# BAT-017 Planning: Import provenance and sensitive-data policy

- Status: draft
- Created At (KST): 2026-07-08 11:33 KST
- Included REQ: REQ-021

## Planning Summary

BAT-017 is a standard batch because REQ-021 affects data persistence policy, import adapters, MCP/Streamlit response contracts, tests, and baseline documentation.

## Work Slices

1. Define provenance policy helpers
   - masked basename helper
   - source hash propagation helper
   - raw payload minimization/filter helper
   - exposure-level sanitizer for `minimal`, `operator-summary`, `debug-gated`
2. Apply policy to import write paths
   - KB Bank PDF
   - Toss Bank PDF
   - Toss Securities PDF
   - SeoulPay XLSX
   - CSV preview/import boundary
3. Apply policy to read/response surfaces
   - import batch listing
   - preview responses
   - MCP tool contracts/default output
   - Streamlit/operator summary where applicable
4. Add regression checks
   - source-specific import tests
   - forbidden-field filtering tests
   - default MCP response privacy checks
   - fixture/log/static scan checks
5. Update baseline documents
   - `docs/data-model.md`
   - `docs/project-structure.md`
   - `docs/runtime-flows.md`

## Design Gate

Design is required. The batch changes persistence semantics and interface response boundaries, so `design.md` must be completed before implementation.

## Risks

- Over-sanitizing may weaken duplicate/rollback/audit traceability.
- Under-sanitizing may leak file paths, raw financial evidence, or arbitrary CSV columns.
- Existing operating DB rows may contain legacy data; this batch must avoid destructive cleanup and keep the follow-up Todo separate.

## Verification Plan

- `python -m ruff check .`
- source-specific pytest for import provenance behavior
- full `python -m pytest -q`
- MCP contract/default privacy checks
- dashboard/AppTest/HTTP smoke if affected
- DB safety smoke if schema/migration changes are included
- StagePilot doctor with current-batch findings separated from pre-existing debt