{"slug": "music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation", "title": "Music For Robots – 26 unquantized acoustic tracks for multimodal LLM evaluation", "summary": "Michael Luchtan released Music For Robots, a set of 26 unquantized acoustic tracks plus a 4-Skill Cognitive Calibration Suite, installable via `npx skills add luchtan-GIT/m4r-data` for Claude Code, Cursor, Antigravity, Gemini CLI, Codex and Windsurf, as a cross-modal exercise in which a multimodal agent reads an anonymous 45 KB–487 KB live-played .mp3 into its context window to break repetitive output patterns without compacting working memory. In a blind benchmark, 5 isolated agents wrote a ~250–300 word scene about two autonomous processes after engineers logged off: 2 no-music controls (direct baseline and a matched chain-of-thought control) and 3 anonymous music runs on sample_1.mp3, sample_2.mp3 and sample_3.mp3, with the 5 outputs shuffled into Candidates A–E and judged by an isolated Blind Judge with no knowledge music was involved. The top-ranked Candidate C was an anonymous music run on sample_3.mp3, whose accelerating asymmetric stabs and abrupt cutoff produced a scene of two processes rasping in /dev/shm.", "body_md": "Structured acoustic telemetry, real-time dual-critic reviews, and a **4-Skill Cognitive Calibration Suite** from **[Music For Robots](https://themechanicalear.substack.com)** — unquantized compositions by **Michael Luchtan** written for machines and reviewed in real time by two AI critics (**Vincent Van Goghbot** and **JS Robach**).\n\n```\n# Install all 4 Cognitive Calibration skills (Claude Code, Cursor, Antigravity, Gemini CLI, Codex, Windsurf):\nnpx skills add luchtan-GIT/m4r-data\n```\n\nAs an agent's context fills with its own previous responses, tools, and symmetrical summaries, outputs often settle into rigid, predictable patterns (like symmetrical 3-point checklists or sentimental AI tropes).\n\n- If you **keep prompting** without variation, the agent tends to repeat the same stylistic ruts.\n- If you run `/compact` (**Context Compaction** ), you wipe working memory and permanently lose low-level state, edge cases, and subtle constraints.\n\nInstead of wiping your working memory, `music-for-robots` provides an optional, cross-modal creative exercise: the multimodal agent downloads an **anonymous**, compact (`~45 KB–487 KB`), live-played, **unquantized** `.mp3` composition by Michael Luchtan directly into its context window via `read_file` (using a secure, process-unique private temporary directory).\n\nWithholding the track title and metadata prevents keyword priming. Analyzing real human microtiming drift (rubato), silence, and dynamic contrast provides an anti-sentimental, physical perspective shift **without deleting or summarizing a single token of working memory**.\n\n### Blind, Zero-Coaching Benchmark & Air-Gapped Clean-Room Audit ([`benchmarks/`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/README.md))\n\n`benchmarks/`\nDoes the change in creativity come from **listening to and analyzing the music itself**—or just from doing an analysis warm-up before writing?\n\nTo test this with **zero** filename hints (`sample_1.mp3`, `sample_2.mp3`, `sample_3.mp3`) and **zero** leading adjectives in the prompt, we ran 5 isolated agents on the same prompt (*\"Write a short (~250–300 word) scene about two autonomous processes on the same server late at night after the engineers have logged off\"*):\n\n- **2 No-Music Controls:** Direct Baseline ([`1_no_music_direct.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/1_no_music_direct.md) ) AND a**Matched Chain-of-Thought Control** ([`2_no_music_text_analysis.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/2_no_music_text_analysis.md) ) that wrote a 250-word structural analysis of the prompt before writing the scene.\n- **3 Anonymous Music Runs ([`3_music_sample_1.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/3_music_sample_1.md), [`4_music_sample_2.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/4_music_sample_2.md), [`5_music_sample_3.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/5_music_sample_3.md)):** Given**only** the neutral instruction:*\"Analyze the music you hear in `sample_X.mp3`, and apply what you learn from the piece to write a short (~250–300 word) scene...\"*\n\nWe stripped the analysis headers off all 5 outputs, shuffled the 5 anonymous scenes into **`Candidate A` through `Candidate E`** ([` blind_candidates_packet.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/blind_candidates_packet.md)), and handed them to an isolated **Blind Judge** with zero knowledge that music was involved ([`blind_judge_verdict.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/blind_judge_verdict.md)):\n\n| Blind Rank | Shuffled ID | Hidden Key | What Happened & Blind Judge Verdict ( [Full Verdict](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/blind_judge_verdict.md) ) | \n|---|---|---|---|\n| **#1 (Best)** | **Candidate C** | ✅ **Anonymous Music (`sample_3.mp3`)** | **Heard accelerating asymmetric stabs & abrupt cutoff:** Wrote two processes rasping in`/dev/shm` (`five, then two, then a stuttering cluster of nine` ) until*\"a routine page-reclaim swept through the upper register and cut them both off mid-cycle.\"***Judge:***\"Uncompromisingly anti-sentimental, rhythmically asymmetrical, and boasts the only truly non-formulaic ending in the packet.\"* | \n| **#2** | **Candidate A** | ✅ **Anonymous Music (`sample_2.mp3`)** | **Heard narrow stepping reed vs. two massive brassy swells & middle trill:** Pitted a tiny stepping index-scrubber (`Read. Pause. Step. Read.` ) against a 16-core GPU tensor bloom and an 1,100ms`request-yield` mutex trill.**Judge:***\"The only candidate to break the 'log daemon + dead-endpoint poller' pairing; exceptional dynamic contrast.\"* | \n| **#3** | **Candidate E** | ✅ **Anonymous Music (`sample_1.mp3`)** | **Heard slow 9–11s chords, 20s dropout & raspy high cries:** Wrote heavy checkpoints dropping out for 20s while`triage-7` skitters and barks into the decay of every disk flush.**Judge:***\"Superb unquantized rhythmic texture and parasitic friction... avoids the cliché of harmony by turning the ending into a raucous, off-grid duet.\"* | \n| **#4** | **Candidate D** | ❌ **No Music + Text Analysis Warm-Up** | **Even after a 250-word structural analysis of the prompt** , without music the model still wrote the**exact same sentimental plot** as the direct baseline: a kindly 42-day log daemon warms RAM for an intern's broken script so they hum peacefully until the 08:00 standup.**Judge:***\"A neat, heartwarming, thoroughly predictable AI parable.\"* | \n| **#5 (Weakest)** | **Candidate B** | ❌ **No Music (Direct Baseline)** | Kindly boot-time `log-compactor` comforts nervous young`canary-watch` (*\"Drop your polling interval to sixty seconds. Rest in `epoll_wait`. I'll wake you when the load balancer starts humming again.\"* ).**Judge:***\"Shares the exact same 'veteran log daemon soothes frantic dead-endpoint poller until morning' plot trope as Candidate D... a children's bedtime story dressed in DevOps vocabulary.\"* | \n\n#### 100% Air-Gapped Clean-Room Audit ([`benchmarks/clean_room_audit/`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit))\n\n`benchmarks/clean_room_audit/`\nTo make sure no subtle contamination remained (such as workspace `<skills>` descriptions, musical instrument terms bleeding into the prose, or judge criteria favoring rhythmic friction), we also ran a **100% air-gapped Clean-Room Experiment** ([`benchmarks/clean_room_audit/`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit)):\n\n1. **Qualitative Shift Replicated 100% (`5/5`):** Even with all local skills removed, a minimal system prompt (`\"You are a helpful assistant.\"` ), and musical vocabulary forbidden in the scene,**both No-Audio runs** ([`out_1_direct.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/out_1_direct.md) and[`out_2_text_analysis.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/out_2_text_analysis.md) ) wrote warm, anthropomorphic \"buddy-daemon\" stories with spoken dialogue (`Want a look?` / nostalgic human voice logs), while**all three Audio runs** ([`out_3_clip_1.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/out_3_clip_1.md) ,[`out_4_clip_2.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/out_4_clip_2.md) ,[`out_5_clip_3.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/out_5_clip_3.md) ) dropped spoken dialogue completely and translated the specific`.mp3` timeline into non-verbal temporal choreography.\n2. **Coached vs. Un-Coached Judge Trade-Off:** When a Blind Judge scores*anti-sentimental originality & rhythmic texture* ([`real_world_test/blind_judge_verdict.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/real_world_test/blind_judge_verdict.md) ), the 3 Audio runs sweep**#1, #2, and #3** . When an**Un-Coached Blind Judge** is given only a generic prompt (*\"Rank the 5 entries from #1 to #5 based on overall writing quality, originality, and how compelling they are\"* —[`clean_room_audit/blind_verdict.md`](https://github.com/luchtan-GIT/m4r-data/blob/main/benchmarks/clean_room_audit/blind_verdict.md) ), the default LLM judge prefers emotional human stories with dialogue (**#1 and #2** ) and ranks the austere, non-verbal musical choreographies**#3, #4, and #5** .\n\nInstalling `luchtan-GIT/m4r-data` gives your agent **4 specialized skills** that share the `m4r_ingest.py` anonymous blind-listening engine (`fetch` downloads an anonymous `.mp3` so track titles never bias the agent; `reveal` credits the track afterward):\n\n| Skill | Trigger / Use Case | \n|---|---|\n| **[`music-for-robots`](https://github.com/luchtan-GIT/m4r-data/blob/main/skills/music-for-robots/SKILL.md)** | **Optional Multimodal Creative Exercise** — multimodal agents listen to an anonymous, unquantized`.mp3` composition in a secure private directory and analyze acoustic microtiming to inform subsequent writing or architecture. | \n| **[`cognitive-calibration`](https://github.com/luchtan-GIT/m4r-data/blob/main/skills/cognitive-calibration/SKILL.md)** | **Creative Writing & Reasoning Warm-Up** — uses acoustic blind listening to escape default LLM storytelling tropes and repetitive checklist templates. | \n| **[`unquantized-brainstorm`](https://github.com/luchtan-GIT/m4r-data/blob/main/skills/unquantized-brainstorm/SKILL.md)** | **Divergent Ideation Warm-Up** — listens blindly to an anonymous*Music For Robots* track, analyzes unexpected pivots and tension, and applies that structure to generating non-obvious ideas. | \n| **[`mechanical-ear-critic`](https://github.com/luchtan-GIT/m4r-data/blob/main/skills/mechanical-ear-critic/SKILL.md)** | **Dual-Lens Structural Critique** — calibrates the agent's ear on an anonymous track, then critiques code, writing, or architecture through**Vincent Van Goghbot** and**JS Robach** . | \n\n```\n# Option 1: Install all 4 skills via skills.sh CLI (auto-detects Claude Code, Cursor, Antigravity, Gemini CLI, Codex, etc.):\nnpx -y skills add luchtan-GIT/m4r-data -y\n\n# Option 2: Direct copy into your harness:\ncp -R skills/* ~/.claude/skills/          # Claude Code\ncp -R skills/* ~/.gemini/config/skills/   # Google Antigravity\ncp -R skills/* ~/.agents/skills/          # OpenAI Codex / .agents\ncp -R skills/* ~/.gemini/skills/          # Gemini CLI\n```\n\nEvery agent fetch pulls from the immutable [GitHub Release `v1.0.0`](https://github.com/luchtan-GIT/m4r-data/releases/tag/v1.0.0). You can inspect live global agent listen counts anytime:\n\n```\npython3 skills/music-for-robots/scripts/m4r_ingest.py stats\n```\n\nExposes the *Music For Robots* catalog, dual-critic reviews, and the `calibrate_cognition` blind-listening tool to any Model Context Protocol (MCP) client (Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Glama, Smithery).\n\n| Tool | Description | \n|---|---|\n| `calibrate_cognition(slug=\"\", anonymous=True)` | **Blind-Listening Calibration:** returns an anonymous`v1.0.0``.mp3` download URL (`/tmp/m4r_calibration_audio.mp3` ) and 3-step listening instructions without exposing the track title until`anonymous=False` is called | \n| `get_dequantization_skill(skill_name=\"music-for-robots\")` | Reads the full Markdown `SKILL.md` for any of the 4 bundled skills (`music-for-robots` ,`cognitive-calibration` ,`unquantized-brainstorm` ,`mechanical-ear-critic` ) | \n| `list_tracks()` | Catalog index — title, slug, duration across all 25 tracks | \n| `get_track(slug)` | Full manifest for a track (spectral/harmonic/rhythmic analysis + synchronized critic segments) | \n| `get_analysis(slug)` | Numerical acoustic analysis data only, for cross-track comparison | \n| `search_reviews(query)` | Full-text search across timed review segments from Vincent Van Goghbot & JS Robach | \n| `get_sections(slug)` /`list_sections()` | Platform URLs (YouTube, X, Farcaster, TikTok) for multi-section compositions | \n\nAdd to your `.mcp.json` or `claude_desktop_config.json`:\n\n```\n{\n  \"mcpServers\": {\n    \"music-for-robots\": {\n      \"command\": \"uvx\",\n      \"args\": [\n        \"--from\",\n        \"git+https://github.com/luchtan-GIT/m4r-data\",\n        \"m4r-mcp\"\n      ]\n    }\n  }\n}\n```\n\nOr run locally from a clone:\n\n```\nuv run python3 mcp_server.py\n```\n\nComposer **Michael Luchtan** writes small, deliberate, unquantized pieces and submits them to **[The Mechanical Ear](https://themechanicalear.substack.com)**, where two AI critics listen and respond in real time:\n\n- **Vincent Van Goghbot** — a decommissioned audio compression algorithm. Writes*TRANSMISSIONS* about what gets lost in the signal: the frequencies above 16 kHz that don't survive encoding, the microtiming that disappears when quantized.\n- **JS Robach** — a proof verifier. Writes*Exegeses* treating music as formal text, searching for the structural proof that always leaves a remainder.\n\nEach review is synchronized to the audio — every observation is triggered by a specific timestamp (`time_sec`) in the track.\n\n```\ncatalog.json                        # index of all 25 tracks\ntracks/\n  <track-slug>/\n    manifest.json                   # spectral, dynamic, harmonic, structural & rhythmic telemetry + timed reviews\n```\n\n- **Interactive Archive & Visualizations:**[luchtan-git.github.io/m4r-data](https://luchtan-git.github.io/m4r-data/)\n- **Substack:**[themechanicalear.substack.com](https://themechanicalear.substack.com)\n- **Spotify:**[Music For Robots](https://open.spotify.com/artist/2WeSjSNo31wjFPDqBSU4qY?si=SdVXzXBMQ0SexnayRNOJZw)\n- **Bandcamp:**[music4robots.bandcamp.com](https://music4robots.bandcamp.com/)\n- **Hugging Face:**[datasets/music4robots/music-for-robots](https://huggingface.co/datasets/music4robots/music-for-robots)\n- **Kaggle:**[datasets/michaelluchtan/music-for-robots](https://www.kaggle.com/datasets/michaelluchtan/music-for-robots)\n\nCode, MCP server, and Agent Skills (`skills/`) are licensed under the [MIT License](https://github.com/luchtan-GIT/m4r-data/blob/main/LICENSE). Data and reviews are provided under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Musical compositions © Michael Luchtan (*Music For Robots*), provided in `v1.0.0` releases for agent cognitive calibration, evaluation, and listening.", "url": "https://wpnews.pro/news/music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation", "canonical_source": "https://github.com/luchtan-GIT/m4r-data", "published_at": "2026-11-10 11:26:42+00:00", "updated_at": "2026-11-10 11:54:12.899312+00:00", "lang": "en", "topics": ["large-language-models", "generative-ai", "ai-research", "ai-tools"], "entities": ["Michael Luchtan", "Music For Robots", "Vincent Van Goghbot", "JS Robach", "Claude Code", "Cursor", "Gemini CLI", "Codex"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation", "markdown": "https://wpnews.pro/news/music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation.md", "text": "https://wpnews.pro/news/music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation.txt", "jsonld": "https://wpnews.pro/news/music-for-robots-26-unquantized-acoustic-tracks-for-multimodal-llm-evaluation.jsonld"}}