Reproducibility bundle
One download with everything needed to rerun a real reconstruction and check what it has in common with ours: https://api.apertureapi.com/static/samples/reproducibility-bundle.zip
What is inside
| File | What it is |
|---|---|
inline-sample-512.png |
The input: a 512 × 512, 16-bit synthetic in-line hologram |
acquisition_metadata.json |
Wavelength, pixel pitch, hologram type, object distance, and how the frame was made |
settings.json |
The exact form fields sent with the request |
run.py |
The Python script that submitted it (no account needed) |
reference-output.zip |
The result package the live service returned for that request |
RECOMPUTE.md |
How to rerun it, what matches, and what currently does not |
Rerun it
unzip reproducibility-bundle.zip -d bundle && cd bundle
pip install requests
python run.py # writes my-output.zip
Then compare provenance.json in my-output.zip against the same file in
reference-output.zip. The provenance ID, settings hash and input commitment
should match: same input, same settings, same software version. The output
arrays and the evidence report ID will not match today. On the in-line path,
two runs of the same request currently return different arrays and quality
metrics; we measured this on 2026-10-12 and list it in the Validation
docket. RECOMPUTE.md has the details.
Why this matters
A methods section says what was done. This bundle lets a reader check the input, settings and software, and it keeps the exact output that was reported. Because outputs are not yet reproducible run to run, the archived output is the record; a rerun is a check of the inputs, not of the numbers. Use the same layout for your own data: input, metadata, settings, script, output.
Next step
Cite the result with the Methods supplement. Questions to system@apertureapi.com.