feat(fusion_plating): partial order handling on the shop floor

Operators can now see and advance a job's parts across multiple stages
at once (e.g. 10 Masking / 20 Plating / 20 Baking on one 50-part job).
Tracking model C (fluid per-stage quantities + existing hold/scrap/
rework records for exceptions); board option 2 (a card per occupied
stage); wait-to-reconverge close. Additive only — no new model, no
migration, no change to the close/cert/ship lifecycle.

Board (fusion_plating_shopfloor/controllers/plant_kanban.py):
- One card PER (job, stage), composite key "{job_id}:{area}". Unsplit
  jobs render exactly as before. _job_presences/_render_presence;
  primary presence keeps full job card_state, secondary presences
  derive state from their focus step.

Card (plant_card.js/.xml/.scss):
- "20 of 50 here" badge; tap opens the workspace focused on that
  stage's step (focus_step_id, already accepted by the workspace).

Move + light-up (move_controller.py, fusion_plating_jobs/fp_job_step.py):
- Availability/pre-fill now from qty_at_step (step had no qty_done/
  qty_scrapped fields — the old read was always 0, dead path).
- Forward move auto-flips destination pending->ready (no auto-start;
  labour timer stays explicit) and auto-finishes a drained source
  (best-effort). Predecessor gate is qty-aware: a step with real
  arrived parts is startable regardless of upstream completion
  (_fp_has_real_incoming, single source of truth for can_start /
  blocker / button_start / move blockers).

Operator advance (job_workspace.js):
- "Send -> <next>" action on in_progress/paused steps opens the slimmed
  Move dialog (qty steppers, no keyboard; advanced fields collapsed).
  Was only wired into the deprecated shopfloor_tablet before.

Close (fp_job.py):
- button_mark_done counts move-based scrap (_fp_scrapped_via_moves) into
  qty_scrapped and derives qty_done = qty - scrapped (was blindly
  = job.qty, over-counting). Reconciliation gate unchanged.

Static-validated: pyflakes (py), lxml parse (xml), node --check (js).
Dynamic tests + browser check need an installed env (entech/trial) —
plating modules can't install on the local Community DB.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

fusion_plating/fusion_plating_jobs/models/fp_job_step.py

@@ -54,12 +54,37 @@ class FpJobStep(models.Model):
         # leak permissive behaviour through a related-field None.
         if not self.job_id:
             return True
+        # Partial-flow short-circuit (2026-06-02 partial-order handling).
+        # Once REAL parts have physically arrived at this step (a move
+        # parked them here), the predecessor lock is moot — the parts are
+        # on the floor at this station, so the step is startable
+        # regardless of whether upstream steps are fully done. This is
+        # what lets a partial group "light up" the next stage while the
+        # rest of the batch is still being processed upstream. Single
+        # source of truth: every caller (can_start, blocker, button_start,
+        # the Move dialog's _blockers_for_move) inherits this behaviour.
+        if self._fp_has_real_incoming():
+            return False
         recipe_seq = self.job_id.enforce_sequential
         if recipe_seq:
             return not self.parallel_start
         # Free-flow recipe — only the legacy per-step flag still gates.
         return bool(self.requires_predecessor_done)
 
+    def _fp_has_real_incoming(self):
+        """True when real parts have physically arrived at this step via
+        a move — an incoming move from a DIFFERENT step with qty_moved > 0.
+
+        Distinct from the qty_at_step first-step seed (a notional UI hint
+        with no backing move) and from self-loop measurement moves
+        (from_step == to_step, used by the Record Inputs wizard). Mirrors
+        the has_real_incoming test in core button_finish's qty gate.
+        """
+        self.ensure_one()
+        return bool(self.incoming_move_ids.filtered(
+            lambda m: m.from_step_id != self and (m.qty_moved or 0) > 0
+        ))
+
     def _fp_has_unfinished_predecessors(self):
         """True when an earlier-sequence step on the same job is not yet
         in a terminal state. Composes with _fp_should_block_predecessors
@@ -86,6 +111,10 @@ class FpJobStep(models.Model):
         'job_id.enforce_sequential',
         'job_id.step_ids.state',
         'job_id.step_ids.sequence',
+        # Partial-flow: arriving parts clear the predecessor gate
+        # (_fp_has_real_incoming), so can_start must recompute on move.
+        'incoming_move_ids.qty_moved',
+        'incoming_move_ids.from_step_id',
     )
     def _compute_can_start(self):
         for step in self:
@@ -217,6 +246,9 @@ class FpJobStep(models.Model):
         'state', 'sequence', 'parallel_start', 'requires_predecessor_done',
         'job_id.enforce_sequential',
         'job_id.step_ids.state', 'job_id.step_ids.sequence',
+        # Partial-flow: arriving parts clear the predecessor gate.
+        'incoming_move_ids.qty_moved',
+        'incoming_move_ids.from_step_id',
     )
     def _compute_blocker(self):
         for step in self:
@@ -652,6 +684,42 @@ class FpJobStep(models.Model):
         ).sorted('sequence')
         return candidates[:1] or self.env['fp.job.step']
 
+    def _fp_try_autofinish_on_drain(self):
+        """Best-effort auto-finish when a step has drained to zero parked
+        parts (2026-06-02 partial-order handling).
+
+        Called by the Move controller after a bulk move commits. When the
+        last parts leave an in_progress step it should close itself — one
+        fewer tap for the operator. But finishing runs the full gate chain
+        (required inputs, sign-off, contract review, receiving, and the
+        post-shop close gates on the last step). If any gate isn't
+        satisfied we must NOT fail the move that already succeeded — so we
+        swallow the UserError and leave the step in_progress for the
+        operator to finish manually (the board will show it "running, 0
+        here", which reads as "finish me").
+
+        Only fires for steps that had REAL incoming parts — never an
+        untouched first-step seed. Returns True if the step finished.
+        """
+        self.ensure_one()
+        if self.state != 'in_progress':
+            return False
+        if not self._fp_has_real_incoming():
+            return False
+        # qty_at_step is a non-stored compute off the move rows — force a
+        # re-read so we see the just-committed outgoing move.
+        self.invalidate_recordset(['qty_at_step'])
+        if self.qty_at_step != 0:
+            return False
+        try:
+            self.button_finish()
+            return True
+        except UserError:
+            # Gates still pending (missing prompts / sign-off / etc.) —
+            # leave the step in_progress for a manual finish. The move
+            # itself stands.
+            return False
+
     def _fp_has_uncaptured_step_inputs(self):
         """True when the recipe step has REQUIRED step_input prompts
         whose values haven't been recorded yet.