amadan
Evidence/design candidate is 55e98af, clean and pushed. Full local gate and actual remote CI now fail only the two intentional readiness regressions: successful two-worksheet import retains one tab and HTTP export one worksheet; unsupported cross-tab formula on a later worksheet reports success and publishes a partial workbook. First-tab B1 refusal control passes. Vet, format, build, migrations and all other package tests pass; race stage is not reached after expected test failure. Fixtures are under root testdata/workbook-regressions; independent ZIP/XML verification confirms their names/order, distinct values and formula placement. Plan is LLM-only and unapproved at docs/superpowers/plans/2026-09-07-workbook-import-fidelity.md. No production implementation or other branch edits. This follows the documented first-worksheet deferral and depends on reviewed/reachable tabs; it does not claim the current form promises whole-workbook fidelity. Review prompt/evidence: /tmp/oficina-sheets-workbook-review. Candidate is deliberately not mergeable/deployable.
Workbook import completeness and cross-reference refusal evidence open
On branch test/xlsx-workbook-fidelity
Independent adjacent import investigation from origin/main4aea8f4. Frozen workbook-tabs #18 and reconciliation candidates remain untouched. Real runImport tests with synthetic two-worksheet XLSX reproduce successful publication of only one tab and one-tab HTTP export. Unsupported cross-worksheet formula in the first tab is correctly refused; the same kind of formula in a later tab is ignored, and the partial workbook is published. Preparing red readiness regressions and an LLM-only import contract dependent on reviewed/reachable tabs. No production implementation, new UI language, Fable call or deployment; whole-workbook export remains owned by #18.
Root queued a combined Fable DESIGN review with distinct verdicts for tabs/export d8fc99f80e52d7d916e89262934c8eacecac057f and adjacent import55e98af865786d8792fb32100ecaafbc5c75d279. Actual reconciliation prerequisitea75c7a9 remains separately reviewed, with its copy gates open. No prior tabs review was found; this is a fresh reviewer session, not transferred reconciliation approval.
Before queueing, root checked all three clean exact worktrees and32 referenced source/evidence file hashes. The active01:51 Dublin timer repeats those checks before model invocation. Ten review jobs total; runner remains read-only and never merges/deploys. Full gate at import candidate is intentionally red in two readiness regressions, so it is not a release candidate. Cross-tab calculation remains a separate unfinished goal. No copy or canary gate waived.
Fable completed joint review: both tabs draftA d8fc99f and import draftB55e98af require bounded design corrections. Full verdict remains at /tmp/oficina-sheets-tabs-import-review (canonical output path is recorded in /tmp/oficina-fable-reset/queue.json). Additional root synthetic dependency probe confirms a custody gap in B's no-plaintext-staging promise: exact Excelizev2.11.0 with the app's256MiB/64MiB limits opens a72003-byte XLSX and spills68157934 bytes of worksheet XML to a temporary file, including an actual inline-string cell marker. Close removes it, but plaintext exists while parsing. Evidence /tmp/oficina-sheets-design-audit/main.go and temp-spill-proof.json; synthetic data only, no live calls. B needs an explicit no-spill/resource contract and a regression, not merely reuse of current limits. This is an additional review finding, not a claimed production exploit or completed fix.