Context
The original strategy review raised dependency-stack risks:
- The local TOB repo venv could use an older
policyengine-us than the intended 6830-port/current stack.
policyengine-us-data projected datasets came from a feature branch and need validation against the active policyengine-us version.
- Batch requirements did not pin
policyengine-us, risking different local/batch/Modal versions.
Work
- Define the intended model stack for the CRFB rerun:
policyengine-us, policyengine-core, policyengine-us-data, branch/commit, and dataset snapshot.
- Validate H5 compatibility and variable availability across the intended 75-year scoring range.
- Pin or record versions for batch, Modal, local scripts, and saved-H5 scoring.
- Add a smoke test proving local and Modal use the same dependency stack.
Acceptance Criteria
- Every production result artifact has dependency-stack metadata.
- Local, batch, Modal, and saved-H5 paths cannot silently use different PE-US versions.
- Any remaining custom branch dependency is documented with merge/pin status.
Related: #84, #86, #98, #99.
Context
The original strategy review raised dependency-stack risks:
policyengine-usthan the intended 6830-port/current stack.policyengine-us-dataprojected datasets came from a feature branch and need validation against the activepolicyengine-usversion.policyengine-us, risking different local/batch/Modal versions.Work
policyengine-us,policyengine-core,policyengine-us-data, branch/commit, and dataset snapshot.Acceptance Criteria
Related: #84, #86, #98, #99.