beffroi-reporting/docs/migration-baseline.md

66 lines
3.3 KiB
Markdown

# First migration baseline
## Run
Requirements: Python 3, Git, Typst, Poppler's `pdftotext`, DejaVu Sans and DejaVu
Sans Mono fonts. Dependencies are checked, not installed automatically.
Choose a new output directory:
```bash
python3 scripts/check-legacy-baseline.py --output /tmp/beffroi-baseline-review
```
The runner reads an explicit source allowlist from legacy renderer commit
`1d381904fff3d378a7c0e04588e75310a9b7ed60`, stages it in a temporary directory,
and executes synthetic fixtures owned by the destination components. It never
materializes legacy real captures/reports or modifies reference working trees.
Temporary sources are removed after execution.
Outputs: five diagnostic PDFs, two example report PDFs, observed diagnostic JSON,
and a manifest of source revision, fixture hashes, Typst version, available
DejaVu fonts, and negative checks. Existing output directories are rejected.
PDFs are review artifacts, not committed content.
## Verification
- Five snapshots match reviewed expected diagnostic output: collection coverage,
findings, grouping, severity/confidence, and system facts.
- Independent assertions cover schema-1 service failure, schema-2 event counts and
grouping, incomplete-collection confidence, Compact unavailable checks,
unexpected collection status, and six watchlist candidate matches.
- Seven PDFs compile, have PDF signatures, and contain expected extracted titles.
- Missing author/date and an unknown theme fail with expected errors.
This is a legacy comparison harness, **not the new integration pipeline**.
The new diagnostic schema, renderer, adapter and full pipeline remain
unimplemented. The old regression entry point mixes synthetic tests with private
captures and is deliberately not executed by this harness.
## Observations and limits
Initial run: 2026-09-15, Linux, Typst `0.15.1 (586e1bd4)`. PowerShell is absent;
no native collection ran. Font versions and metadata affect PDF bytes, so byte
identity is not asserted. The fixture dates explicitly fix analysis reference time.
Schema-1 produces one service failure. Full/partial schema-2 preserves unknown
events and caps incomplete-collection confidence at 50. Compact unavailable checks
with zero counts remain unavailable. Its legacy display still shows a high numeric
score despite partial collection: this is observed behavior, not an accepted new
health contract. Generic synthetic reports still acquire the Beffroi logo.
First-page visual review covered the technical report and Compact diagnostic:
both were readable without visible clipping and the diagnostic showed its partial
collection warning. Other pages, long-content pagination and alternate branding
remain open. Synthetic checks do not verify historical content or advisory
accuracy. Linked vendor guidance has not been revalidated.
## Next pass
- [Diagnostic inventory and draft contract](../win-diagnostics/docs/migration-inventory.md)
- [Renderer inventory and draft field mapping](../reportist/docs/migration-inventory.md)
- [Private preservation and adapter decisions](../beffroi-reports/docs/migration-inventory.md)
Implement versioned schemas with valid/invalid synthetic examples, then extract
one offline diagnostic and generic rendering slice against this baseline. Finish
private incident evidence mapping before claiming historical migration is verified.