Idempotent post-migrate that moves mid-flight in_progress jobs whose
recipe steps are all terminal into the appropriate new state:
- draft cert exists → awaiting_cert
- no cert required → awaiting_ship
done jobs left alone (historically completed, already shipped).
Card_state + mini_timeline_json recomputed for affected rows so the
plant kanban renders correctly on first page load.
Version bump 19.0.10.31.0 → 19.0.11.0.0 triggers the migration on -u.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two follow-up fixes caught during the entech deploy of recipe cleanup:
1. RESOLVER_KIND_TO_ACTIVE_KIND was missing a self-pass entry for
'wet_process'. The new aliases added in 19.0.21.3.0 (Chemical
Conversion, Trivalent Chromate Conversion, Strip Process - AL,
Plug The Threaded Holes via mask) directly return 'wet_process'
from the resolver — without the passthrough they didn't translate
to any active kind and stayed as 'other'. Added 'wet_process':
'wet_process' so the migration's Phase 2 backfill catches them.
2. Migration 19.0.10.26.0 Phase 3 was using batch unlink
(clone_recipes.unlink()) which tripped a PostgreSQL FK cascade
ordering bug on entech ("insert or update on parent_id violates
FK ..." during the CASCADE chain). Rewrote Phase 3 to delete one
clone at a time with SAVEPOINT per clone — slower but immune to
the batching bug, and one failed clone doesn't roll back the
whole transaction.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Migration 19.0.10.26.0/post-migrate.py runs in 5 phases:
1. Resequence recipe 3620 ENP-ALUM-BASIC ops to fix the duplicate-
sequence bug (Contract Review=10, Incoming Inspection=20,
Masking=30, Racking=40, then the rest). Also delete the empty
duplicate ENP-Alum Line sub_process (id 4056).
2. Backfill kind on all kind=other nodes via the extended resolver
from fusion_plating 19.0.21.3.0
3. Delete all per-part clone recipes (name contains em-dash)
4. Recompute fp.job.step.area_kind on all steps
5. Recompute fp.job.active_step_id + card_state on in-flight jobs
Plant kanban: no_parts cards now always land in the Receiving column
regardless of active_step area_kind. The receiver works Receiving;
that's where the card belongs when parts haven't arrived.
Spec: docs/superpowers/specs/2026-05-24-recipe-cleanup-design.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- fp.step.kind.area_kind: drop tracking=True (model doesn't inherit
mail.thread; tracking was a no-op emitting a startup warning).
- Migration 19.0.10.25.0: anchor the De-Masking ILIKE so it doesn't
wildcard-match "Ready For De-Masking" (which the earlier "Ready %"
rule already routes to gating). Also drop the cur_code='mask' filter
so the 4 De-Masking nodes still classified as 'other' get picked up
on fresh re-runs too.
Direct SQL applied the 4-row fix on entech (post-migrate doesn't
re-run for already-applied versions); this commit keeps fresh
installs and any future re-runs consistent.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Fix the Shop Floor plant kanban so cards land in the right column:
- fp.job._compute_active_step_id walks priority chain
(in_progress > paused > ready > pending), not just in_progress
- fp.job._compute_card_state edge case respects job.state='done'
(no more bogus 'contract_review' label on done jobs)
- fp.job.step._compute_area_kind reads kind.area_kind directly;
legacy _STEP_KIND_TO_AREA dict removed (50+ lines deleted)
- /fp/landing/plant_kanban filters out done/cancelled jobs from
the live board
Migration 19.0.10.25.0 backfills template metadata (codes,
descriptions, icons, kind_id) on 30 unfinished library templates
and repoints recipe nodes for 6 unambiguous name patterns
(Blasting -> blast, Ready For X -> gating, De-Masking -> demask,
Scheduling -> gating, Nickel Strip -> wet_process,
Pre-Meas/Check Sulfamate -> inspect).
Battle test bt_s24_between_steps.py covers between-step routing,
paused step lifecycle, and done-job board filter.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
