Evidence dashboard
Show the state.
Keep the claims separate.
This page combines current repository evidence with the historical v0.2 release record. Its release source revisions identify artifact builds; newer numerical and R3 metrics describe development evidence.
Compiler backend
Native lowering has a visible boundary.
The ratchet counts functions generated through IR and HIR, with the remaining AST fallback kept explicit.
Ownership
Memory behavior is tracked by evidence.
The ownership matrix separates tested cells, promoted support and known limitations so a broad claim cannot hide an unverified boundary.
Numerical validation
Residuals stay attached to the run.
The analytic matrix and FFT checks show interpreter/native parity, per-case tolerances, independent DFT residuals, inverse reconstruction and native allocation cleanup. A local report is useful evidence; CI status is shown separately. Inspect the validation contract ↗ · Read methods and limits ↗
Scientific effects
Effect coverage stays conservative.
The inventory publishes resolved and unresolved operation edges separately. Counts are distinct function-operation pairs, not source callsite totals; an incomplete inventory is visible instead of being presented as a static purity guarantee.
Reproducibility
R-levels describe what can be replayed.
R0 records exact CI replay; R1 and R2 add seed or input-snapshot contracts. R3 needs auditable, run-bound evidence. CI checks host capture and closed entry effect rows for all 30 registered bundles. Its archived package inventories and replays all 30 R0/R1/R2 bundles with the recorded WASM; surface also carries a run-bound receipt. The compiler records function-operation resolution categories and local targets, but these are edge-level summaries rather than per-callsite or runtime dispatch traces. A source scan checks literal input paths but does not observe runtime reads or resolve aliases, imports, FFI, or host access. Replay results do not promote any fixture to R3: the count remains zero. Symbolic callbacks stay visible, and a closed row does not prove purity.
GitHub Linguist
Recognition follows public use.
The local proposal, grammar and representative MIT samples are prepared. The external snapshot currently reports the public extension count and keeps the upstream PR pending until Linguist's usage threshold and distribution review are met.
Release evidence
v0.2 readiness and limits.
The published v0.2.0 met all 12 release criteria. Open research and platform limits remain visible in the linked evidence.
How to read this
Use the source artifacts.
The RC source revision identifies the older candidate archives. The published release source identifies the tagged binaries. Numerical and R3 counts come from the linked repository evidence and may postdate those binaries. A local numerical report without matching CI provenance is labeled accordingly.