Typical handoff
final-figure-4-revised.png
- unknown source bytes
- instrument identity in a notebook
- calibration certificate elsewhere
- analysis settings no longer reconstructable
Open research software · measurement provenance
From a measured value to reviewable evidence.
BenchLineage keeps research intent, instrument identity, calibration coverage, raw data, deterministic analysis, uncertainty, and publication in one local chain.
Python 3.11+ · zero runtime dependencies · offline · MIT License
Run rc-baseline-001SYNTHETIC EXAMPLE
StudyRC low-pass response
→
Instrumentscope-01 · SYN-0042
CalibrationValid at run time
→
Raw observation41-point sweep · CSV
Analysisfc = 1.5749 kHz
→
UncertaintyU = 0.00667 · k = 2
Sealed evidence rootsha256:db35e8d47945… · verified
9durable artifact layers
4built-in analyses
0runtime dependencies
1verifiable publication ZIP
The plot is not the experiment record.
Six months after a measurement, a figure may survive while the exact instrument, calibration window, run conditions, raw file, and uncertainty chain disappear into filenames and private notes.
BenchLineage handoff
A sealed, audited evidence graph.
- plain JSON and original raw files
- run-to-instrument and run-to-study links
- calibration coverage at acquisition time
- manifested report and deterministic ZIP
One chain, six explicit claims.
Each transition leaves a durable artifact. Audit checks relationships and timing; seal checks current bytes; bundle packages the complete handoff.
Study
Objective, hypothesis, protocol, and tags.
studies/*.jsonTraceability
Instrument identity and calibration window.
instruments/*.jsonRun
Conditions, deviations, operator, and raw paths.
runs/*.jsonAnalysis
Deterministic outputs and uncertainty budget.
analysis/*.jsonAudit + seal
Semantic links, date coverage, and SHA-256 root.
seals/*.jsonBundle
Report, records, raw bytes, and member manifest.
evidence.zipA portable evidence layer, not another laboratory platform.
Mature ELNs and data systems solve broader institutional problems. BenchLineage optimizes for one researcher or a small EE group that needs better provenance without operating a server.
| System type | Best at | Infrastructure | BenchLineage boundary |
|---|---|---|---|
| ELN / LIMS | Collaboration, permissions, inventory, scheduling | Institutional service or managed deployment | Use it when governance is the main need |
| Instrument automation | Drivers, benches, scheduling, real-time control | Hardware-specific stack | BenchLineage intentionally does not control acquisition |
| ML / data versioning | Runs, models, pipelines, large artifacts | Local or remote tracking/storage | Calibration and uncertainty are not their central record |
| BenchLineage | Small-team EE evidence completeness and handoff | Python and ordinary files | Local graph, calibration audit, seal, report, bundle |
No individual feature is novel. The practical hypothesis is that a smaller, inspectable workflow can prevent evidence loss with less setup than a general platform. Adoption and independently verified bundles—not this comparison table—would validate that hypothesis.
One archive a reviewer can verify.
The bundle command refuses a failing audit or invalid latest seal. It then writes fixed member order, timestamps, permissions, compression, and a plain SHA-256 manifest.
Publication handoff
Deterministic by construction.
Rebuilding from unchanged workspace bytes produces the same ZIP bytes. Verification reads the archive without extracting and reports unsafe, duplicate, missing, added, or changed members.
benchlineage bundle thesis-bench --output evidence.zip
benchlineage verify-bundle evidence.zip
benchlineage verify-bundle evidence.zip
BUNDLE-MANIFEST.jsonsha256
workspace/benchlineage.json45440da0…
workspace/studies/rc-filter.json57dfb43b…
workspace/data/raw/rc-baseline.csv4266851a…
workspace/analysis/rc-baseline-001.jsona7720b91…
workspace/reports/demo-report.htmlself-contained
workspace/seals/seal-2026…jsonroot valid
The claim boundary is part of the artifact.
Byte integrity and scientific validity are different questions.
Not a calibration authority.It records certificate references and date coverage; it does not authenticate the certificate.
Not a trusted timestamp.A local clock and SHA-256 root do not prove when the experiment happened.
Not an instrument controller.Acquisition remains independent so labs can keep their existing drivers and formats.
Not scientific validation.An internally consistent bundle can still describe a flawed or fabricated experiment.
Inspect the complete synthetic evidence chain.
All instrument names, serials, certificates, and measurements in the demonstration are fictional. The report demonstrates the data model and audit surface; it makes no physical measurement claim.
python -m pip install benchlineage