test(core): pin the shared validation corpus to every envelope value domain - #2315
Merged
Conversation
…domain Pinning the corpus to the envelope key list closed one half of the drift that made an 89-cell benchmark run export a one-line summary per cell: a key nobody wrote a case for. The other half was still open — a value nobody wrote a case for. The Harbor exporter spells `origin`, `modelVisibility` and `status` out again in Python, so a member added on this side that no corpus case carries leaves the two disagreeing about what is valid with nothing red. `status` already derived its type from a runtime constant; `origin` and `modelVisibility` were inline unions the decoder compared against by hand, which made their domains unenumerable — a chain of `!==` is invisible to anything asking what a field may hold. Give them the same shape as `status`, have the decoder read the constant through a shared `isOptionalMember`, and hold the corpus to what those constants contain. The corpus gains the four `status` members it never carried.
`role` and `author` are closed envelope domains that the Python exporter spells out too, so leaving them out of the map made the invariant false where it was stated: three of five domains were pinned. The holes were real, not hypothetical. The corpus never carried `role` or `author` as `tool` or `system`, and events with `role: tool` are among the most common a run emits — narrowing the exporter's role set to drop it degrades a real event stream to a summary, and nothing was red.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
#2314 fixed a drift that made every one of an 89-cell benchmark run's Maka trajectories export as a one-line summary: the Harbor exporter re-implements the RuntimeEvent envelope check in Python and had never learned
originormodelVisibility. It closed that class of drift at the key level — the shared validation corpus is now held toruntimeEventEnvelopeKeys(), so a key added to the interface cannot ship without a corpus case, and a corpus case the Python side rejects turns the exporter's silence into a red test.The same drift is still open one level down. Python spells the closed value domains out again (
{"provider", "code_mode"},{"visible", "hidden"}, the status set), so a member added on the TypeScript side that no corpus case carries leaves the two disagreeing about what is valid with nothing red — and every event carrying that member degrades to a summary exactly as an unknown key did.Two changes, and the first is what makes the second possible:
statusalready derived its type fromRUNTIME_EVENT_STATUSES.originandmodelVisibilitywere inline unions that the decoder compared against by hand, which left their domains unenumerable: a chain of!==is invisible to anything asking what a field may hold. They now have the same shape asstatus, and the decoder reads the constants through a sharedisOptionalMemberinrecord-schema.tsrather than repeating their members.runtimeEventEnvelopeValueDomains()exports those three domains, and a new test requires an accepted corpus case for every member.branchis free text and has nothing to enumerate, so it does not appear.The corpus gains the four
statusmembers it never carried (streaming,failed,aborted,cancelled) — onlycompletedwas ever exercised.Refs #2314.
Verification
Both halves of the chain proved by reverting, not by reading:
'harness'toRUNTIME_EVENT_ORIGINSwithout touching the corpus →no corpus case carries origin: harness.modelVisibilityset to{"hidden"}→ the headless corpus round-trip fails onvalid-visible-model-visibilitywithruntime_event_schema_invalid— the exact degradation fix(headless): teach the trajectory exporter the two envelope fields it never learned #2314 was about, now caught by a test instead of by a benchmark run.@maka/core799 pass / 0 fail,@maka/headless1431 pass / 0 fail (includes the Python corpus round-trip),@maka/runtime3293 pass / 0 fail.npm run lintandnpm run formatclean.Not run: desktop E2E and Storybook — this touches no renderer surface.
Review focus
The value-domain test is only as strong as the corpus round-trip it feeds: it guarantees a case exists, and the Python side is what turns that case into a cross-language assertion. It does not reach nested shapes (
content,actions,refs), whose domains are still pinned only by the cases someone thought to write.