BenchLineage turns an ordinary project directory into a verifiable chain from research intent to instrument identity, calibration coverage, raw measurement, derived analysis, uncertainty budget, and published report.

It does not replace a full electronic laboratory notebook, instrument-control framework, or institutional data repository. It focuses on a smaller failure mode that is especially common in electrical-engineering research: six months after a measurement, the plot survives but the exact instrument, calibration window, run conditions, raw file, and analysis lineage do not.

The default workflow is offline. There is no account, server, database, API key, telemetry, or runtime dependency.

When the record needs to leave the project directory, BenchLineage can produce either its strict byte-verifiable evidence bundle or an interoperable .eln archive for electronic laboratory notebook exchange.

Live demonstration · Synthetic report · PyPI package · Methodology · CI · Python 3.11+ · MIT · zero runtime dependencies

Important

BenchLineage is a research recordkeeping tool, not a calibration authority, regulatory compliance system, or safety certification. A cryptographic digest shows whether recorded bytes changed; it cannot prove that a physical measurement was performed correctly or honestly.

See it before installing

• Project site
• Synthetic interactive report
• Example workspace
• Methodology
• Competitive landscape

Every instrument name, serial number, certificate, and measurement in the public demonstration is synthetic. The demo illustrates the evidence model; it makes no claim about physical hardware.

The evidence chain

_{Every stage writes an inspectable file; the final bundle keeps those files together.}

Layer	Durable artifact	Question it answers
Workspace	benchlineage.json	Who owns this record and which format does it use?
Study	studies/*.json	Why was the experiment run and what was planned?
Instrument	instruments/*.json	Which physical device produced the observation?
Calibration	calibrations/*.json	Was traceability recorded for the run date?
Run	runs/*.json	Who did what, when, under which conditions?
Raw data	data/raw/*	What was actually observed?
Analysis	analysis/*.json	Which deterministic derivation produced the result?
Seal	seals/*.json	Do the current bytes match the published evidence root?
Report	reports/*.html	Can another person inspect the chain without this package?

A result is more than a number

The figure below uses one fictional RC-filter result to show the five kinds of context that travel with it. This is the smallest useful mental model for BenchLineage: the reported value sits in the middle, and every surrounding statement has a recorded source.

_{Integrity checks answer whether recorded files changed; they do not certify whether the experiment was scientifically correct.}

Install

Install from PyPI:

python -m pip install benchlineage
benchlineage --version

For an isolated one-off demonstration with pipx:

pipx run benchlineage demo my-bench --seed 20260804

PyPI distributions are published from the tagged GitHub workflow with a short-lived OIDC credential and a public provenance attestation. Each GitHub Release also carries the same wheel, source archive, and a SHA-256 checksum file.

For development:

git clone https://github.com/CAOShurong/benchlineage.git
cd benchlineage
python -m pip install -e .

Try the complete demonstration

benchlineage demo my-bench --seed 20260804
benchlineage audit my-bench
benchlineage verify my-bench

Open my-bench/reports/demo-report.html. It is a single HTML file with embedded CSS, JavaScript, charts, audit evidence, and source records. It uses no CDN and sends no network requests.

The generated demonstration includes:

• a 41-point RC low-pass sweep with a baseline component value;
• a second sweep after a deliberate resistor substitution;
• a ten-point buck-converter load and efficiency sweep;
• three synthetic instruments and three synthetic calibration records;
• three derived analyses and three uncertainty budgets;
• a SHA-256 evidence inventory and root digest.

Start a real workspace

The command blocks in this section use the POSIX-shell \ line continuation. In PowerShell, replace each trailing \ with a backtick (`), or put the command on one line. Use a JSON file for structured arguments as shown below; this avoids shell-dependent quote handling.

benchlineage init thesis-bench \
  --title "Wide-bandgap converter characterization" \
  --owner "Your Name" \
  --owner-email "you@example.org" \
  --owner-given-name "Your" \
  --owner-family-name "Name" \
  --data-license-url "https://spdx.org/licenses/CC-BY-4.0.html" \
  --data-license-name "CC BY 4.0" \
  --data-license-description "Experimental data released under CC BY 4.0."

benchlineage add-instrument thesis-bench \
  --id scope-01 \
  --kind oscilloscope \
  --manufacturer Tektronix \
  --model MSO58 \
  --serial C012345

benchlineage add-calibration thesis-bench \
  --id scope-01-2026 \
  --instrument scope-01 \
  --performed-at 2026-01-18T09:00:00+08:00 \
  --due-at 2027-01-18T09:00:00+08:00 \
  --certificate CAL-2026-001 \
  --standard-uncertainty 0.0025 \
  --unit V

benchlineage create-study thesis-bench \
  --id gate-loop \
  --title "Gate-loop ringing characterization" \
  --objective "Measure overshoot across three gate resistances." \
  --hypothesis "Higher gate resistance reduces overshoot at the cost of switching loss." \
  --protocol "Record 200 switching events per resistor at fixed bus voltage and load." \
  --tag power-electronics \
  --tag switching

Raw evidence is intentionally added as an ordinary file. This keeps acquisition independent from BenchLineage and avoids locking a lab into one driver ecosystem:

thesis-bench/data/raw/gate-loop-rg2p2.csv

Save the run conditions as conditions.json:

{
  "bus_voltage_v": 400,
  "load_current_a": 20,
  "gate_resistance_ohm": 2.2
}

Record the run. The --conditions option accepts either a JSON file path or inline JSON; the file form works unchanged in POSIX shells and PowerShell:

benchlineage add-run thesis-bench \
  --id gate-loop-rg2p2-001 \
  --study gate-loop \
  --operator "Your Name" \
  --started-at 2026-08-04T14:00:00+08:00 \
  --instrument scope-01 \
  --raw-file data/raw/gate-loop-rg2p2.csv \
  --conditions conditions.json

Then analyze, audit, seal, and report:

benchlineage analyze thesis-bench \
  --run gate-loop-rg2p2-001 \
  --kind column_summary

benchlineage audit thesis-bench --output audit.json
benchlineage seal thesis-bench --label "Preprint figure 4 evidence"
benchlineage verify thesis-bench

benchlineage report thesis-bench \
  --title "Gate-loop characterization evidence" \
  --note "Dataset frozen before manuscript submission." \
  --output gate-loop-report.html

Publish a reviewable evidence bundle

After audit and sealing, create one deterministic ZIP for a collaborator, reviewer, or data repository:

benchlineage bundle thesis-bench --output thesis-bench-evidence.zip
benchlineage verify-bundle thesis-bench-evidence.zip

The bundle includes the workspace records, raw files, reports, latest seal, a human-readable README, and BUNDLE-MANIFEST.json. The manifest lists every included path, byte length, and SHA-256 digest. Rebuilding from unchanged bytes produces the same ZIP bytes.

The verifier refuses unsafe archive paths and reports duplicate, missing, added, or changed members without extracting the bundle. This is an integrity check, not a digital signature or a claim that the experiment was valid.

Move the record into an electronic lab notebook

The ELN Consortium format is a ZIP-packaged RO-Crate exchange format implemented by products including eLabFTW, Kadi4Mat, PASTA, RSpace, and SampleDB. Export the same sealed workspace without copying records into a vendor-specific form by hand:

benchlineage export-eln thesis-bench --output thesis-bench.eln
benchlineage verify-eln thesis-bench.eln

The exporter preserves every original workspace file and adds flattened JSON-LD that describes the workspace owner, studies, instruments, calibrations, experimental runs, raw files, and analyses. Instruments become RO-Crate IndividualProduct entities; runs and analyses become CreateAction provenance. Each local file carries its byte length and SHA-256 digest.

When init receives a data-license URL, export-eln links the root Dataset to a RO-Crate CreativeWork license entity with its declared name and description. The URL must be an absolute HTTP(S) URL without embedded credentials. BenchLineage does not infer a license: existing or new workspaces that omit the declaration retain an explicit contact-the-author warning. The public synthetic demo declares the repository's MIT license; that declaration does not license user data.

The Dataset identifier carries the latest BenchLineage evidence root. As with a native seal, that root excludes generated reports and seal records; the ELN's per-file SHA-256 entries still cover every member of the finished archive.

verify-eln reads the archive without extracting it. It rejects unsafe or duplicate paths, multiple archive roots, malformed graph references, missing or unlisted payloads, and byte or digest mismatches. Standard ro-crate-preview.html, ro-crate-preview_files/, and ELN metadata signature members are reported as unverified_ancillary instead of evidence: their presence is allowed, but BenchLineage does not claim their bytes or signature trust were verified unless the metadata separately declares them as hashed File entities. Exports are deterministic for the same sealed bytes and output filename.

BenchLineage targets the ELN Consortium's currently exercised RO-Crate 1.1 compatibility surface, which maximizes compatibility with existing importers. It does not claim that every target ELN will preserve every BenchLineage-specific field in its own interface; the original JSON, CSV, reports, and seals remain in the archive even when an importer ignores richer metadata. The exact mapping and independent validation evidence are documented in ELN interoperability. BenchLineage does not convert a third-party .eln archive into its stricter workspace schema; verify-eln is a read-only archive check, not an import command.

Built-in analyses

BenchLineage deliberately implements a small, inspectable analysis core:

Analysis	Detected columns	Main outputs
Frequency response	frequency_hz, vin_v, vout_v, optional phase_deg	gain, dB curve, passband estimate, interpolated −3 dB cutoff
Power efficiency	vin_v, iin_a, vout_v, iout_a, optional load_ohm	input/output power, loss, efficiency, peak operating point
Linear calibration	reference, observed	slope, intercept, R², RMSE, error summary
Column summary	any all-numeric CSV	count, mean, median, extrema, standard deviation, standard error

These are reference implementations, not an attempt to replace NumPy, SciPy, SPICE, MATLAB, or domain-specific analysis. The JSON analysis artifacts are designed so more specialized tools can write compatible results.

Uncertainty budgets

An analysis can include Type A or Type B components:

[
  {
    "name": "vertical accuracy",
    "limit": 0.005,
    "distribution": "rectangular",
    "source": "certificate CAL-2026-001"
  },
  {
    "name": "repeatability",
    "standard_uncertainty": 0.0016,
    "distribution": "normal",
    "source": "twenty repeated acquisitions"
  }
]

benchlineage analyze thesis-bench \
  --run gate-loop-rg2p2-001 \
  --uncertainty uncertainty.json \
  --coverage-factor 2

The engine converts stated limits for normal, rectangular, and triangular distributions into standard uncertainties, applies sensitivity coefficients, combines independent contributions by root-sum-square, and reports component variance shares. See Uncertainty model for assumptions and limitations.

Auditing and sealing are different

benchlineage audit checks semantic relationships:

• referenced study, instrument, raw-data, and analysis records exist;
• calibration dates cover each run date when possible;
• raw paths remain inside the workspace;
• unreferenced raw files are reported;
• the latest seal still matches the evidence.

benchlineage seal checks byte identity. It records the SHA-256 digest and byte length of each evidence file, then hashes the ordered inventory into one root digest. Reports and previous seals are excluded so presentation can be regenerated without changing the recorded evidence.

Compare two recorded states without manually reading their inventories:

benchlineage diff-seals \
  thesis-bench/seals/seal-before.json \
  thesis-bench/seals/seal-after.json

The command returns machine-readable added, removed, and changed path lists plus both labels, timestamps, and root digests. It exits with status 0 only when the roots match, so it can gate a handoff or publication script.

A passing seal does not prove scientific truth. It proves only that the inventoried bytes match the declared root.

Why plain files?

Large ELN and LIMS platforms solve collaboration, permissions, inventory, scheduling, and institutional deployment. BenchLineage instead optimizes for:

• one researcher or a small team;
• offline operation on Windows, macOS, or Linux;
• Git-friendly review and long-term readability;
• instrument-agnostic ingestion;
• explicit, deterministic audit rules;
• a migration path rather than another data silo.

All core records are readable JSON and CSV. If BenchLineage disappears, the evidence remains usable.

Repository map

src/benchlineage/       CLI, analysis, audit, report, sealing, and bundle engine
schemas/                JSON Schema contracts for durable artifacts
demo/workspace/         committed synthetic example
site/                   public project site and generated report
docs/                   methodology, workflows, comparison, and design notes
tests/                  unit, integration, CLI, tamper, and report tests
scripts/                reproducible demo and repository checks

Project boundaries

BenchLineage does not currently:

• control instruments or promise SCPI-driver compatibility;
• sign records with a trusted timestamping authority;
• provide multi-user permissions or electronic signatures;
• store secrets safely;
• validate laboratory safety or regulatory compliance;
• infer uncertainty models automatically;
• replace raw binary formats with CSV;
• guarantee FAIR compliance merely because metadata fields exist.

Those boundaries are intentional and documented in the roadmap.

Development

python -m pip install -e .
python -m unittest discover -s tests -v
ruff check src tests scripts
ruff format --check src tests scripts
python -m compileall -q src tests scripts
python scripts/build_demo.py --check
python scripts/check_repository.py

The package targets Python 3.11 and newer and has zero runtime dependencies.

Responsible citation

If BenchLineage contributes to a published evidence workflow, cite the software version and the root digest of the sealed workspace. A software citation file is provided in CITATION.cff.

License

MIT. Contributions are welcome under the same license. Questions and concrete laboratory workflows belong in Discussions; reproducible defects and inspectable analysis proposals belong in Issues.