QA run · platform-core.seed-integrity · a86db175 · 2026-08-10 · PASS

[baozhoutao]

Selector platform-core.seed-integrity → 1 item (surface api, revision 1) · app showcase
Result: ✅ PASS (5/5 clauses)

First real end-to-end run of the checklist-test skill (introduced in #7304), recorded per the one-issue-per-run model — results stay out of the repo tree. Convention label for these is qa-run (not yet created in this repo; add bug when a run fails).

Environment

framework	a86db175 — branch claude/platform-test-checklist-ocwugl (PR #7304)
app · driver	showcase · SqlDriver(better-sqlite3)
db	file:/tmp/qa-seed-run1/data.db (isolated, fresh boot → restart against same DB)
port	3467 · seed --seed-admin (admin@objectos.ai)
when	2026-08-10T06:39Z

Per-clause verdicts

#	clause	verdict	oracle evidence (text)
1	every seeded object's row count equals the authored seed count	✅ pass	18/18 objects: API total == re-derived seed count; 0 mismatches (table below)
2	seed VALUES land verbatim	✅ pass	Specimen — Full: f_multiselect ["red","green"]; f_json {"nested":{"k":"v"},"list":[1,2,3]}; f_lookup "Northwind" → resolved to real account id wIetZ1J9twPOdROY (exact match)
3	boot log reports every seed load as success	✅ pass	[Seeder] Seed loading complete {"inserted":130,"updated":0,"skipped":0,"errored":0} — no rejection line
4	seed replay is idempotent	✅ pass	restart same DB: {"inserted":0,"updated":122,"skipped":8,"errored":0}; count sweep 1 ≡ sweep 2 (identical)
5	deliberately-unseeded fields stay unseeded	✅ pass	f_secret / f_user / f_users / task.cover all null — recorded expected-absent, not findings

Negative traps ruled out: no nav-visible object at count 0; no count grew on restart.

Count table (expected == actual, first sweep ≡ restart sweep)

showcase_account 14 == 14 showcase_invoice 12 == 12
showcase_announcement 2 == 2 showcase_invoice_line 5 == 5
showcase_business_unit 9 == 9 showcase_preference 1 == 1
showcase_category 4 == 4 showcase_product 4 == 4
showcase_contact 32 == 32 showcase_project 5 == 5
showcase_expense_line 13 == 13 showcase_project_membership 3 == 3
showcase_expense_report 4 == 4 showcase_task 10 == 10
showcase_field_zoo 2 == 2 showcase_team 2 == 2
showcase_inquiry 3 == 3 sys_business_unit 5 == 5

18 objects · 130 rows total.

Note — checklist drift (NOT a platform defect)

The item's authored baseline reads showcase_contact = 33 (9 named + 24 bulk prospects), but the current seed module (examples/app-showcase/src/data/seed/index.ts) declares 32, and the server serves 32. Re-derived per the item's own step-1 rule ("re-derive rather than trusting a stale list if seeds changed"), so the run used 32 as expected and passed. → refresh the item baseline 33 → 32 on the next checklist-test item revision. No platform action.

Screenshots

None. Every clause here is an api/log oracle (server truth); per RUNNER.md the report is text-only — screenshots are live judgment aids that never enter the record. A reproduction rule is included only for failures; this run had none.