing
+  files raised `soundfile.LibsndfileError` through the MCP
+  framework as an internal server error. Every other tool in the
+  module returns `{"error": ...}` dicts. Now wrapped in
+  `try/except` for consistent error shape.
+  [`mcp_server/tools/analyzer.py:571-581`](mcp_server/tools/analyzer.py).
+
+### High-severity fixes (API consistency)
+
+- **`batch_set_parameters` rejects negative `parameter_index`**
+  (BUG-audit-H3). `set_device_parameter` validates this at the MCP
+  layer; `batch_set_parameters` didn't, leaking an unstructured
+  `IndexError` from the Remote Script. Now rejected with a clear
+  ValueError at normalisation time.
+  [`mcp_server/tools/devices.py:318-328`](mcp_server/tools/devices.py).
+
+### Medium-severity fixes
+
+- **`_enrich_slice_response` uses positional fallback for missing
+  `index` field** (BUG-audit-M2). Direct `s["index"]` access in a
+  list comprehension raised `KeyError` on bridge version skew.
+  Now uses `s.get("index", i)` with `enumerate` fallback.
+  [`mcp_server/tools/analyzer.py:57-62`](mcp_server/tools/analyzer.py).
+- **`test_identify_returns_none_for_free_port` is no longer flaky**
+  (BUG-audit-M4). The test hardcoded port 59999 as "almost
+  certainly free"; when another process on the machine held it
+  (hitting Claude Desktop during the audit), the test failed
+  without diagnosing a real code bug. Now uses
+  `socket.bind(("127.0.0.1", 0))` to get a kernel-assigned free
+  port, releases it, then verifies.
+  [`tests/test_startup_safety.py:50-70`](tests/test_startup_safety.py).
+
+### .amxd binary patched in place
+
+- `m4l_device/LivePilot_Analyzer.amxd` had the `ping` response
+  version string patched from `"1.12.0"` → `"1.12.2"` via direct
+  byte replacement (same-length delta, size preserved). No Max
+  re-export needed.
+- `m4l_device/livepilot_bridge.js` source version also updated for
+  the next full rebuild.
+
+---
+
+## 1.12.1 — Silent-failure fixes + slice classifier (April 18 2026)
+
+Reconciles the "separate git stash" called out under v1.12.0's Known
+limitations — the 2026-04-18 Villalobos-groove session surfaced four
+silent-failure bugs and the need for a drum-slice spectral classifier.
+
+**+1 tool (397 → 398): `classify_simpler_slices`.** +43 regression
+guards (pure-Python, run without a live Ableton).
+
+### Ship-stoppers fixed
+
+- **`get_master_rms.pitch.midi_note` clamped** (BUG-F1) — polyphonic
+  pitch detector emitted values up to 319.15 with amplitude 0. Now
+  drops readings with zero amplitude or out-of-range MIDI.
+  [`mcp_server/tools/analyzer.py:128-147`](mcp_server/tools/analyzer.py).
+- **`get_simpler_slices` discloses base MIDI pitch** (BUG-F2) — Simpler
+  Slice mode uses C1 (MIDI 36) as slice 0, NOT C3. Response now
+  includes `base_midi_pitch` at top level and `midi_pitch` per slice.
+  Prevents the class of silent-audio bugs where MIDI notes at pitch
+  60+ trigger nothing. Docstring updated to mandate using the
+  returned `midi_pitch`.
+  [`mcp_server/tools/analyzer.py:37-62, 462-487`](mcp_server/tools/analyzer.py).
+- **`delete_track` last-track error message** (BUG-F3) — Ableton's
+  default rejection message was unrelated to the real cause ("you
+  can't add notes to a clip that doesn't exist yet"). Now pre-checks
+  `track_count` and raises a clear ValueError.
+  [`mcp_server/tools/tracks.py:93-112`](mcp_server/tools/tracks.py).
+- **`batch_set_parameters` accepts aligned schema** (BUG-F4) — supports
+  `parameter_index` / `parameter_name` (matching `set_device_parameter`)
+  in addition to the legacy `name_or_index`. Rejects ambiguous entries
+  with clear errors.
+  [`mcp_server/tools/devices.py:292-328`](mcp_server/tools/devices.py).
+
+### New tool
+
+- **`classify_simpler_slices(track, device, file_path?)`** — runs
+  FFT-based spectral analysis on a Simpler's slice boundaries, returns
+  each slice labeled as KICK / SNARE / HAT / ghost plus feature
+  breakdown (peak, rms, band %). Validated thresholds from the
+  2026-04-18 Villalobos-groove session on "Break Ghosts 90 bpm":
+    - KICK: sub+low ≥ 45%, high < 40%
+    - HAT: high ≥ 70% AND mid < 25%
+    - SNARE: mid ≥ 25% AND high ≥ 40% AND peak ≥ 0.6
+    - ghost: peak < 0.35
+  Eliminates the "assume slice 0 = kick" class of bug.
+
+### New module
+
+- **`mcp_server/sample_engine/slice_classifier.py`** — pure-Python
+  band-energy + peak classifier. Testable without Ableton
+  (`tests/test_slice_classifier.py` uses synthesized drum hits).
+  Exported `classify_segment()` and `classify_slices()` for direct
+  use outside MCP as well.
+
+### Documented bug entries
+
+- `BUGS.md` gains a new **"F. 2026-04-18 Villalobos-groove creative
+  session"** section: F1-F4 fixed here, F5-F7 scoped to v1.13+, F8
+  wontfix (workaround documented).
+
+### Not included in this release
+
+- `package.json` / `server.json` / plugin manifests / skill tool-count
+  sync — release-discipline is a separate task per CLAUDE.md's
+  "Version Bump" section. Run `python3 scripts/sync_metadata.py --fix`
+  before the next release.
+- Registering `classify_simpler_slices` in
+  `tests/test_tools_contract.py::test_analyzer_tools_registered` — the
+  total-count assertion there already passes if regenerated; the
+  registration list is additive and can land with the metadata sync.
+
+---
+
+## 1.12.0 — Live 12 LOM completeness (April 18 2026)
+
+Thirteen chunks closing the gap between LivePilot and Ableton Live 12.3.6's
+Live Object Model. **325 → 397 tools (+72). 45 → 51 domains (+6).**  Every
+addition is hasattr-probed where the underlying API varies by Live version
+— Core (12.0+), Enhanced (12.1.10+), and Full Intelligence (12.3+) tiers
+keep their graceful degradation contract.
+
+### New tools by chunk
+
+- **Chunk 1 — Song scale awareness (4 tools):** `get_song_scale`,
+  `set_song_scale`, `set_song_scale_mode`, `list_available_scales` — exposes
+  Live 12.0's Scale Mode at the song level. Remote Script probes
+  `song.scale_name` / `song.root_note`; missing on pre-12 falls back to
+  "Scale awareness unavailable".
+- **Chunk 2 — Per-clip scale override (3 tools):** `get_clip_scale`,
+  `set_clip_scale`, `set_clip_scale_mode` — Live 12.0 MIDI clip-level override
+  for the song scale. Honors `clip.scale_name` and `clip.scale_mode`.
+- **Chunk 3 — Tuning System (4 tools):** `get_tuning_system`,
+  `set_tuning_reference_pitch`, `set_tuning_note`, `reset_tuning_system` —
+  Live 12.1 microtonal tuning. Writes to `song.tuning_system`,
+  `tuning.reference_pitch`, and individual tuning-note cents.
+- **Chunk 4 — Follow Actions (8 tools):** Clip-level Live 12.0 revamp
+  (multi-action enum, `follow_action_time`, `follow_action_enabled`) +
+  scene-level Live 12.2+ (`scene.follow_action`, `scene.follow_action_time`)
+  + preset wrapper for "A→B→C chain" common shapes.
+- **Chunk 5 — Groove Pool (7 tools):** Pool enumeration, per-clip assignment,
+  master groove dial. Exposes Live 11+ `song.groove_pool` and the swing /
+  timing / random / velocity amount on each groove.
+- **Chunk 6 — Take Lanes (6 tools):** Enumeration, creation, per-lane clip
+  creation. Live 12.0 read surface + 12.2 write surface — both paths handled
+  with `hasattr` probes so older hosts degrade cleanly.
+- **Chunk 7 — Rack Variations + Macro CRUD (8 tools):** Variation
+  store/recall/delete + macro add/remove/randomize on Instrument/Audio-Effect
+  Racks (Live 11+).
+- **Chunk 8 — Sample Slice CRUD (6 tools):** `insert_slice`, `move_slice`,
+  `remove_slice`, `clear_slices`, `reset_slices`, `import_slices_from_onsets`.
+  Writes to `SimplerDevice.sample.slices` (Live 11+).
+- **Chunk 9 — Wavetable Modulation Matrix (5 tools):** Targets,
+  routing, amounts — completes the Wavetable surface alongside the existing
+  parameter tools (Live 11+).
+- **Chunk 10 — Song/Track long-tail primitives (12 tools):** `tap_tempo`,
+  `nudge_tempo_down/up`, exclusive arm/solo, `capture_and_insert_scene`,
+  `count_in`, Ableton Link state, `jump_in_session_clip`, performance-impact
+  read, `appointed_device`.
+- **Chunk 11 — Device A/B Compare (3 tools):** State read, toggle, copy
+  direction. Uses Live 12.3+ `Device.is_ab_state_enabled` / `ab_state`
+  where available; all three tools hasattr-probe so they return a clear
+  "unsupported on this Live build" error on 12.2 and older.
+- **Chunk 12 — ControlSurface enumeration (2 tools):** `list_control_surfaces`,
+  `get_control_surface_info`. Always-available diagnostic for multi-surface
+  setups.
+- **Chunk 13 — MIDI Tool bridge (4 tools):** `install_miditool_device`,
+  `set_miditool_target`, `get_miditool_context`, `list_miditool_generators`.
+  Exposes Live 12 MIDI Tools (Generators + Transformations) backed by
+  LivePilot generators (euclidean_rhythm, humanize, tintinnabuli). Ships
+  with both `.amxd` files pre-built from Live's factory templates — install
+  via `install_miditool_device()` which copies to the correct User Library
+  subfolders. **Note**: end-to-end Max-side integration is a known
+  follow-up; Max's `[js]` object may not locate `miditool_bridge.js` on
+  every machine without the folder being added to Max's File Preferences →
+  File Search Path. Server-side tools and config dispatch work standalone;
+  full round-trip notes-in-clip requires that Max path setup. Hence:
+  server shipped, Max-side user-setup step documented.
+
+### New domains (45 → 51)
+
+Source of truth is module layout — six new files registered @mcp.tool()
+decorators:
+
+- `mcp_server/tools/scales.py` — serves both scales (Chunks 1–2) AND tuning
+  (Chunk 3) since both live on the Song object.
+- `mcp_server/tools/follow_actions.py` — Chunk 4.
+- `mcp_server/tools/grooves.py` — Chunk 5.
+- `mcp_server/tools/take_lanes.py` — Chunk 6.
+- `mcp_server/tools/diagnostics.py` — Chunk 12.
+- `mcp_server/tools/miditool.py` — Chunk 13.
+
+Chunks 7–11 extended existing domains (devices, clips) and did not introduce
+new modules.
+
+### Known limitations
+
+- **MIDI Tool bridge (Chunk 13) — Max-side file search**: the `.amxd` files
+  reference `js miditool_bridge.js` relatively. Max normally searches the
+  .amxd's folder first, but Live's MIDI Tool instantiation context can
+  bypass that. If `Max → Window → Max Console` shows "can't find file
+  miditool_bridge.js" when you fire the tool: open Max → Options → File
+  Preferences → add `~/Music/Ableton/User Library/MIDI Tools/Max Generators`
+  and `~/Music/Ableton/User Library/MIDI Tools/Max Transformations` to the
+  File Search Path, save, reload the device.
+- A separate git stash ("pre-existing drift before LOM completeness work"
+  with `classify_simpler_slices` in flight) was left intact; the user will
+  reconcile it in its own release.
+
+### Not a breaking change
+
+Every new tool is additive. No existing tool names, parameters, or return
+shapes changed. Remote Script still boots cleanly on Live 12.0 — the 12.1+
+and 12.3+ tools just return `STATE_ERROR` with a clear message when the host
+lacks the underlying LOM attribute.
+
+## 1.10.9 — Second-pass audit + deferred-bugs shipped (April 18 2026)
+
+Completes every non-feature item on the v1.10.8 audit backlog. 2116 → 2132
+passing tests (+16 regression guards). 324 → 325 tools (`check_clip_key_consistency`
+lands from BUG-D1). Every deferred BUG-C and BUG-D entry either ships or is
+scoped to a follow-up feature; BUG-C4 is filed upstream as
+[PrefectHQ/fastmcp#3967](https://github.com/PrefectHQ/fastmcp/issues/3967).
+
+### Ship-stoppers fixed
+
+- **`send_capture` phantom 35s hang when receiver is None** —
+  [`mcp_server/m4l_bridge.py:441`](mcp_server/m4l_bridge.py). `send_command`
+  had a receiver-None guard; `send_capture` didn't. When UDP 9880 failed
+  to bind but the cache still reported connected, the OSC packet was
+  sent, the capture future was never registered, and the full 35s
+  timeout fired with a misleading "device may be busy" error. Added the
+  matching 5-line guard so the real cause surfaces immediately.
+- **`utils.py` stale after Control Surface toggle** —
+  [`remote_script/LivePilot/__init__.py:43`](remote_script/LivePilot/__init__.py).
+  Every handler does `from .utils import get_track, get_device`.
+  `importlib.reload(devices)` rebinds those names, but because `utils`
+  was not in `_HANDLER_MODULES`, the re-import resolved against the
+  stale `sys.modules["LivePilot.utils"]`. Added `utils` first in the
+  reload order so edits to the shared helpers pick up on toggle too,
+  honoring the dev-workflow guarantee documented at the top of the file.
+
+### New tools
+
+- **`check_clip_key_consistency`** (BUG-D1) — parses Splice-style
+  filename key tokens (`_D#min`, `_Ebmaj`, `_Cm`, …), cross-checks
+  against `get_detected_key`, returns the exact `set_clip_pitch(coarse=±N)`
+  call that would realign on mismatch. Handles `#`/`b` accidentals,
+  `min`/`m`/`maj` suffixes, absent tokens gracefully.
+- **`get_session_diagnostics(check_clip_keys=True)`** — opt-in scan
+  across every audio clip, appending `clip_key_mismatch` warnings with
+  the one-call fix attached.
+
+### Features shipped
+
+- **Session continuity persistence wired at startup** (BUG from external
+  audit). `bind_project_store_from_session()` now computes a project
+  fingerprint, opens the matching `ProjectStore`, AND rehydrates
+  `_threads` + `_turns` from disk. Wired into `server.py` lifespan with
+  a lazy rebind on first `record_turn_resolution` / `open_creative_thread`.
+  Creative threads and turn history now survive server restarts — the
+  README's "return to a project" claim is now load-bearing.
+- **Taste-aware `propose_next_best_move`** — replaces keyword-only
+  matching with `0.55 × keyword + 0.30 × taste_alignment + 0.15 ×
+  (1 − avoidance) ± 0.10 × family_bonus`. Cold-start users identical to
+  before; users with history get personalized ranking via
+  `dimension_weights` and `dimension_avoidances`. New return fields:
+  `score_breakdown`, `taste_active`, `taste_evidence_count`.
+- **Evaluation `taste_fit`** — previously hardcoded to `0.0`. Now
+  computed from `outcome_history` by matching same-direction deltas on
+  the same dimensions: ±0.2 per kept/undone match, neutral 0.5 baseline.
+- **`AutomationGraph.coverage_pct`** (BUG-D2, detection half) — new
+  fields `coverage_pct`, `clip_envelope_count`, `clips_scanned`
+  distinguish "no automation exists" from "we couldn't probe" from
+  "sparse but present". Surfaced in `get_project_brain_summary` as
+  `automation_coverage_pct`.
+
+### Refactor (BUG-C1)
+
+- **`mcp_server/tools/analyzer.py`** 1069 → 913 LOC. All 32
+  `@mcp.tool()` decorators stay in the module (FastMCP registration
+  order unchanged), helpers moved to `_analyzer_engine/`:
+  `context.py` (SpectralCache/bridge accessors, analyzer health check),
+  `sample.py` (Simpler post-load hygiene, filename heuristics),
+  `flucoma.py` (FluCoMa hint text, pitch-name table). Same
+  package-facade pattern as `_composition_engine` and `_agent_os_engine`.
+- **`sync_metadata::get_domains()` now skips `_*`** directories + files
+  under `mcp_server/tools/`. Matches Python private-package convention,
+  prevents internal helpers from registering as false domains.
+
+### Resilience (BUG-C3)
+
+- **`_get_all_tools()` probe chain extended** in `mcp_server/server.py`
+  to 4 paths: existing `_tool_manager._tools` and
+  `_local_provider._components`, plus speculative 3.3+ rename
+  `_local_provider._tools` and the future public `mcp.list_tools()`.
+  Each wrapped in try/except. All-empty fall-through now prints
+  `fastmcp.__version__` + the attempted probe labels — prior silent `[]`
+  return would disable schema coercion with no signal.
+- **New `_assert_tool_registry_accessible()`** self-test runs at module
+  import. Empty registry or a count mismatch against
+  `tests/test_tools_contract.py` fails loudly via stderr.
+
+### Upstream (BUG-C4)
+
+- **FastMCP feature request filed** —
+  [PrefectHQ/fastmcp#3967](https://github.com/PrefectHQ/fastmcp/issues/3967)
+  "Feature request: public tool-enumeration API". Migration is a
+  no-op once upstream lands a `mcp.list_tools()` or `mcp.tools`
+  surface — the probe chain already anticipates both.
+
+### Metadata + docs
+
+- **`sync_metadata.py` extended** — catches both `N tools` and hyphenated
+  `N-tool`, now covers `manifest.json` + `docs/manual/intelligence.md` +
+  `.claude-plugin/marketplace.json`. New prose-claim checks for
+  bridge-command count, enriched-device YAML count, and `GENRE_DEFAULTS`
+  key count — every narrative number now traces to a code derivation.
+- **README / CLAUDE.md / docs drift closed** — 28→30 bridge commands,
+  81→71 enriched devices, 7→4 genre defaults, 323→325 tool count in
+  stale marketplace/plugin/manifest descriptors. Intelligence manual
+  example signatures for `record_positive_preference`,
+  `record_anti_preference`, `evaluate_move` updated to match the
+  shipping APIs.
+- **Skill cleanups** — `livepilot-release` tool-count drift fixed;
+  `livepilot-wonder` reference paths corrected to point at their real
+  home in `livepilot-core`; arrangement vs composition-engine triggers
+  deduplicated (constructive vs analytical split).
+- **New test** `tests/test_claim_consistency.py` — 12 guards running
+  `sync_metadata --check` from pytest, verifying `manifest.json` and
+  `intelligence.md` stay in the sweep, and asserting that
+  `bind_project_store_from_session` keeps a non-test caller.
+
+### Quality-of-life
+
+- **`scripts/test.sh`** — blessed test entrypoint always uses `.venv/bin/python`,
+  closes the contributor trap where bare `pytest` failed 28 tests on
+  system Python.
+- **`.gitignore`** additions: `m4l_device/*.pre-presentation-backup`,
+  `m4l_device/*.pre-*-backup`, `.mcp.json.disabled`.
+- **Stale `livepilot-1.10.5.tgz`** removed from the repo root.
+
+---
+
 ## 1.10.8 — Deep audit fix pass (April 18 2026)
 
 Outcome of a cross-subsystem audit (Remote Script, MCP server, M4L bridge,

package/README.md

@@ -17,7 +17,7 @@
 
 <p align="center">
   An agentic production system for Ableton Live 12.<br>
-  324 tools. 45 domains. Device atlas. Splice integration. Auto-composition. Spectral perception. Technique memory.
+  398 tools. 51 domains. Device atlas. Splice integration. Auto-composition. Spectral perception. Technique memory.
 </p>
 
 <br>
@@ -38,7 +38,7 @@ Most MCP servers are tool collections — they execute commands. LivePilot is an
 | Layer | What it provides |
 |-------|-----------------|
 | **Deterministic Tools** | Direct control: transport, tracks, clips, notes, devices, scenes, mixing, arrangement, browser, automation |
-| **Device Atlas** | Knowledge of every device in Ableton's library — 1305 devices indexed by name, URI, category, tag, and genre. 81 enriched with sonic intelligence. 683 drum kits mapped |
+| **Device Atlas** | Knowledge of every device in Ableton's library — 1305 devices indexed by name, URI, category, tag, and genre. 71 enriched with sonic intelligence. 683 drum kits mapped |
 | **Sample Engine** | Three-source sample intelligence — searches Ableton's browser, your filesystem, and Splice's catalog simultaneously. 6 fitness critics score every result. 29 processing techniques |
 | **Spectral Perception** | Real-time ears via M4L — 8-band FFT, RMS/peak metering, Krumhansl-Schmuckler key detection, pitch tracking. Closes the feedback loop so the AI hears its own changes |
 | **Technique Memory** | Persistent library of production decisions. Save a beat pattern, device chain, or mix template. Recall by mood, genre, or texture across sessions |
@@ -58,7 +58,7 @@ Most MCP servers are tool collections — they execute commands. LivePilot is an
 │                                                                      │
 │  Device Atlas            8-band FFT              recall by mood,     │
 │  1305 devices            RMS / peak              genre, texture      │
-│  81 enriched             pitch tracking          29 techniques       │
+│  71 enriched             pitch tracking          29 techniques       │
 │  683 drum kits           key detection           replay into session │
 │                                                                      │
 │  Sample Engine           Corpus Intelligence     Taste Graph          │
@@ -79,8 +79,8 @@ Most MCP servers are tool collections — they execute commands. LivePilot is an
 │         └─────────────────┼──────────────────┘                       │
 │                           ▼                                          │
 │                  ┌─────────────────┐                                  │
-│                  │   324 MCP Tools  │                                  │
-│                  │   45 domains     │                                  │
+│                  │   398 MCP Tools  │                                  │
+│                  │   51 domains     │                                  │
 │                  └────────┬────────┘                                  │
 │                           │                                          │
 │           Remote Script ──┤── TCP 9878                                │
@@ -100,15 +100,15 @@ Most MCP servers are tool collections — they execute commands. LivePilot is an
 
 **MCP Server** (`mcp_server/`) — Python FastMCP server. Validates inputs, routes commands to the Remote Script over TCP, manages the M4L bridge, runs the atlas, sample engine, composer, and all intelligence engines. This is what your AI client connects to.
 
-**M4L Bridge** (`m4l_device/`) — Optional Max for Live Audio Effect on the master track. Provides deep LOM access through Max's LiveAPI that the ControlSurface API can't reach. UDP 9880 (M4L to server) carries spectral data and LiveAPI responses. OSC 9881 (server to M4L) sends commands. 36 bridge tools (backed by 29 bridge commands) for hidden parameters, Simpler internals, warp markers, display values, and Simpler warp / Compressor sidechain writes that live on child objects Python can't reach.
+**M4L Bridge** (`m4l_device/`) — Optional Max for Live Audio Effect on the master track. Provides deep LOM access through Max's LiveAPI that the ControlSurface API can't reach. UDP 9880 (M4L to server) carries spectral data and LiveAPI responses. OSC 9881 (server to M4L) sends commands. The 32 spectral/analyzer tools strictly require the bridge; device and sample tools that call the bridge also have graceful fallbacks, so core functionality works without it. Backed by 30 bridge commands for hidden parameters, Simpler internals, warp markers, display values, and Simpler warp / Compressor sidechain writes that live on child objects Python can't reach.
 
-**Device Atlas** (`mcp_server/atlas/`) — In-memory indexed JSON database. 1305 devices with browser URIs, 81 enriched with YAML sonic intelligence profiles (mood, genre, texture, recommended chains). 6 indexes: by_id, by_name, by_uri, by_category, by_tag, by_genre. The AI never hallucinates a device name or preset — it always resolves against the atlas first.
+**Device Atlas** (`mcp_server/atlas/`) — In-memory indexed JSON database. 1305 devices with browser URIs, 71 enriched with YAML sonic intelligence profiles (mood, genre, texture, recommended chains). 6 indexes: by_id, by_name, by_uri, by_category, by_tag, by_genre. The AI never hallucinates a device name or preset — it always resolves against the atlas first.
 
 **Sample Engine** (`mcp_server/sample_engine/`) — Searches three sources simultaneously: BrowserSource (Ableton's library), SpliceSource (local Splice catalog via SQLite), FilesystemSource (user directories). Every result passes through a 6-critic fitness battery (key, tempo, spectral, genre, mood, technical). 29 processing techniques (Surgeon precision vs. Alchemist experimentation). Builds complete sample processing plans with warp, slice, and effect recommendations.
 
 **Splice Client** (`mcp_server/splice_client/`) — Searches Splice's catalog through two layers: the local SQLite database (`sounds.db`, already-downloaded samples) and the live gRPC API (full catalog, including samples you haven't downloaded yet). The gRPC client auto-detects Splice's dynamic port via `port.conf`, handles self-signed TLS, and enforces a 5-credit safety floor before any download. Per-call timeouts (5–10s) prevent a hung Splice process from stalling the MCP event loop. Graceful fallback to SQL-only if grpcio isn't installed. No API key needed — authentication comes from the running Splice desktop app.
 
-**Composer** (`mcp_server/composer/`) — Prompt-to-plan pipeline. Parses natural language ("dark minimal techno 128bpm with industrial textures") into a CompositionIntent (genre, mood, tempo, key). Plans layers using role templates (kick, bass, percussion, texture, lead, pad, fx). Compiles to a step-by-step plan of tool calls that the agent executes. Does not execute autonomously — returns the plan. 7 genre defaults (techno, house, ambient, hip-hop, dnb, dub, experimental).
+**Composer** (`mcp_server/composer/`) — Prompt-to-plan pipeline. Parses natural language ("dark minimal techno 128bpm with industrial textures") into a CompositionIntent (genre, mood, tempo, key). Plans layers using role templates (kick, bass, percussion, texture, lead, pad, fx). Compiles to a step-by-step plan of tool calls that the agent executes. Does not execute autonomously — returns the plan. 4 genre defaults (house, techno, trap, ambient) — genres outside this set fall back to a neutral layer plan.
 
 **Corpus** (`mcp_server/corpus/`) — Parsed device-knowledge markdown converted to queryable Python structures: EmotionalRecipe, GenreChain, PhysicalModelRecipe, AutomationGesture. Feeds Wonder Mode, Sound Design critics, and the Composer with deep creative knowledge at runtime — not just LLM prompts, actual structured data.
 
@@ -120,7 +120,7 @@ Most MCP servers are tool collections — they execute commands. LivePilot is an
 
 ## The Intelligence Layer
 
-12 engines sit on top of the 324 tools. They give the AI musical judgment, not just musical execution.
+12 engines sit on top of the 398 tools. They give the AI musical judgment, not just musical execution.
 
 ### SongBrain — What the Song Is
 
@@ -156,7 +156,7 @@ When a session is stuck — repeated undos, overpolished loops, no structural pr
 
 ### Hook Hunter
 
-Identifies the most salient musical idea — ranks by rhythmic distinctiveness, melodic contour, repetition. Tracks whether hooks are developed, neglected, or undermined. Flags when a transition fails to deliver expected payoff.
+Identifies the most salient musical idea — ranks candidates by recurrence across scenes, motif salience, and section placement (payoff-section boost). Tracks whether hooks are developed, neglected, or undermined, and flags when a transition fails to deliver expected payoff. Rhythm-side ranking is currently heuristic (drum-track detection + clip reuse); true onset-based rhythmic features are on the roadmap.
 
 ### Session Continuity
 
@@ -172,7 +172,7 @@ Every engine follows: **measure before → act → measure after → compare**.
 
 ## Tools
 
-324 tools across 45 domains. Highlights below — [full catalog here](docs/manual/tool-catalog.md).
+398 tools across 51 domains. Highlights below — [full catalog here](docs/manual/tool-catalog.md).
 
 <br>
 
@@ -204,7 +204,7 @@ Every engine follows: **measure before → act → measure after → compare**.
 The M4L Analyzer sits on the master track. UDP 9880 carries spectral data to the server. OSC 9881 sends commands back.
 
 > [!TIP]
-> All 281 core tools work without the analyzer — it adds 36 bridge tools and closes the feedback loop.
+> Most tools work without the analyzer — it adds 32 spectral/analyzer tools (frequency, loudness, perception) and closes the feedback loop.
 
 ```
 SPECTRAL ─────── 8-band frequency decomposition (sub → air)
@@ -232,7 +232,7 @@ The atlas is an in-memory indexed database of Ableton's entire device library.
 
 ```
 1305 devices total
-  81 enriched with sonic intelligence (mood, genre, texture, chains)
+  71 enriched with sonic intelligence (mood, genre, texture, chains)
  683 drum kits mapped with note assignments
    6 indexes: by_id, by_name, by_uri, by_category, by_tag, by_genre
 ```
@@ -317,7 +317,7 @@ Prompt-to-plan auto-composition engine.
 └─────────────────┘
 ```
 
-- 7 genre defaults: techno, house, ambient, hip-hop, dnb, dub, experimental
+- 4 genre defaults: house, techno, trap, ambient (unknown genres fall back to a neutral plan)
 - Returns step-by-step plans — the agent executes each tool call in sequence
 - `compose` — plan a multi-layer composition from text prompt
 - `augment_with_samples` — plan sample-based layers for existing session
@@ -360,7 +360,7 @@ The V2 intelligence layer. These tools analyze, diagnose, plan, evaluate, and le
 | Creative Constraints | 5 | constraint activation, reference-inspired variants |
 | Preview Studio | 5 | variant creation, preview rendering, comparison, commit |
 
-> **[View all 324 tools →](docs/manual/tool-catalog.md)**
+> **[View all 398 tools →](docs/manual/tool-catalog.md)**
 
 <br>
 
@@ -587,7 +587,7 @@ See [CONTRIBUTING.md](CONTRIBUTING.md) for architecture details, code guidelines
 
 | Document | What's inside |
 |----------|---------------|
-| [Manual](docs/manual/index.md) | Complete reference: architecture, all 324 tools, workflows |
+| [Manual](docs/manual/index.md) | Complete reference: architecture, all 398 tools, workflows |
 | [Intelligence Layer](docs/manual/intelligence.md) | How the 12 engines connect — conductor, moves, preview, evaluation |
 | [Device Atlas](docs/manual/device-atlas.md) | 1305 devices indexed — search, suggest, chain building |
 | [Samples & Slicing](docs/manual/samples.md) | 3-source search, fitness critics, slice workflows |

package/m4l_device/livepilot_bridge.js

@@ -95,7 +95,7 @@ function anything() {
 function dispatch(cmd, args) {
     switch(cmd) {
         case "ping":
-            send_response({"ok": true, "version": "1.10.8"});
+            send_response({"ok": true, "version": "1.12.2"});
             break;
         case "get_params":
             cmd_get_params(args);

package/mcp_server/__init__.py

@@ -1,2 +1,2 @@
 """LivePilot MCP Server — bridges MCP protocol to Ableton Live."""
-__version__ = "1.10.8"
+__version__ = "1.12.2"

package/mcp_server/evaluation/fabric.py

@@ -31,6 +31,66 @@ from .policy import apply_hard_rules
 # ── Sonic Evaluator ──────────────────────────────────────────────────
 
 
+def _compute_taste_fit(
+    dimension_changes: dict[str, dict],
+    outcome_history: Optional[list[dict]],
+) -> float:
+    """Score how well this move aligns with the user's recent taste.
+
+    Shipped in v1.10.9 — previously hardcoded to 0.0.
+
+    For each dimension that moved (in ``dimension_changes``), look at the
+    user's last few kept/undone outcomes for the same dimension:
+      * kept with the same direction of delta → +0.2 per match
+      * undone with the same direction of delta → −0.2 per match
+
+    Returns a value in 0..1 (0.5 = neutral, neither signal). Empty history
+    returns 0.5, which is the correct "no information yet" neutral the
+    composite score already expects.
+
+    ``outcome_history`` entries are dicts of the shape::
+
+        {"dimension": "punch", "delta": 0.12, "kept": True}
+
+    Callers that pass a richer shape should extract those three fields.
+    Malformed entries are skipped silently so a schema bump upstream can't
+    break the evaluator.
+    """
+    if not outcome_history or not dimension_changes:
+        return 0.5
+
+    # Only weigh the most recent slice — taste drifts, and older signals
+    # shouldn't veto a current evaluation.
+    recent = outcome_history[-10:]
+
+    adjustment = 0.0
+    matched = 0
+    for dim, change in dimension_changes.items():
+        current_delta = change.get("delta", 0.0)
+        current_sign = 1 if current_delta > 0 else (-1 if current_delta < 0 else 0)
+        if current_sign == 0:
+            continue
+        for entry in recent:
+            if not isinstance(entry, dict):
+                continue
+            if entry.get("dimension") != dim:
+                continue
+            past_delta = entry.get("delta")
+            if not isinstance(past_delta, (int, float)):
+                continue
+            past_sign = 1 if past_delta > 0 else (-1 if past_delta < 0 else 0)
+            if past_sign != current_sign:
+                continue
+            kept = bool(entry.get("kept", False))
+            adjustment += 0.2 if kept else -0.2
+            matched += 1
+
+    if matched == 0:
+        return 0.5
+    # Neutral baseline 0.5 + averaged adjustment, clamped to [0, 1].
+    return _clamp(0.5 + adjustment / matched)
+
+
 def evaluate_sonic_move(
     request: EvaluationRequest,
     outcome_history: Optional[list[dict]] = None,
@@ -109,12 +169,13 @@ def evaluate_sonic_move(
     measurable_component = _clamp(0.5 + measurable_delta)
     preservation = _clamp(1.0 - collateral_damage * 5)
     confidence = measurable_count / max(len(targets), 1)
+    taste_fit = _compute_taste_fit(dimension_changes, outcome_history)
 
     score = (
         0.30 * goal_fit
         + 0.25 * measurable_component
         + 0.15 * preservation
-        + 0.10 * 0.0   # taste_fit: placeholder, no history in fabric v1
+        + 0.10 * taste_fit
         + 0.10 * confidence
         + 0.10 * 1.0   # reversibility: 1.0 for undo-able moves
     )