0.4.1 — The Twin Grades Itself

2026-08-20 · additive · every 0.3.0 and 0.4.0 document stays valid

Seven of the nine vNext RFCs are now shipped. 0.4.1 delivers the honesty layer: the twin states its own trustworthiness, and the seam refuses to ship from a twin that isn't trustworthy enough.

RFC-05 — Credibility scorecard

Playback gains a scorecard: backtest window, calibration grade A–F, prediction-interval coverage, and a required credibility score. The seam gates on it — emits[].minCredibility fails the bundle when the twin's own score is below the bar. Twin claims become graded, not asserted.

RFC-06 — Seam version ranges + provenance

fluidVersion at the seam accepts semver ranges (^0.7.3) resolved against the vendored FLUID set — no more version cliff — and each emit may carry a provenance block: contractDigest is validator-enforced against the in-bundle contract, so contract drift after proving fails the build; outputDigest is reserved for engine attestation at the Runtime profile.

Numbers

Regression suite: 97 checks (was 85).
The example universe now exercises the scorecard, the credibility gate, a version range, and the provenance digest end to end.

Per the staged plan, one tranche remains: 0.5.0 — the RFC-04 runtime enforcement contract with conformance test vectors, and RFC-09 signed bundles — after which Level-3 Runtime is certifiable.

Full field-by-field diff: Changelog.