Measured product lead custody receipt.
github.com/Zer0pa/ZPE-BioDEVELOPER-READY
LIVE
PTB-XL OPEN
ECG With Every Beat Encoded
ECG-only staged codec · ZPE-Bio · PyPI zpe-bio v0.2.1 · github.com/Zer0pa/ZPE-Bio
ZPE-Bio compresses heart-rhythm recordings and publishes the replay error on every dataset it touches. On MIT-BIH, all 48 records return inside the 2.32% clinical target with a mean replay error of 1.12%.
PTB-XL processes cleanly but one record drifts to 5.29%, so the wider ECG surface is named, not claimed. This is a developer-ready archival codec, not a clinical device, wearable, or regulatory product.
01 · THE GAPMEASURED, NOT ASSUMED
ECG codecs that report file size without naming their replay error leave clinicians guessing.
02 · MARKETSADJACENT FORECASTS
Where this codec belongs: archive replay and cardiology data stewardship. No clinical-device, wearable, or regulatory readiness claim.
03 · VALUE
48/48
MIT-BIH records replay inside the 2.32% target · mean 1.12% replay error
04 · INSIGHT
Every replayed ECG should carry its measured change.
05.1 · CURRENT TECHSIZE WITHOUT FIDELITY
Lossless compressors preserve bytes and report file-size ratios. They tell a cardiology archive nothing about how closely a replayed waveform tracks the original signal on any named clinical corpus.
05.2 · OUR TECHREPLAY ERROR PUBLISHED
ZPE-Bio reports replay error directly per dataset. On MIT-BIH, 48 of 48 records return at mean 1.12%, inside the 2.32% clinical target a cardiology archive can show its review board. PTB-XL processes 100 of 100 records with full integrity but a worst-case record reaches 5.29%, so that corpus is named openly as out of contract rather than buried.
05.3 · BENCHMARKSFOUR CORPORA
MIT-BIH48/48PASS
NSTDB15/15PASS
EDB90/90PASS
PRD target2.32%MIT-BIH
Reported today: MIT-BIH 48/48 · NSTDB 15/15 · EDB 90/90 · PTB-XL 5.29% out.
06 · MEASUREMENTPRD BY CORPUS
Replay error reported per corpus, with MIT-BIH as the reference.
06.1 · COMPARATIVE PERFORMANCEPRD BY CORPUS
MIT-BIH Arrhythmia (48 records), NSTDB (15), EDB (90), PTB-XL (100 sampled). MIT-BIH target: replay error ≤ 2.32%. File-size winners stay gzip and zstd; this codec competes on fidelity, not bytes.
07 · KEY METRICSMEASURED RESULTS
07.1 · MIT-BIH INTEGRITY
48/48
MIT-BIH records replay clean · integrity pass
07.2 · MEAN CR
1.12%
MIT-BIH aggregate replay error · inside the clinical target
07.3 · PTB-XL INTEGRITY
100/100
PTB-XL 100-record sample · worst 5.29% disclosed
07.4 · PRD TARGET
2.32%
MIT-BIH clinical target · met by mean
07.5 · MEAN PRD
1.12%
MIT-BIH 48-record mean · inside target
08 · DETERMINISMCI PARITY PYTHON + RUST
Python and Rust replay ECG; the surface returns consistently.
08.1 · WHAT DETERMINISTIC MEANSECG SURFACE ONLY
Repo CI runs Python and Rust round-trips on every push across MIT-BIH, NSTDB, the European ST-T Database, and PTB-XL. Replay error is reported per dataset. MIT-BIH sits inside the 2.32% target at a mean of 1.12%.
PTB-XL logs 100 of 100 records with full integrity, but the worst-case replay error reaches 5.29% and stays visible in the public results. PyPI 0.2.1 is stale and is not presented as fresh; the surface is ECG-only.
08.2 · THE FIDELITY GAP
Honest Blocker ·
PTB-XL processes 100 of 100 records, but the worst-case replay error reaches 5.29%, outside the 2.32% MIT-BIH target. PyPI is stale at zpe-bio 0.2.1. Wearable-data scope and the 0x0400 marker policy are open. No clinical, regulatory, EEG, wearable, or raw-size superiority claim.
09
FIVE PATHS FROM ONE clinical contract.
09.1 · THE AMBITION
The hinge is the disclosed boundary, not the passed contract. A codec that names its worst-case excursion alongside its pass rate becomes the substrate cardiology archives can adopt before regulatory engagement — because the open question is already on the table, not waiting in the field.
09.2 · WHAT WORKS NOWEXTERNAL
MIT-BIH 48/48 records pass; the mean replay error of 1.12% sits inside the 2.32% clinical target.
09.3 · WHAT'S STILL OPENEXTERNAL
PTB-XL excursion at 5.29%, stale PyPI release, wearable scope, and marker-policy decision are open.
09.4 · ARCHIVES · NEAR-TERM (12–24 MO)
Cardiology archives gain a fidelity floor
An academic cardiology lab can stop hoarding raw float32 MIT-BIH waveforms because every record returns inside a documented 2.32% replay error. Storage budgets fall and the archived signal still satisfies clinical research review.
09.5 · PROCUREMENT · NEAR-TERM (12–24 MO)
Buyers decide before deployment, not after
A cardiology archive buyer who sees the PTB-XL 5.29% excursion published alongside the MIT-BIH pass rate can choose with eyes open. Disclosure shifts procurement from a post-purchase surprise to a pre-purchase conversation about which corpora the buyer actually needs.
09.6 · CLINICAL RESEARCH · MID-TERM (24–48 MO)
Reproducible studies travel between sites
When two cardiology research groups share compressed waveforms with identical replay error, a finding produced at one institution can be re-run at another without arguing about whether the data was altered in transit. Reproducibility stops depending on raw-byte custody.
09.7 · BOUNDARY CLOSURE · MID-TERM (24–48 MO)
PTB-XL becomes a routine corpus
If codec tuning brings the PTB-XL worst-case replay error from 5.29% under the 2.32% target, the PTB-XL population — broader demographics, more leads, longer recordings — joins the safely archivable set. The cardiology research community gains a second large public corpus with a fidelity guarantee.
09.8 · INDUSTRY STANDARD · PARADIGM (48 MO+)
Fidelity claims arrive with disclosure attached
ECG codecs across the industry begin publishing their per-dataset thresholds, worst-case excursions, and pass rates as standard practice. Hospitals, regulators, and procurement teams stop accepting a single global fidelity number and start asking which corpus, which boundary, which pass rate.