Pin the authoring guide's worked op-scripts against the release binary #344
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Debt class named by the 2026-07-26 harvest audit (architect drift review):
docs/authoring-guide.mdnow carries several worked op-script examples (the tap example, and the harvest-added LinComb and ConstantCost→CostSum wiring examples), each verified by hand at authoring time by piping it throughaura graph build— but none is pinned by a test. A vocabulary or schema change would let them rot silently: the guide would keep teaching op-scripts the binary refuses.Ask: a test that extracts the guide's fenced op-script JSON blocks and drives each through
graph build(and, where the example claims specificintrospect --paramsoutput, checks that too), so a rotting example fails the suite instead of the next reader.context: existing debt class, grown by the harvest's two new examples; no current misbehaviour — the examples build today (verified in-cycle).
First concrete instance (harvest-2026-07-26 fieldtest, B1)
The rot class this issue names has its first observed case:
docs/authoring-guide.md~800-821 transcribed multi-regime family ids in a shape the executor does not produce (suffix-r1with the default regime unmarked; reality is an infix-rKafter the window segment with-r0present even for an implicit default regime). Transcribed output — not just op-script inputs — rots too, so the pin should cover asserted output shapes (family-id/trace-handle lines) where cheap, not only that the examples build. The instance itself was fixed in-cycle.