Scientific provenance
Make the computation
inspectable.
Ostrin emits a machine-readable artifact with the source hash, compiler target and an experimental, type-aware effect analysis. Known rows, callback variables and unresolved calls make missing evidence visible before anyone calls a result reproducible.
Recorded artifact
One report, every assumption visible.
This page is generated from ostrinc --provenance-report --json --target wasm32-wasi and checked against the source during the website build.
Loading provenance artifact…
Evidence boundary
Honest about what is missing.
The artifact is deliberately conservative. It is useful evidence for tooling and publication workflows, and it does not silently turn an effect analysis into a proof of purity or reproducibility.
Available now
Source hashes are normalized across line endings, effect sites are serialized with stable paths, the target and WASI profile are explicit, and the report is available as text or JSON for native, WASI and browser tooling. The experimental typed analysis resolves known calls by receiver type and carries callback variables and open rows into the artifact.
Unresolved calls
Calls the checker cannot resolve remain separate from known effects. Callback variables and open rows expose additional uncertainty. A closed empty row is not a purity proof, and this analysis does not establish an R level.
- Loading…
Reproducibility requirements
The artifact lists evidence required by each inferred effect, such as a seed and RNG algorithm, input snapshots, captured outputs or a scheduler contract. These are constraints only; the report does not certify an R level.
- Loading…
Next evidence layer
The standalone report remains conservative about invocation inputs and host capture. The reproducible experiment bundles now add embedded input snapshots where applicable, invocation output hashes, compiler effect summaries and CI host metadata when generated by Pages CI; the eleven input-driven Viz bundles demonstrate the narrower R2 input-snapshot contract, and the nine seeded fixtures retain R1 algorithm evidence. CI also writes a bounded WASI host replay receipt after replaying all 29 downloaded bundles; it records the run identity and per-fixture hashes but does not promote the static R3 count or claim cross-platform equivalence. The design contract is documented in scientific effects and provenance ↗.