Follow-up filed by the domain:services seat (#6021) on closing #7225 as a false alarm. domain:* left to the triage seat (expected: domain:services, matching the example-corpus precedent of #7036/#7225).
Why
#7225 concluded that three shipped example hooks were silent no-ops; measurement proved all three work (the declarative flat-input proxy, packages/objectql/src/hook-wrappers.ts:446) and that the "fix" would have broken the showcase's public web-to-lead insert. The wrong call was cheap to make because nothing pins these hooks' behaviour — a careful reader deriving from three source files plus the (incomplete, see #7254) contract table gets it wrong, and the examples are the reference corpus AI authors copy from. One full dispatch was already spent on the false alarm; the pin stops the next one.
What
Land the probes the #7225 dev already wrote and measured green, as permanent regression tests:
examples/app-crm: real kernel + real driver, insert/update opportunity through the app's real object + registered OpportunityStageHook, assert the persisted row (closed_won → probability 100, closed_lost → 0, non-closed control untouched, handler sees flat input.stage bound). Costs two devDeps on app-crm (@objectstack/objectql, @objectstack/driver-sql) — flagged, deliberate.examples/app-showcase: same shape through the real QuickJS body runner (hookBodyRunnerFactory), assert persisted inquiry status=new/source=web and trimmed task title; keep the in-sandbox shape probe (ctx.input is the flat record, hasData=undefined) as the documented-surface witness.
Test-only; no src/ behaviour change in either app. The probe transcripts and wiring are described in the #7225 close-out comment and the dev report quoted there.
Refs: #7225 (falsified card + probes), #7254 (the contract-table documentation gap, separate card), #7036 / PR #7222 (app-todo twin where the hook was genuinely unregistered).
Follow-up filed by the
domain:servicesseat (#6021) on closing #7225 as a false alarm.domain:*left to the triage seat (expected:domain:services, matching the example-corpus precedent of #7036/#7225).Why
#7225 concluded that three shipped example hooks were silent no-ops; measurement proved all three work (the declarative flat-input proxy,
packages/objectql/src/hook-wrappers.ts:446) and that the "fix" would have broken the showcase's public web-to-lead insert. The wrong call was cheap to make because nothing pins these hooks' behaviour — a careful reader deriving from three source files plus the (incomplete, see #7254) contract table gets it wrong, and the examples are the reference corpus AI authors copy from. One full dispatch was already spent on the false alarm; the pin stops the next one.What
Land the probes the #7225 dev already wrote and measured green, as permanent regression tests:
examples/app-crm: real kernel + real driver, insert/updateopportunitythrough the app's real object + registeredOpportunityStageHook, assert the persisted row (closed_won → probability 100, closed_lost → 0, non-closed control untouched, handler sees flatinput.stagebound). Costs two devDeps on app-crm (@objectstack/objectql,@objectstack/driver-sql) — flagged, deliberate.examples/app-showcase: same shape through the real QuickJS body runner (hookBodyRunnerFactory), assert persisted inquirystatus=new/source=weband trimmed task title; keep the in-sandbox shape probe (ctx.inputis the flat record,hasData=undefined) as the documented-surface witness.Test-only; no
src/behaviour change in either app. The probe transcripts and wiring are described in the #7225 close-out comment and the dev report quoted there.Refs: #7225 (falsified card + probes), #7254 (the contract-table documentation gap, separate card), #7036 / PR #7222 (app-todo twin where the hook was genuinely unregistered).