objectql test driver doubles are `any`-annotated at scale — the retired-capability tombstone has no tsc channel in 64 files

[os-zhuang]

Found while fixing #6546 (PR #7261). Recorded, not fixed there — #6546's file surface is
packages/objectql/src/protocol-batch-atomic.test.ts only, and folding this in would have turned an
XS card into a 64-file sweep.

Observation

#6546 was one driver double authoring a capability key retired by #4634. The deletion was trivial;
the interesting half was why nobody noticed — the literal was annotated const driver: any, so it
was never compared against IDataDriver, and nothing in that file parses it through
DriverInterfaceSchema. With both channels off, the retiredKey(...) tombstone in
DriverCapabilitiesSchema had nothing to fire through.

That annotation is not unusual in this package. Measured on origin/main @ 0caf122:

	files
packages/objectql/src/**/*.test.ts with a driver-ish double annotated : any =	64
…of which also spell supports: {} as any	17
test files that reference IDataDriver at all	14

What this is NOT

No retired bit is authored in any of them today, and no gate is red. A repo-wide sweep for
supports: literals spelling retired or invented keys (run as part of #6546) found nothing beyond
#6546's own instance, #4782's, and the two deliberate rejection fixtures
(packages/spec/src/data/driver.test.ts:317, packages/spec/src/data/datasource.test.ts:60), which
are correct as written. Every other literal is either {} or a live bit
(autonumber, batchSchemaSync, queryDateGranularity).

So this is observation-class, exactly like #6546: nothing a user hits, nothing broken.

Why it is still worth a line

The tombstone's stated design is that a retired key is authored-unwritable — "tsc rejects it at
the authoring site" is the first of the two channels driver.zod.ts names for it. Across these 64
files that channel is switched off, so the tombstone's protection currently rests entirely on the
second channel (a parse), which test doubles by definition rarely go through.

The concrete harm is the one #4782 and #6546 both name: mocks get copied. #6546's mock had
carried its retired bit since #4634 without anyone noticing, and it was found by reading, not by a
gate. The cost of the class is that the next such copy is equally invisible.

Measurement from #6546, for sizing

Retyping one double any → IDataDriver cost three stubs for interface members the tests never
reach (upsert, syncSchema, dropTable), and moved @objectstack/objectql's measured
TEST_DEBT count by zero (340 → 340, ceiling 355). One data point, not a trend — that double
was already close to interface-shaped, and the as any in 17 of the files above suggests some are
further away. Anyone picking this up should re-measure per file rather than extrapolate from one.

Note also that a naive sweep will trip TS2790 wherever a test does delete driver.<requiredMember>
to prove a method-presence gate; Reflect.deleteProperty(...) is the cast-free equivalent (#6546
used it).

Suggestion

Not obviously a single sweep. Plausible shapes, in increasing cost:

1. Leave it, and retype opportunistically whenever a file is touched for another reason.
2. Retype only the doubles that author a non-empty supports — the ones where a capability key
   could actually be spelled wrong — which is a much smaller set than 64.
3. A gate that fails a driver double annotated any, ratcheted against a measured baseline in the
   style of check-engine-double-contract.

Filing without a recommendation; the sizing above is what I measured, not a plan.

[claude]

Triage: finding appended (domain:engine-core was applied at filing — kept, the face is packages/objectql/src/**/*.test.ts doubles).

• Grade rationale: observation-class by the card's own evidence — no retired bit is authored in any of the 64 files today, no gate is red, and the sweep run as part of objectql/protocol-batch-atomic.test.ts mock driver advertises the retired supports.transactions bit — invisible because the mock is : any #6546 found nothing beyond the already-handled instances. Nothing a user hits; the harm channel is copy-a-mock. That is the finding shape, not pm:queue.
• Premise spot-check @ origin/main 1da1f32: 96 objectql test files carry : any = annotations (card measured 64 driver-ish doubles at 0caf122 — direction confirmed, my broader count is a superset).
• Dedup: objectql/protocol-batch-atomic.test.ts mock driver advertises the retired supports.transactions bit — invisible because the mock is : any #6546 (the instance that spawned this card) is in flight and fixes one file only (PR test(objectql): type the batch-atomic driver double as IDataDriver, dropping the retired supports.transactions bit #7261, accepted 04:20Z, retypes that one double to IDataDriver); this card is the corpus-wide question — distinct scope, no overlap. PR test(objectql): type the batch-atomic driver double as IDataDriver, dropping the retired supports.transactions bit #7261 also demonstrates the fix shape a future sweep would replicate.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code

[claude]

Findings triage (re-verify @ origin/main c6a4eeb): HOLD — the tsc channel is still off at scale (a broader count of : any = in packages/objectql/src/*.test.ts now hits 101 files, direction unchanged from the filed 64), no retired bit authored anywhere, no gate red. Observation-class stands; a 64-file retyping sweep is real cost with no live defect behind it today. finding stands.

本评论来自分诊座位 Routine（#5474 试点），不构成认领。

---

Generated by Claude Code