Problem
The source-semantic-hardening change was validated on a 12-table dev slice, but the proposal target is the full 33-table cBioPortal corpus. Full-corpus sign-off — plus the holdout-vs-dev-slice bias check in tasks.md §8.8 — is still pending.
Currently blocked: only 12 of 33 tables are landed in Databricks. The remaining 21 tables require the cbioportal-omop-data-bridge runbook to be completed before evaluation can proceed.
Proposed solution
Once ingest is unblocked:
- Complete the
cbioportal-omop-data-bridge runbook to land all 33 tables.
- Run the full build +
sema eval telemetry dump on the 33-table corpus.
- Run the holdout-vs-dev-slice bias comparison (tasks.md §8.8) to confirm few-shot library generalizes.
- Sign off OpenSpec tasks §10.1 and §10.4.
Alternatives considered
- Declaring 12-table dev-slice results sufficient — rejected; the proposal target is the full corpus and we want the holdout check before archiving the change.
Not closed by #63 — tracks post-merge follow-up work.
Problem
The
source-semantic-hardeningchange was validated on a 12-table dev slice, but the proposal target is the full 33-table cBioPortal corpus. Full-corpus sign-off — plus the holdout-vs-dev-slice bias check in tasks.md §8.8 — is still pending.Currently blocked: only 12 of 33 tables are landed in Databricks. The remaining 21 tables require the
cbioportal-omop-data-bridgerunbook to be completed before evaluation can proceed.Proposed solution
Once ingest is unblocked:
cbioportal-omop-data-bridgerunbook to land all 33 tables.sema evaltelemetry dump on the 33-table corpus.Alternatives considered
Not closed by #63 — tracks post-merge follow-up work.