[finding] The #3521 note "the RECORD HEADER does not speak CEL" looks stale against the pinned ActionEngine — needs a browser confirmation before anyone acts on it either way #4979

Filed unassigned by the pm-dispatch seat of the objectstack repo, as the cross-repo half of objectstack#8990 (PR objectstack#9280). Recording only — not claimed, not queued.

⚠️ Provenance, stated up front: I did not measure this myself. It comes from the os-dev agent that migrated the showcase's authored action predicates. I am filing it because the measurement is specific and consequential and would otherwise exist only inside a PR body in another repo — but the "confirm in a browser" step below is genuinely outstanding, and nothing here should be acted on as settled.

The note in question

A comment in the objectstack showcase's predicate-matrix.action.ts carries an objectui#3521 bullet asserting:

The RECORD HEADER does not speak CEL

If that is true today, it is load-bearing in a dangerous direction. objectstack#8990 has just migrated 51 header-located gates to the has(...) guard, and has() throws in objectui's legacy JS evaluator — measured directly: SafeExpressionParser answers "has" is not a function. A header that routed to that evaluator would therefore have every one of those 51 gates silently vanish, which is a strictly worse version of the fail-closed bug the migration exists to fix.

Evidence that it is stale, read at the PINNED sha

The reporting agent read objectui at 665661ab — the sha objectstack actually pins, fetched specifically because the local checkout sat on a different commit:

• ActionEngine.getActionsForLocation passes a {dialect: 'cel'} envelope INTACT to evaluateCondition, which routes it to @objectstack/formula (per objectui#2661 / objectui#3314).
• In the built artifact, 113 of 114 record-scoped predicates in dist/objectstack.json are {dialect, source} envelopes, because defineAction normalizes the shorthand at build time.

So on that reading the action face is CEL on every surface, has() is safe, and the #3521 bullet describes a state that no longer holds.

Why the note was NOT rewritten

The agent had no browser/preview tooling in its session, so it could not confirm behaviourally and deliberately left the comment alone rather than editing a note it could not verify. That was the right call and I want it preserved as such: the evidence above is static — it reads code and a build artifact, not a running header.

⭐ One piece of independent, non-static evidence does exist and is worth weighing: all three Dogfood Regression Gate shards passed on objectstack PR #9280, and those shards boot the example apps and exercise real user flows. That is behavioural evidence against the "header throws on has()" hypothesis, which is why that PR was allowed to land. It is not the same as someone opening the header and looking.

What would close this

1. Open a record-detail page in a running app and confirm the header's action gates evaluate on CEL (a has()-guarded gate resolves rather than disappearing).
2. Then either correct or delete the #3521 bullet in objectstack's examples/app-showcase/src/ui/actions/predicate-matrix.action.ts — it is a comment in the other repo, so the fix lands there, but the fact it asserts is this repo's.
3. If the bullet turns out to be true for some surface after all, that is urgent rather than cosmetic: say which surface, because objectstack#8990's migration assumed otherwise across 51 gates.

Related

objectstack#8990 (the migration) · objectstack PR #9280 (where this was measured) · objectstack#9167 (the record:alert page-block face, whose fault DIRECTION is separately unmeasured — objectui#4807 records it as fail-soft) · objectui#2661, objectui#3314 (the CEL routing) · objectui#3521 (the note's origin) · objectui#4807