chore(tiles): stop exporting the frozen incident bundles

[yorickdewid]

Follows #75, which turned tile generation off but deliberately left enabled/upload_dataset on so the GPKG exports kept feeding the static-model history. For the incident layers that history has nothing left to record.

        tileset        | enabled | generate_tileset | upload_dataset
-----------------------+---------+------------------+----------------
 analysis_full         | t       | f                | t
 facade_scan           | t       | f                | t
 incident              | f       | f                | t
 incident_district     | f       | f                | t
 incident_municipality | f       | f                | t
 incident_neighborhood | f       | f                | t

Why these four

• report.incident is frozen: 2,756 rows, newest create_date 2026-03-03 — 157 days ago. Intake stopped when the KCAF loket and municipal portals moved off our infra, so every nightly export since March has been a byte-identical copy.
• The three aggregations are pure derivations of report.incident joined to the geocoder areas — reconstructable from the incident export, and from the table itself.
• ~700 daily snapshots already cover the entire period incidents were being filed.

enabled=false makes processOne take the both-flags-off early return from #53 — no ogr2ogr dump, no queue job, no log line. Verified after applying that Martin still serves all four layers (200 on incident, incident_district, incident_municipality, incident_neighborhood).

Why facade_scan is NOT included

• Live data — new QuickScan samples in Aug 2026 (178 in April, 37 in June, 2 in August). Not frozen.
• Not in analysis_full — I checked the column list: 33 columns of risk model, none of skewed_parallel_facade, skewed_perpendicular_facade, facade_type, settlement_speed, facade_scan_risk, risk, priority. The export is the only archived record of them.
• It faces a planned wipe — the delete-and-reimport of all 5,200+ QuickScan records would destroy every trace of the current state. The archive is the only place it would survive.

Costs 1.2 MB and ~17s a night. analysis_full stays enabled as the static-model history itself.

Result

Nightly process_mapset is now analysis_full (~9.5 min) + facade_scan (~17s), and nothing else.

Applied to prod 2026-08-07.

🤖 Generated with Claude Code