2026-09-09
Captured the first deterministic tree hash and public file inventory after O027 generation.
The measured public tree contained 1754 files.
Evidence records
evidence/universe/o027/O027_REPRODUCIBILITY.json
Evidence · experiment
Deterministic Rebuild Experiment: O027
A software-only experiment testing whether the O027 guided-experience build reproduces the same public files byte-for-byte when run again against the same packaged descendant.
Goal of the experiment
Test one bounded reproducibility claim for the O027 packaged build.
Hypothesis
Running the O027 builder twice against the same packaged descendant will produce identical public paths and byte streams.
Recorded hypothesis state: tested supported in scope.
Materials & setup
- Frozen O027 build tools and packaged descendant · software checkpoint
- Path-and-byte hash comparison · software measurement
Dated record
2026-09-09
Ran the O027 builder again against the same packaged descendant and rehashed every public path and byte stream.
Byte-identical files: 1754; changed: 0; added: 0; removed: 0.
Evidence records
evidence/universe/o027/O027_REPRODUCIBILITY.json
2026-09-09
Compared the resulting deterministic tree hash.
The recorded deterministic tree SHA-256 is 929bdb0773aa15880bf17ba58e0d17fb297039914bc0472d0d63ca3b5884e493.
Evidence records
evidence/universe/o027/O027_REPRODUCIBILITY.json
Observations
- 1754 public files matched byte-for-byte on rerun.
- No public files were added, removed, or changed in the reproducibility comparison.
Related ICU resources
Result
The O027 rebuild was deterministic for the tested packaged descendant and build procedure.
| Measure | Recorded value |
|---|---|
| Public Files | 1754 |
| Byte Identical | 1754 |
| Changed | 0 |
| Added | 0 |
| Removed | 0 |
| Tree Sha256 | 929bdb0773aa15880bf17ba58e0d17fb297039914bc0472d0d63ca3b5884e493 |
Limitations
- This result applies to the tested O027 packaged descendant and does not prove every future ICU build will be deterministic.
- The experiment did not test live deployment infrastructure, external network services, or concurrent writers.
- It measures file reproducibility, not product quality or user outcomes.
Evidence boundary
This is a first-party ICU editorial record built from retained project evidence. It is not a member testimonial, user outcome, physical trial, or community submission. Future community claims will remain attributed and unverified by default unless their own evidence is reviewed.
How ICU separates first-party evidence from future community contributions →