Measured product lead custody receipt.
github.com/Zer0pa/ZPE-MusicDEVELOPER-READY
LIVE
REAL-CORPUS OPEN
Encoding The Music Inside The Music
The score codec — what the composer wrote, not what the orchestra played · ZPE-Music · PyPI zpe-music 0.1.0 · github.com/Zer0pa/ZPE-Music
A recording captures what an orchestra sounded like on one Tuesday night. A score captures what the composer meant — every event, voice, articulation and expression beneath the sound. Every audio codec eats the score and emits a waveform.
ZPE-Music does the opposite: it encodes the score itself. On the declared MusicXML score surface, all six exactness axes resolve at 1.0 under 11/11 release checks. Real-corpus work against MuseScore and IMSLP catalogues is active, not done.
01 · THE GAPSCORE BECOMES SOUND
When a score is encoded as a waveform, the composer's intention disappears — it can't come back.
02 · MARKETSADJACENT FORECASTS
Symbolic-score archives have no published analyst category — MusicXML volume sits inside music-publishing and rights-management forecasts.
03 · VALUE
$10.8B
Music publishing grows; the symbolic-score archive sits unpriced inside publisher and rights workflows.
04 · INSIGHT
A waveform keeps the sound. a score keeps the music.
05.1 · CURRENT TECHCODECS EAT THE SCORE
DAWs encode audio. Streaming platforms deliver waveforms. AI music systems learn from recordings. The written structure beneath the sound — events, voices, articulation, expression — is outside every test. Codecs eat it and emit sound.
05.2 · OUR TECHKEEP THE SCORE
ZPE-Music encodes the score itself. On the declared MusicXML score surface it preserves events, parts, voices, articulations, expression fields, performance tuples and repeated-pitch notes, decoding byte-identical to the input. All six exactness axes resolve at 1.0 across 11/11 release checks in 3.39 s. Audio interpretation is explicitly out of scope.
05.3 · BENCHMARKSMUSICXML SCORE SURFACE
Axes6 / 6exactness 1.0
Checks11 / 11release pass
Verify3.39seconds
PyPI0.1.0stale pending
Surface: bounded MusicXML · real-corpus benchmark against MuseScore + IMSLP still pending.
06 · MEASUREMENTRELEASE ARTIFACT PROOFS
Six axes. Six exactness claims. each checked against a release artifact.
06.1 · COMPARATIVE PERFORMANCEMUSICXML 4.0 SCORE SURFACE
music_release_metrics.json + release_verification.json · 11 passed · 3.39 s · PyPI 0.1.0 stale. Six axes on the MusicXML 4.0 score surface. Audio, MIDI, continuous dynamics, pedal and performer state are out of scope. Source: github.com/Zer0pa/ZPE-Music.07 · KEY METRICSMUSICXML RELEASE PROOFS
07.1 · SCORE AXES
6 / 6EXACT
Every axis at exactness 1.000
07.2 · RELEASE CHECKS
11 / 11PASS
pytest score suite · strict
07.3 · VERIFY RUNTIME
3.39s
Release suite · 2026-04-25
07.4 · REAL-CORPUS
null
MuseScore + IMSLP · active, not done
07.5 · PUBLIC PYPI
0.1.0
Connected · stale pending release
08 · SCORE FIDELITYDECLARED FIELDS ONLY
Encode the declared score. Decode the declared score. the intention survives.
08.1 · WHAT DETERMINISTIC MEANSPER-ROUNDTRIP, SCORE NOT SOUND
Deterministic here means per-roundtrip, and it means the score, not the sound. On the declared MusicXML score surface, encode and decode reproduce the canonical fields across all six axes — SCORE_EVENT, PART, ARTICULATION, EXPRESSION, PERF_TUPLE, REPEATED_NOTE — resolving at exactly 1.0 in 3.39 seconds.
Audio waveforms, MIDI rendering, continuous tempo, dynamics curves, pedal state and performer state are outside the verified scope. We do not yet make the byte-identical claim against public real-corpus MusicXML; the MuseScore and IMSLP benchmark is active, not done.
08.2 · HONEST BLOCKER
Honest Blocker ·
Out of scope: audio waveforms, MIDI benchmarks, continuous tempo and dynamics curves, pedal and sustain state, performer state, raw MusicXML part-name identity. Next step: close the public MuseScore and IMSLP corpus benchmark, refresh the stale 0.1.0 PyPI release, and publish the corrected build with the real-corpus result attached.
09
What the composer wrote stays recoverable.
09.1 · THE AMBITION
The aim is a score that travels as the score — every event, voice, articulation and expression intact — instead of being flattened into a waveform every time it moves. A music industry built on recordings finally gets a first-class symbolic carrier for the work beneath the sound.
09.2 · WHAT WORKS NOW
On declared MusicXML today: six exactness axes at 1.0, eleven of eleven release checks pass in 3.39 s.
09.3 · WHAT'S STILL OPEN
The public MuseScore and IMSLP corpus benchmark is active; the 0.1.0 PyPI release is stale pending refresh.
09.4 · INTERCHANGE · NEAR-TERM (12–24 MO)
Scores move between editors without loss
A composer hands a Finale file to an engraver who opens it in Dorico, the engraver returns it through a publisher in Sibelius, and every voice, articulation and tied note arrives in the same place it started — no rebuild pass.
09.5 · ARCHIVES · NEAR-TERM (12–24 MO)
Catalogue librarians get a fidelity receipt
A music library that holds a publisher's MusicXML catalogue can prove the Beethoven sonata in the 2024 archive is exactly the score on the 2026 reading stand. The proof is a packet comparison, not a side-by-side render.
09.6 · VERSIONING · MID-TERM (24–48 MO)
Commissioned works carry a revision history
A composer revising a string quartet for a chamber orchestra and the publisher tracking changes both see exactly which bars moved between draft three and draft four. Score editing becomes a tracked process, the way code review is, instead of a stack of PDFs.
09.7 · MUSIC AI · MID-TERM (24–48 MO)
Symbolic music models train on clean ground truth
A music-AI team building a transformer over symbolic scores stops fighting the inconsistencies of heterogeneous MusicXML exports. The training corpus arrives in one encoded form, and the model learns the music a composer wrote rather than the quirks of the editor that exported it.
09.8 · IDENTITY · PARADIGM (48 MO+)
A musical work gets a fingerprint
Two encodings of the same Bach prelude — one from a 19th-century scholarly edition, one from a modern engraver — produce comparable packets. Rights societies, publishers and academic catalogues can talk about the work itself, not the file format that happened to carry it.