Re-scope the contradiction scanner and schedule it

[markfreeman-dot-dev]

v2 migration #100 › M7-lifecycle (epic #15) · Size S · Architecture

Goal

Keep the scanner checks v2 needs, run them over petri.sqlite on optional cron, keep petri scan on-demand with exit codes intact.

Context

This issue owns the scanner rework; M4-dbos only disables the queue-dependent categories and points here. Most of the v1 scanner's 10 categories die in v2 (queue-schema checks with M4-dbos, verdict-drift checks with M2-agents); what survives is domain-data integrity over petri.sqlite.

Scope

In:

• Category inventory naming keep/retire and the obsoleting mechanism (M2-agents typed verdicts, M4-dbos queue deletion), committed as a docs note.
• auto_fix and scan_loop decision (scanner.py:152-187): manual-only behind an explicit flag, or retired; scheduled runs never auto-fix.
• Surviving checks rewritten in petri/analysis/scanner.py over petri/query/; ScanIssue replaced with a Pydantic model.
• DBOS.create_schedule in the v2 app-bootstrap module, cron from a petri.yaml key (croniter); petri scan runs the same workflow synchronously, exit-1-on-findings.
• scan_completed events via the domain-event seam (payload in EVENT_DATA_MODELS); root span petri.* (dish id, findings count, trigger source). Tests: test_scanner_v2.py, test_scan_schedule.py.

Out:

• No scanning of generated adapter config beyond what v2 still generates (D5).
• No dashboard scan-findings UI beyond the events.

Acceptance criteria

• All 10 v1 categories accounted for: each rewritten for v2 or mapped to the mechanism that makes it impossible, with a one-line rationale; the auto_fix/scan_loop decision is recorded.
• A seeded phantom edge (edge to a nonexistent cell id) is detected in unit tests.
• A malformed event row (data payload failing its registered model under envelope {id, cell_id, timestamp, type, agent, iteration, data}) is detected in unit tests.
• With config key absent, no schedule registers; with a test cron, a scan fires within the poll window appending exactly one scan_completed event per firing, no duplicates across a mid-interval restart; scheduled runs never invoke auto_fix.
• petri scan reports the same findings as the scheduled run on identical state, and exits 1 when findings exist and 0 otherwise.
• Both paths emit a root span with petri.* attributes (dish id, findings count, trigger source).

Pointers

• petri/analysis/scanner.py:23-30,66-149,152-187,339,364,387,435 — hierarchy, dispatch, retiring checks
• petri/cli/scan.py; petri/models.py:199-226 EVENT_DATA_MODELS
• petri/query/ (M6 read layer); petri/storage/edge_registry.py (edges table)
• v2 app-bootstrap — register DBOS.create_schedule before launch; never @DBOS.scheduled; M2-agents/M4-dbos obsoleting mechanisms

---

#N in this issue refers to entries in docs/field-reports.md, not issues in this repo.
To claim this issue, leave a comment. Questions welcome — see CONTRIBUTING.md.