Two issues on the Process Tree client action:
1. Back to Work Order kept growing breadcrumbs (WO -> Tree -> WO ->
Tree -> ...) because onBack used action.doAction() which PUSHES
a new act_window onto the stack instead of popping. Fixed by
trying action.restore() first (pops the Tree off the stack and
returns to the parent WO/Step controller). Falls through to
explicit doAction only when there's no parent in the stack
(direct URL access).
2. The empty-state banner referenced productionId, a dead variable
from the bridge_mrp era when the tree was tied to mrp.production.
Since the component now uses jobId (fp.job context key), the
"No manufacturing order selected" message ALWAYS fired regardless
of whether a job was loaded. Fixed by using jobId and updating
the message to "No work order selected".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two compounding issues introduced by the persistence audit work:
1. fields.Text mismatch: sale.order.x_fc_internal_note and
x_fc_external_note are actually fields.Html. I declared the
sale.order.line related mirrors and the fp.job stored copies as
fields.Text, so setup_related raised:
TypeError: Type of related field
sale.order.line.x_fc_internal_note is inconsistent with
sale.order.x_fc_internal_note
Fixed by switching both Note fields on fp.job and sale.order.line
to fields.Html.
2. Module-load-order: Tier 3 fields (x_fc_delivery_method,
x_fc_ship_via, x_fc_invoice_strategy) are defined in
fusion_plating_jobs (related to sale.order via _inherit), but I
referenced them in fusion_plating core's fp_job_views.xml — which
loads BEFORE fusion_plating_jobs registers the fields. View
validator raised "Field x_fc_delivery_method does not exist".
Fixed by removing those 3 fields from the core view group and
adding them via xpath in fusion_plating_jobs's fp_job_form_inherit
(which loads after the fields are registered).
Both fixes deployed and verified — registry loads in 2s, all field
types match, related path resolves correctly. No data loss; the
fp.job rows that already had stored Text content for internal_note /
external_note will carry over into the Html field intact.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Tier 3 of the SO->fp.job persistence audit. Three logistics/billing
fields surface on fp.job as related read-only (not stored) mirrors:
- x_fc_delivery_method - Local Delivery / Shipping Partner / Customer
Pickup. Cargo classification used by logistics planning.
- x_fc_ship_via - Carrier name (UPS, FedEx, customer pickup, etc.).
- x_fc_invoice_strategy - Deposit / Progress / Net Terms / COD-Prepay.
Read by the invoicing module's hooks; mirroring on the WO is for
manager visibility only.
These were intentionally chosen as related (not stored persisted)
because the SO is the authoritative source - the existing downstream
code (delivery + invoicing modules) already reads them off SO directly.
A stored copy would risk drift. Related auto-follows SO updates.
Same three fields also mirrored on sale.order.line as stored related
for per-line list visibility.
Closes the SO->fp.job persistence audit. All 10 operational fields
identified now flow through to the WO (7 stored + populated at confirm,
3 related read-only).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Tier 2 of the SO->fp.job persistence audit. Four operational metadata
fields mirrored from sale.order:
- x_fc_internal_deadline (Date) - shop's internal target finish date,
ahead of the customer-facing deadline. Kept separate from
date_deadline (which scheduling code may adjust).
- x_fc_planned_start_date (Date) - customer-quoted planned start date.
Kept separate from date_planned_start (Datetime, capacity-adjusted).
- x_fc_internal_note (Text) - shop-internal notes from the order.
- x_fc_external_note (Text) - customer-facing notes, printed on
traveller / BoL / cert.
All four populate at SO confirm via _fp_auto_create_job, and surface
on sale.order.line as stored related fields for per-line visibility.
fp.job form view gets a Notes group alongside the Customer References
group from Tier 1.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Tier 1 of the SO->fp.job persistence audit. Three customer-reference
fields entered on sale.order's Plating tab were not flowing through
to fp.job (or SO lines), so the shop floor and printed paperwork
(traveller, BoL, cert) had to round-trip via sale_order_id every time.
Changes:
- fp.job: new x_fc_customer_job_number (Char, tracking), x_fc_po_number
(Char, tracking), x_fc_rush_order (Boolean, tracking). All three
populated by _fp_auto_create_job at SO confirm time.
- sale.order.line: x_fc_customer_job_number / x_fc_po_number added as
stored related fields off order_id so per-line list views show the
customer's references without navigating to the order header
(x_fc_rush_order was already on lines).
- fp.job form view: small Customer References group under the title
surfaces the three fields where the user expects them.
Verified end-to-end: SO -> SO line related fields -> fp.job direct
fields all carry the same value.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two improvements to the Process Tree visualization opened from the
Work Order's Process Tree header button:
1. Back button returns to the Work Order (job form) instead of
Plant Overview. fp.job.action_open_process_tree now passes
back_job_id in the client-action context; process_tree.js
reads it via a new backJobId getter, updates the button label
to "Back to Work Order", and routes onBack to fp.job form.
The Plant Overview fallback stays for callers that don't pass
either back_step_id or back_job_id.
2. Completed operation/step cards now have a green fill (#1e8449)
and a subtle pulsing glow (box-shadow animation, 2.6s alternate)
so finished work pops against still-pending dark cards. Hover
pauses the animation so the click target is steady. Reuses the
same green the workflow-state slice already used.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User wanted Finish & Next to drop its text label like the other row
buttons, but stand out visually as the primary action. Solution:
icon-only with a vivid green color and a subtle pulse animation.
- New SCSS: fp_finish_btn.scss with branch-on-$o-webclient-color-
scheme so the dark bundle uses green-400 (pops on dark bg) and
light bundle uses green-600. Pulse animation 1.8s ease-in-out
infinite, scale 1.0 ↔ 1.18. Pauses on hover/focus so the click
target is steady.
- Registered in both web.assets_backend and web.assets_web_dark
per the project's dark-mode rule (CLAUDE.md).
- View: string="Finish & Next" → title="Finish & Next",
class drops "text-primary", gains "o_fp_finish_btn".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
UI: secondary row buttons in the embedded step list are now icon-only
with tooltips (Pause / Complete 1 → Next / Record / Skip / Move…).
Saves ~70% horizontal space. "Finish & Next" stays text+icon as the
primary action.
Fix: removed the racking-inspection gate from button_finish. Racking
is now a recipe step (not a separate inspection workflow), so the
"Racking inspection for ... is Inspecting — must be Done" error no
longer fires. _fp_check_racking_inspection_complete() helper is
preserved for diagnostics but no longer called from the finish path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User feedback: operators with small parts (e.g. valve bodies) batch
them through the whole recipe. The previous behavior — Finish & Next
raising "use Complete 1 → Next or Move..." when qty>1 — forced N
clicks for a workflow that's naturally one click.
Change: _fp_record_one_piece_auto_move now ALWAYS bulk-moves
qty_at_step parts to the next step in one move record, regardless of
whether the qty is seed-only (first / paperwork step) or real (parked
from an upstream move). Audit trail is preserved (one move row per
finish), operator gets one click.
Three buttons now map cleanly to the three workflows:
- Finish & Next: bulk all parts forward, finish, auto-start next
- Complete 1 -> Next: streaming flow, move 1 part, stay open
- Move...: explicit qty + destination wizard for partial batches
Verified end-to-end on entech: seed qty=6 + real-incoming qty=6 both
move forward in a single click each.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The qty gate I added refused Finish on steps where qty_at_step > 0,
to force operators to move parts forward first. But the first-step
seed in _compute_qty_at_step gives the earliest non-terminal step
a notional qty = job.qty — a UI hint, not actual parked parts.
Paperwork steps (Contract Review, Inspection-by-paperwork, etc.)
sit on that seed, and the gate was blocking Finish with a misleading
error.
Fixes:
- button_finish gate now checks for REAL incoming moves before
refusing. Seed-only qty (no incoming_move_ids filtered to non-
self-loop) is exempt.
- _fp_record_one_piece_auto_move detects seed-only qty and bulk-
moves ALL parts in one shot to the downstream step. Correct for
paperwork / first steps where parts don't physically wait
per-piece — one click finishes the paperwork and pushes the whole
batch forward.
For steps with REAL incoming moves (parts actually moved here via
a Move record), the original gate semantics still apply: qty == 1
auto-moves one part; qty > 1 raises with the "use Complete 1 → Next
or Move…" message.
Verified on entech: Contract Review with seed qty=6 now finishes
cleanly, bulk-moving all 6 parts to the next step in one move.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Received milestone was tied to recipe steps tagged
default_kind='receiving'. But receiving in this system is a pre-
recipe inbound logistics flow (fp.receiving model in
fusion_plating_receiving). When parts physically arrive, the flow
sets sale_order.x_fc_receiving_status to partial or received.
Changes:
- New trigger_on_parts_received Boolean on fp.job.workflow.state.
- _fp_is_passed_for_job branch: passes when sale_order's
x_fc_receiving_status is in (partial, received).
- _compute_workflow_state_id depends extended with
sale_order_id.x_fc_receiving_status so the bar recomputes
automatically when the receiving flow updates the SO.
- DB seed update: Received state drops trigger_default_kinds=
'receiving' and gains trigger_on_parts_received=True.
Verified end-to-end on entech: bar moves Confirmed → Received on
status change, regresses on rollback, accepts both 'partial' and
'received' as satisfying the milestone.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two coupled fixes so the workflow bar shows "In Progress" when work
is actually underway, even on recipes without kind tagging:
B. Auto-promote fp.job.state on first step start.
fp.job.step.write hook detects step transitions to in_progress
and promotes parent job state from 'confirmed' → 'in_progress'.
Without this, fp.job.state never reached 'in_progress' anywhere
in the codebase, so the trigger_on_job_state='in_progress'
path was dead code.
C. Smarter trigger_first_step_started for untagged recipes.
For tagged recipes (any step has kind in wet/bake/mask/rack),
keep the strict kind-based check. For untagged recipes (all
steps kind='other' or similar), fall back to "any step in
in_progress/paused/done" so the milestone fires regardless of
recipe authoring quality.
Verified end-to-end on entech with untagged steps:
- confirmed → in_progress when first step starts
- workflow bar tracks at in_progress through the work
- workflow bar advances to done when all steps done/skipped
Recipe authoring still encouraged for full Received / Inspected
intermediate states (those keep their default_kind triggers).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Root cause: two compounding bugs in fp.job.workflow_state_id.
- The "Confirmed" state seed has no trigger fields set, so it
never passes _fp_is_passed_for_job.
- The _compute_workflow_state_id loop breaks on the first
non-passed state — so when Confirmed fails, every later
state stays unevaluated and the bar is stuck at Draft.
Fixes:
- Add trigger_on_job_state Selection field on fp.job.workflow.state
with values confirmed/in_progress/done. Passes when fp.job.state
>= the chosen value ("at least" semantics with explicit ordering
that treats on_hold==in_progress and cancelled outside the
progression). Lets workflow states key off the job's own state
when recipe default_kind tagging isn't present.
- Extend _fp_is_passed_for_job with the new branch.
- Change _compute_workflow_state_id from first-non-pass-breaks to
highest-passed-wins. Untagged/not-applicable states no longer
block the cascade — the bar shows the furthest milestone the
job has actually reached.
- Seed update (DB-side, since data is noupdate=1): Confirmed now
has trigger_on_job_state='confirmed'.
Result: Work Order # 00011 (state=confirmed, all 11 steps done/
skipped) now correctly shows the bar at "Done" instead of "Draft"
(via the existing trigger_all_steps_done on Done). Mid-flight
confirmed jobs without recipe tagging will show at least
"Confirmed" now.
Recipe authoring note (out of scope here): for accurate Received /
In Progress / Inspected intermediate states, recipe nodes still need
default_kind tagging (receiving / wet|bake|mask|rack / final_inspect).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
When a supervisor edits a timelog's date_started/date_finished (or
deletes a stale timelog), the parent step's "Actual Min" column
was showing stale data — duration_actual is a regular Float set
once by button_finish.
Adds:
- fp.job.step._fp_resum_duration_actual: quiet helper that re-sums
duration_actual from time_log_ids.duration_minutes. Skip no-op
updates so write traffic is minimised.
- fp.job.step.timelog.create/write/unlink hooks: call the helper
on the affected parent step(s) so duration_actual stays
consistent. Write hook only fires when date_started/date_finished/
step_id changed (notes edits skip resync). step_id reassignment
resyncs both old and new parent.
- Existing action_recompute_duration_from_timelogs (manual button)
still posts a chatter entry for audit-trail use cases.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three coupled shop-floor corrections:
- fp.job._compute_display_name: renders "Work Order # 00011" in
form header, breadcrumbs, M2O dropdowns, and error messages.
DB name stays as WH/JOB/00011 - existing chatter/cert/delivery
references unchanged.
- fp.job.step.button_finish: refuses if qty_at_step > 0 AND a
downstream pending/ready step exists. Last runnable step is
exempt (parts complete in place). Manager bypass via
fp_skip_qty_gate=True context key.
- fp.job.step.action_complete_one_to_next: new per-row button
"Complete 1 -> Next" for streaming flow (large parts going
one-by-one). Records move(qty=1) to next step; if drain takes
qty_at_step to 0, auto-finishes source + auto-starts destination
via existing action_finish_and_advance.
- fp.job.step._fp_record_one_piece_auto_move: auto-move shim
wired into action_finish_and_advance. qty=1 + downstream =>
silently record move(1). qty>1 + downstream => raise pointing
at Complete 1 -> Next. Last step always allowed.
- 16 new TestQtyGate tests covering gate / shim / auto-finish /
last-step exemption / display rename / Move wizard zero-qty.
Spec: docs/superpowers/specs/2026-05-12-step-qty-gate-and-display-rename-design.md
Plan: docs/superpowers/plans/2026-05-12-step-qty-gate-and-display-rename.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Implements the milestone-cascade design (Phase 1) and catches the
fusion_plating_jobs / fusion_plating_certificates source up to entech.
Milestone cascade (this PR's core):
- fp.job: new computes all_steps_terminal, next_milestone_action,
next_milestone_label; dispatcher action_advance_next_milestone with
3 helpers (_action_open_draft_certs, _action_open_draft_delivery,
_action_mark_active_delivery_delivered); _resolve_required_cert_types
resolver; _fp_create_certificates rewritten to honour
part.certificate_requirement + partner flags + loop over resolved
cert types
- fp.job.workflow.state: new trigger_on_delivery_state Boolean;
_fp_is_passed_for_job extended with delivery-state branch;
Shipped state seed reroutes from default_kind=ship to the new trigger
- View: hide Finish & Next when all_steps_terminal; add 4 mutually-
exclusive milestone buttons (Mark Job Done / Issue Certs / Schedule
Delivery / Mark Shipped) bound to one dispatcher
- Cert gate (fusion_plating_certificates/models/fp_delivery.py):
action_mark_delivered hard-blocks on draft certs; manager bypass
via fp_skip_cert_gate=True context key
- 24 unit tests in test_fp_job_milestone_cascade.py covering computes,
resolver, dispatcher, cert gate
- Spec: docs/superpowers/specs/2026-05-12-job-milestone-cascade-design.md
- Plan: docs/superpowers/plans/2026-05-12-job-milestone-cascade.md
Other entech changes caught up in this sync (from earlier session
patches not previously committed):
- fp.job version bump series 18.x → 19.0
- res_users_views.xml addition (signature widget in user prefs)
- racking inspection smart button removal
- various view/manifest touch-ups
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three fixes from user feedback:
1. Chatter posting raw HTML
_AUDIT_BODY in migration 19.0.18.8.0 was a plain str with <p>
tags. message_post escaped it for safety, so the chatter pill
rendered '<p><strong>...</strong></p>' literally to the recipe
author. Wrapped in markupsafe.Markup so Odoo recognises it as
safe HTML. Going forward: ANY message_post body containing HTML
tags MUST be wrapped in Markup() — most callers already do this,
the migration script was the outlier.
2. Library template editor showed raw <p> tags
onOpenLibraryEdit was JSON-cloning the payload directly without
running description through the existing _htmlToText helper that
the per-step editor uses. Added the conversion. Save path
(onSaveLibraryEditor + library_save) already wraps via
_textToHtml so storage stays HTML-compatible.
3. Per-step inline form was missing critical fields — user had to
delete + re-add a step to change Type/workflow trigger/parallel/signoff
onToggleEdit now also captures default_kind, triggers_workflow_state_id,
parallel_start, requires_signoff into the edit state. onSaveStep
sends them in the write vals. Added _fpResetStepEdit helper to
keep open/cancel/save reset paths in sync.
New per-step form has:
* Step Type (Default Kind) dropdown — drives workflow milestone
triggers + step-kind routing (e.g. contract_review opens QA-005)
* Triggers Workflow State dropdown (Sub 14) — per-step override
* Parallel Start checkbox (Sub 13)
* Require QA Sign-off checkbox
step_write controller endpoint also gained a field whitelist —
was previously accepting any vals dict from the client (security
hole + opaque to maintainers).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three issues from user testing on entech:
1. RPC error: column fp_step_template.triggers_workflow_state_id
does not exist
Root cause: the field was declared in fusion_plating CORE, but
its target model fp.job.workflow.state lives in fusion_plating_jobs.
Odoo loads core BEFORE jobs (jobs depends on core), so when core's
field declaration runs, the comodel doesn't exist yet — and Odoo
silently skips creating the column.
Fix: moved the field to fusion_plating_jobs/models/fp_job.py via
_inherit. Now the column is added when jobs loads (after core),
and the FK target is resolvable.
2. No chatter on the Workflow State form
Added _inherit = ['mail.thread', 'mail.activity.mixin'] to
fp.job.workflow.state. Tracking enabled on name/code/sequence so
admins see who changed the milestone vocabulary. <chatter/> widget
added to the form view.
3. Form layout still showed cramped 2-col help text
The XML file on disk had my new alert-info card, but Odoo's DB
ir_ui_view still held the old arch. The -u didn't refresh it
(likely because the file's mtime didn't change between deploys).
Fix: bump version + the next deploy will run a SQL DELETE on the
ir_ui_view record so Odoo recreates it from XML on -u.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two follow-ups on the workflow state work:
1) Form layout
The "How triggers combine" help text was crammed into a 2-column
group, taking ~25% of the available width. Pulled it out of the
group and rendered as a full-width <div class="alert alert-info">
below the trigger fields. Same fix applied to Notes — uses a
<separator> + bare <field> for full sheet width.
2) Simple Recipe Editor support
The trigger field was only exposed in the Tree Editor. Added it
to the Simple Editor's inline library form too:
* fp.step.template.triggers_workflow_state_id (new Many2one) —
per-template default, snapshot-copied to recipe nodes when
dropped into a recipe (added to _SNAPSHOT_FIELDS).
* /fp/simple_recipe/workflow_states/list — new endpoint to feed
the dropdown. Soft-fails when fusion_plating_jobs isn't
installed (returns []).
* Library editor JS — _fpEnsureWorkflowStatesLoaded helper
caches the catalog on first open (create + edit paths both
warm it). Save vals carry the trigger id.
* Library editor XML — dropdown rendered after the flag
checkboxes. Hidden when the catalog is empty so the form
doesn't show a useless "— None —" pick.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The compute appended '[code]' so admin pages could disambiguate
states at a glance. But display_name is what the status-bar widget
uses to render each pill, so every pill came out as 'Received
[received]', 'In Progress [in_progress]', etc.
Removed the compute. Admin list view already shows code as a
separate column.
Earlier commit parented the new menu directly under menu_fp_config,
making it appear at the top alongside the 7 themed buckets instead
of inside one. Workflow milestones map directly to recipe-step
kinds, so 'Recipes & Steps' is the natural home.
Replaces the generic Draft/Confirmed/In Progress/Done statusbar with
a shop-configurable list of plating-specific milestones. Bar advances
automatically as recipe steps complete; no manual button clicks.
What ships
==========
* New model: fp.job.workflow.state
Catalog of milestones (name, code, sequence, color, triggers).
Triggers can be:
- trigger_default_kinds: "receiving,inspect" matches by step.default_kind
- trigger_first_step_started: any wet/bake/mask/rack step started
- trigger_all_steps_done: every non-cancelled step in done/skipped
- block_when_quality_hold: held back while NCR/hold open
Plus per-recipe-node override (see below).
* Default 7-state seed (data/fp_workflow_state_data.xml):
Draft → Confirmed → Received → In Progress → Inspected → Shipped → Done
noupdate=1 so per-shop edits survive module upgrade.
* Recipe-side trigger field on fusion.plating.process.node:
triggers_workflow_state_id (Many2one, optional)
Wins over default_kind matching. Lets the recipe author pin a
specific step as a milestone trigger even when default_kind isn't
set or doesn't match. Exposed in the Recipe Tree Editor properties
panel (dropdown sourced from the catalog).
* fp.job.workflow_state_id (computed, stored)
Iterates the catalog in sequence order; lands at the highest passed
milestone. Recomputes on step state / kind / recipe_node / quality
hold changes. Replaces fp.job.state on the form's statusbar.
* Settings UI: Configuration > Workflow States
Standard list+form pages so admins can add / edit / deactivate
states. Manager-group write permission, supervisor read.
What this does NOT do
=====================
* Doesn't drop fp.job.state — that field still drives the internal
state machine (button_confirm, action_cancel, etc.). Only the
UI statusbar is reassigned.
* No migration for existing jobs — they auto-recompute on next read
because workflow_state_id is a stored compute with the right
api.depends. Existing WH/JOB/00342 will display its current
workflow state on next page load.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User feedback: operators kept asking why their work order said "Step 10"
for the first row. The 10-spacing was originally there to allow midpoint
inserts (insert sequence 15 between 10 and 20 without renumbering).
Tradeoff is operator confusion, and recipe authors rarely insert in the
middle anyway. Switching to 1-based contiguous sequences.
Files changed (every step-sequence allocation in the codebase):
fusion_plating_jobs/models/fp_job.py
_generate_steps_from_recipe — seq_counter starts at 1, increments by 1.
This is the path that builds fp.job.step records, so new jobs now show
Step 1, 2, 3, ... in the work order.
fusion_plating_bridge_mrp/models/mrp_production.py
Same change for the legacy MRP bridge so customers still on
mrp.production also get 1-based numbering.
fusion_plating/controllers/recipe_controller.py
- create_node: max_seq + 1
- reorder_nodes: idx + 1
- swap renumber: i (was i * 10)
- paste-import renumber: i (was i * 10)
- move_node: max_seq + 1
- _copy_subtree (recipe duplicate/import): i (was i * 10)
fusion_plating/controllers/simple_recipe_controller.py
- _sequence_for_position rewritten — always renumbers siblings to
keep them contiguous. Returns pos + 1 for the inserted node.
Old code used midpoint-with-fallback-to-renumber (10/20/30 spacing).
- step_reorder: i (was i * 10)
- library_input_add + step_add_input: existing_max + 1
What this DOESN'T do
Existing fp.job.step records keep their old sequences (10, 20, ...).
Re-confirm the SO to spawn a fresh job if you want the clean 1-based
numbering on a current test job. No data migration — we're in dev
and the user explicitly said test data is disposable.
What this DOES do
Every NEW job created from this commit forward shows Step 1, 2, 3, ...
Every NEW recipe step inserted via the simple editor / tree editor
also gets sequence 1, 2, 3, ...
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User reproduced on WH/JOB/00342: clicked Start on Incoming Inspection
while Contract Review was still in_progress. Sub 13 should have raised
UserError. It didn't. Both steps ended up in_progress.
Investigation:
$ grep "def button_start" fusion_plating_jobs/models/fp_job_step.py
88: def button_start(self): ← Sub 13 gate code
876: def button_start(self): ← Policy B + Sub 8 (older)
Two definitions of the same method in the same class. Python uses the
SECOND. My Sub 13 gate at line 88 was dead code from the moment it
landed. WH/JOB/00342's Contract Review and Incoming Inspection both
ran in_progress because the live button_start (line 876) only did
Policy B Contract Review auto-open and Sub 8 Racking auto-open — no
predecessor check.
Fix:
* Removed the duplicate button_start at line 88 (left a marker
comment so the next person doesn't redo this footgun)
* Merged the Sub 13 predecessor gate AND the receiving soft check
into the line-876 button_start so all four behaviours run from
one method:
1. Predecessor gate (raise UserError if blocking)
2. Contract Review auto-open (route to QA-005)
3. Racking auto-open (route to inspection)
4. super().button_start() + receiving check + serial promotion
Helpers _fp_should_block_predecessors / can_start / _compute_can_start
preserved (used by view + Move wizard too).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bug: action_finish_current_step (the header-level Finish & Next
button on the job form) called button_start() without capturing its
return value. So when button_start returned an action (e.g. the new
QA-005 redirect for contract_review steps from 21e42e7), the header
method threw it away and returned True. Result: operator clicked
Finish & Next, the step started, but no navigation. They had to
click again — the second click found the in_progress step, called
action_finish_and_advance, which returned the QA-005 action.
Two clicks instead of one to land on QA-005.
Fix: capture button_start's return value. If it's a dict (= an
action), return it. Otherwise return True (the normal case).
User reproduction (WH/JOB/00341):
Header > Finish & Next (1st click) → step starts + QA-005 opens
Sign / dismiss QA-005 → back to job
Header > Finish & Next (2nd click) → step finishes + next starts
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User feedback on WH/JOB/00341 (S00279 retest): clicking Start on
the Contract Review step changed state to in_progress but didn't
take them to QA-005. They had to then click Finish & Next twice
to land on the form — confusing flow.
Better UX: when an operator clicks Start on a step where
recipe_node.default_kind='contract_review', the step starts AND
the QA-005 form opens immediately. Operator signs/dismisses,
navigates back, hits Finish & Next once → step finishes + advances.
Implementation:
fp.job.step.button_start, after super() returns and the
receiving check runs, calls _fp_contract_review_redirect()
(existing helper). If it returns an action, return that
instead of the parent's result. Single-record only — bulk
button_start (job-level start-all) shouldn't navigate.
Helper logic unchanged — same gate matrix:
* recipe_node.default_kind == 'contract_review'
* job has part_catalog_id
* review state NOT in (complete, dismissed)
When review is already complete, the gate clears: button_start
returns the normal True so the operator can advance the step
without bouncing through QA-005 again.
Tests:
test_button_start_routes_cr_step_to_qa005 — start opens QA-005
test_button_start_does_not_route_when_review_complete — start
does NOT redirect once review is signed off
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Scrapped the v2/v3 form-view + list-as-cards CSS approach after
extensive failure to make Odoo's editable list look like cards.
Built a proper OWL Dialog component instead, mirroring the pattern
used by fusion_plating_shopfloor's move_parts_dialog.js.
What changed
============
* New OWL Dialog: fp_record_inputs_dialog.js
- Loads step + prompt definitions via /fp/record_inputs/load
- Renders each prompt as a semantic <div class="o_fp_ri_card">
- Per-row widget chosen by input_type:
numeric/temperature/thickness/time_seconds/ph -> number input
boolean/pass_fail -> custom CSS toggle (clearer than Bootstrap)
date -> datetime-local input
photo -> file picker w/ preview + clear
multi_point_thickness -> 5-cell grid + live average
bath_chemistry_panel -> pH/Conc/Temp/Bath grid
selection -> dropdown sourced from selection_options
text/signature/... -> text input
- Live in-range hint for numeric prompts
("in range" / "below target" / "above target")
- Save validates ad-hoc rows have a Prompt label
- Save dispatches the next_action returned by the wizard model
(e.g. action_finish_and_advance for the Finish & Next flow)
* New XML template: fp_record_inputs_dialog.xml
Full DOM control. No fighting Odoo's list view, no class-stripping
bugs from canUseFormatter, no read-mode-vs-edit-mode CSS dance.
* New SCSS: fp_record_inputs_dialog.scss
- Dark mode aware (compile-time @if $o-webclient-color-scheme==dark)
- Pure semantic selectors (.o_fp_ri_card, .o_fp_ri_input, etc.)
- 14 surface tokens with light/dark hex pairs
- Tablet polish via @media (max-width: 768px)
- Custom toggle widget (no <input type="checkbox"> hidden trick)
* New controller: controllers/record_inputs.py
- /fp/record_inputs/load: returns step + prompts payload
- /fp/record_inputs/commit: creates a wizard, populates lines,
calls action_commit (reuses existing audit-trail / synthetic
move semantics — no commit logic duplicated)
* fp_job_step.py wired to dispatch the new action
- _fp_open_input_wizard returns
{ type: 'ir.actions.client', tag: 'fp_record_inputs_dialog' }
- action_open_input_wizard same
- Contract-review redirect gate preserved (Sub 4 work intact)
* Manifest registers JS/XML/SCSS in BOTH backend + dark bundles
per the dark-mode pattern in CLAUDE.md.
What was kept
=============
* fp.job.step.input.wizard TransientModel — UNCHANGED. The new
controller's commit endpoint creates a wizard record and calls
action_commit() on it, so all the audit-trail / synthetic-move
/ chatter logic stays in Python where it belongs.
* v2 + v3 form views still exist in the XML file. If the OWL
dialog ever fails, switch action_open_input_wizard back to
ir.actions.act_window with view_id=v2 or v3.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
I was wrong about the DOM. Verified from Odoo 19 source on entech:
web/static/src/views/fields/float/float_field.xml
web/static/src/views/fields/char/char_field.xml
web/static/src/views/list/list_renderer.xml
Float/Char fields render as a BARE <span> (read mode) or BARE
<input class="o_input"> (edit mode) directly inside the <td>.
There is NO .o_field_widget wrapper. So all my prior CSS targeting
.o_field_widget matched nothing.
Also discovered: Odoo's getCellClass() in list_renderer.js calls
canUseFormatter() which strips custom <field> classes when the
column has widget="..." set:
canUseFormatter(column, record) {
if (column.widget) {
return false; // ← class stripped here
}
...
}
So o_fp_iw_value class doesn't even land on cells with
widget="boolean_toggle"/"image". Those cells render natively;
boolean toggle and image styling now targets the widgets directly
wherever they appear (.o_boolean_toggle, .o_field_image).
Fix: put visible chrome (border, bg, padding, min-height) on the
<td> itself for prompt/meta/value/extras cells. Make inner span
and input transparent + inherit. Focus ring travels up via
:focus-within on the td.
Cells now look like obvious input boxes from first paint, regardless
of whether the user has clicked into edit mode.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Root cause user kept seeing inputs as bare/borderless text:
Odoo's <list editable="bottom"> renders each cell as a read-mode
<span> inside .o_field_widget UNTIL the user clicks the cell.
Only then does an <input> swap in. My CSS was targeting
`td.o_fp_iw_value input { ... }` so the chrome only appeared on
focus. Every other (unclicked) cell looked like dead text.
Fix:
Move all input chrome (border, bg, padding, min-height) to the
.o_field_widget wrapper which is ALWAYS in the DOM. Then make
the inner <input> / <span> transparent so they inherit. Effect:
the cell looks like an input box from first paint, regardless
of focus state. Focus ring travels up via :focus-within.
Special widgets (boolean toggle, photo upload, multi-point,
bath panel) opt OUT of the wrapper chrome via :has() so they
keep their own visual treatment.
Same fix applied to .o_fp_iw_extra cells (composite types).
User reproduction: WH/JOB/00339 → Record on Masking step. After
hard-refresh + this build, every value cell should read as an
obvious input box even before the operator clicks.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Four visible bugs reported by user after deploy:
1. Type + Unit pills overlapped at top-right of every card.
Root cause: both <field>s carried the same .o_fp_iw_meta class
AND both mapped to grid-area: meta. CSS Grid stacked them on
top of each other so the labels rendered as overlap garbage
(e.g. "eachnber", "Time(secs)Time(seconds)").
Fix: distinct classes (.o_fp_iw_meta_type / .o_fp_iw_meta_unit)
each in its own grid column. Grid is now 4 columns wide:
"prompt | type | unit | trash"
2. Input borders barely visible in dark mode (#343942 on #22262d).
Operators couldn't tell where to click.
Fix: brighter border using $fp-iw-ink-faint instead of $fp-iw-border.
Hover bumps to $fp-iw-ink-mute. Focus uses brand purple. Also
added a slight surface tint ($fp-iw-page) so empty inputs read
as obviously-interactive instead of blending into the card.
3. Photo widget rendered enormous (full card width).
Root cause: max-width applied only to the preview image, not
to the .o_field_image container itself.
Fix: max-width 240px on .o_field_image AND its inner controls.
4. Numeric values floated centered in empty space.
Root cause: input width wasn't stretching to its grid cell;
default Odoo numeric-cell text-align: right plus our missing
width: 100% left tiny inputs centered in the value area.
Fix: explicit width: 100%, text-align: left, and 420px
max-width on the .o_field_widget container.
Bonus polish:
* Trash icon hidden by default (opacity: 0), reveals to 0.6 on
row hover, full opacity on direct hover. Reduces visual noise
for the common case where operator just types and saves.
* Boolean toggle scale bumped from 1.4 to 1.5 + adds left margin
so the switch sits properly inside the value cell.
* Mobile (<900px) grid collapses to: prompt|trash / type|unit /
value / extras — keeps the type+unit pair on one row but lets
them flow naturally below the prompt.
No model changes. SCSS + XML view only. v2 view still in place
for instant rollback.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two coherent feature drops shipping together because their fp_job_step
edits overlap. Both target operator workflow correctness.
## Sub 13 — Sequential step enforcement (recipe + per-step)
Background:
Investigation on WH/JOB/00339 showed operators starting Incoming
Inspection while Contract Review was still in_progress. Audit:
98.7% of recipe operations system-wide had requires_predecessor_done
= false (the legacy per-step opt-in defaults off, recipe authors
rarely tick the box).
Architecture:
Recipe-level toggle + per-step opt-out (Option A from /investigate).
* fusion.plating.process.node.enforce_sequential — Boolean on the
recipe root. Default True. When True, every operation under this
recipe waits for earlier-sequence steps to finish before it can
start.
* fusion.plating.process.node.parallel_start — Boolean on operation
nodes. When True, this step bypasses the sequential gate (e.g.
paperwork or QA review that runs alongside production).
* Mirrored on fp.step.template (parallel_start) so library steps
carry the flag into snapshots.
* fp.job.enforce_sequential — related from recipe_id. Snapshotted
at job creation so a recipe author flipping the recipe's flag
AFTER job generation does NOT change behaviour mid-run.
* fp.job.step.parallel_start — related from recipe_node_id.
* Decision matrix (encapsulated in
fp.job.step._fp_should_block_predecessors):
recipe.enforce_sequential | step.parallel_start | step.req_pred_done | block?
--------------------------|---------------------|--------------------|------
True | False | any | YES
True | True | any | no
False | any | True | YES
False | any | False | no
* Manager bypass via context fp_skip_predecessor_check=True (existing).
Runtime gates:
* fp.job.step.button_start — calls _fp_should_block_predecessors;
raises UserError naming the blocking earlier step(s).
* fp.job.step.can_start — computed Boolean for view-side disable.
* Move wizard predecessor check
(fusion_plating_shopfloor/controllers/move_controller.py) — uses
the same helper so tablet + backend behave identically.
UI surface:
* Recipe form (fp_process_node_views.xml) — enforce_sequential
toggle on recipe root, parallel_start checkbox on operations.
* Step template form — parallel_start checkbox.
* Simple Recipe Editor (inline library form) — Parallel Start
checkbox + legacy flag demoted with muted styling + supervisor
group gate.
* Recipe Tree Editor (properties panel) — both flags exposed,
only-show on the right node_type.
* Controllers updated to allowlist + payload the new fields.
Migration:
fusion_plating/migrations/19.0.18.12.0/post-migrate.py — sets
enforce_sequential = TRUE on every existing recipe-root node.
Idempotent. User confirmed dev-stage data, so retroactive flip
is safe (no production jobs to disrupt).
Tests:
TestSequentialEnforcement (10 tests) covering:
* sequential mode blocks out-of-order start
* first step always startable
* predecessor finish/skip unlocks next
* parallel_start opts out of gate
* free-flow mode bypasses gate
* legacy requires_predecessor_done still honoured in free-flow
* manager bypass via context
* can_start compute reflects state correctly
* library template parallel_start snapshots into recipe-node
## Sub 12e — Record Inputs Wizard v3 (card layout, dark-mode aware)
Background:
v2 wizard was a 17-column wide editable table. Operators got lost
finding which value column applied to their row's type, horizontal
scroll required on tablets, composite types crammed into one row.
New layout:
* Each measurement renders as a stacked card (CSS Grid + display
transformation on the existing list widget — preserves inline
editing, no JS rewrite).
* Card header: prompt name (large, bold) + type/unit pills.
* Card body: ONLY the value widget for this row's type
(number / boolean / date / text / photo / multi-point / panel).
* Composite types (multi-point thickness 5x reading + avg, bath
panel 4 fields) get inline sub-grid inside the card.
* Empty state ("no measurement prompts") with friendly CTA.
Dark mode:
* SCSS branches at compile time on $o-webclient-color-scheme
(per fusion-plating/CLAUDE.md note).
* Tokens: 7 surface colours + 4 ink levels with light/dark hex
pairs, all behind var(--fp-*) custom properties for per-deploy
override.
* Registered in BOTH web.assets_backend AND web.assets_web_dark
so each bundle compiles its own palette.
Tablet polish:
@media (max-width: 900px) — collapse meta below prompt + bump
numeric input min-height to 56px.
Defensive:
* v2 view kept in the XML file (instant rollback by changing one
view_id ref).
* `:has(.o_invisible_modifier)` rule drops empty cells out of the
grid so Odoo's invisible="..." doesn't punch holes in layout.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two bugs fixed in one drop, both targeting the contract review (QA-005)
enforcement gap reported on entech.
## Bug 1 — WO step routed to wrong wizard
Symptom: clicking Finish & Next or Record on a Contract Review step in
WH/JOB/00339 opened the generic measurement wizard with three fake
prompts (Reviewer Initials / Date Reviewed / QA-005 Approved). No path
to the actual QA-005 form from the work order.
Root cause: action_finish_and_advance + action_open_input_wizard had no
branch for recipe_node.default_kind == 'contract_review'. The step.kind
mapping collapses contract_review -> 'other' so kind-based detection
wouldn't have worked either; gate has to live at the recipe-node layer.
Fix in fusion_plating_jobs/models/fp_job_step.py (v19.0.8.14.6):
- action_finish_and_advance:329 calls _fp_contract_review_redirect
before the input-wizard branch
- action_open_input_wizard:844 same gate, keeps Record button consistent
- _fp_contract_review_redirect:866 (new) returns the part's
action_start_contract_review() unless review.state in
(complete, dismissed) — gate clears so the step can finish after
the operator signs QA-005.
## Bug 2 — Part create did not enforce contract review
Symptom: spec called for a banner-only UX. User wanted true automatic
enforcement on first part creation under an enforced customer.
Fix in fusion_plating_quality/models/fp_part_catalog.py (v19.0.4.10.0):
- @api.model_create_multi def create() override
- _fp_enforce_contract_review_on_create() helper auto-stages the
fp.contract.review record AND surfaces three prominent reminders:
1. Sticky bus.bus warning toast (top-right, doesn't auto-dismiss)
2. mail.activity (To Do) on the part for the current user
3. Smart button on the part form lights up (review now exists)
- Idempotent: skips parts that already carry a review id
- Soft-fails: bus or activity outage doesn't block part creation
- create()-only — write/update flows never re-trigger
Sub 4's existing info banner stays as a fourth surface.
## Tests
- fusion_plating_jobs/tests/test_fp_job_extensions.py:
+TestContractReviewStepRouting (5 tests covering both routing methods,
the complete/dismissed gate-clear, and non-CR step regression)
- fusion_plating_quality/tests/test_part_catalog_contract_review_enforcement.py
(NEW): 9 tests covering auto-create, batch create, idempotency,
activity surface, bus surface, write-must-not-retrigger, soft-fail.
- docs/superpowers/tests/2026-04-22-sub4-smoke.py: flipped the
"no review yet" assertion to "review auto-created" to match new
behavior. Sign-flow assertions unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Bucket 1 — Generation bug fix
- post-migrate.py for 19.0.18.8.0 promotes flat 'step' children of
recipes to 'operation' so fp.job._generate_steps() picks them up.
Filter is narrow: only direct children of node_type='recipe' get
flipped, tree-editor sub-steps (parent.node_type='operation') are
untouched. Idempotent. Posts an audit chatter note on each affected
recipe.
- Simple Editor controller hardcodes node_type='operation' on insert
+ snapshot-import path so future recipes start correct.
Bucket 2 — Inline library authoring
- 6 new JSONRPC routes (/fp/simple_recipe/library/load + save +
seed_defaults + input/{add,write,remove}, /fp/simple_recipe/tank/list).
- + New Step button in the right pane opens an inline form with name /
kind / icon / instructions / stations / flags / prompts table.
- Pencil icon on each library row reopens the same form prefilled.
- Step Kind picker leads with 'Generic — no automatic behaviour'.
- 'Seed defaults from kind' calls action_seed_default_inputs server-side
for kinds that have curated default prompts.
Bucket 3 — Back nav
- '← Recipes' button in the header (or '← Part' when opened from
Process Composer) mirrors recipe_tree_editor.js, with
clearBreadcrumbs:true to avoid stack pollution.
Verified on entech: LGPS1104's 19 'step' children now show as
'operation', migration chatter note posted on the recipe, asset cache
busted.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Spec for the upcoming Simple Recipe Editor refinement:
- Fix node_type bug so Simple-Editor recipes generate job steps
- Inline + New Step / pencil-edit library authoring with prompts
- Back button + breadcrumb-aware navigation (mirrors tree editor)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Plating Jobs list was sorted priority-desc, deadline-asc, id-desc — which
mixed Done (closed) jobs in with Confirmed (open) jobs and made the list look
chaotic to managers. Done jobs from weeks ago surfaced above active work.
Two changes:
1. New stored compute fp.job.state_priority (Integer, indexed) ranks states
by managerial relevance: in_progress=0, confirmed=1, draft=2, on_hold=3,
done=4, cancelled=5. _order now leads with state_priority asc, then
priority desc, then date_deadline asc, then id desc. Active work bubbles
to the top automatically.
2. Plating Jobs action defaults to a new 'Open' filter
(state not in done, cancelled). Managers see only active work by default;
they untick the filter to see history. Added On Hold + Cancelled filters
too for full state coverage.
Verified on entech: top 10 jobs are now all in_progress, sorted by deadline
ascending. Existing 26-row list goes from chaotic to focused.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Operator Instructions panel had a hardcoded inline style
(background: #f8f9fa) which became a white-on-dark unreadable blob
in dark mode. Replaced with a CSS class backed by an SCSS file that
branches at compile-time via $o-webclient-color-scheme — registered
in both web.assets_backend (light) and web.assets_web_dark (dark)
bundles per the CLAUDE.md pattern.
Tokens: panel bg #f8f9fa light / #22262d dark; border #d8dadd /
#3a3f47; text #212529 / #e8eaed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Click a step's name in the embedded job-form list → opens a read-only
modal with everything a manager wants in one scroll: equipment,
schedule, master collect-measurements banner, operator instructions
(rich-text from recipe_node.description), measurement prompts list,
and values recorded so far.
Implementation: separate read-only form view bound to the embedded
field via context={'form_view_ref': '...'}. The standalone editable
form view stays registered for the Job Steps menu, so direct
navigation still loads the editable variant.
Three new computed/related fields on fp.job.step:
- quick_look_instructions (Html, related from recipe_node_id.description)
- quick_look_prompt_ids (filtered+sorted recipe_node.input_ids, step_input only)
- quick_look_recorded_value_ids (search across moves: input_value rows
whose move.from_step_id == self.id)
Plus a small action_open_full_form method that escapes from the modal
to the editable form when the manager actually needs to edit.
Edge cases:
- No recipe_node_id → instructions panel shows empty-state hint
- collect_measurements=False → amber banner: "Master switch off — no
values will be collected at runtime"
- Multiple moves on same step → values list shows all, newest first
Spec: docs/superpowers/specs/2026-04-30-step-details-modal-design.md.
Verified on entech: step "11. Hard Anodize Type III" populates with
516 chars instructions + 7 prompts.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Critical UX gap discovered in production-environment battle test: when
operator hits "Mark Done" and the input wizard fires, they only saw the
measurement prompts list. The rich-text instructions written by the
office (recipe_node.description) never reached the operator at the
exact moment they need them.
Fixed: wizard model gains instructions (Html, computed from
step.recipe_node_id.description) + has_instructions flag.
Form view renders the instructions in a prominent blue alert at the
top of the wizard, above the Measurements list. Hidden when blank
so operators on instruction-less steps don't see noise.
Also: extend default_kind Selection on fusion.plating.process.node to
match fp.step.template — both models now have the same 24 kinds. Without
this, recipe authors could pick a kind in the library template form
that the recipe-node Selection rejected with a ValueError.
Battle test artifact:
- Recipe "Hard Anodize Type III + Dye + Seal" (id=1863) — 23 steps,
105 measurement prompts, rich-text operator instructions per step
- SO S00278 for ABC Manufactoring confirmed → fp.job 1236 / WH/JOB/00337
with all 23 steps materialized, 105 prompts visible to operators
- Wizard test: step "11. Hard Anodize Type III" → 516 chars of
instructions render + 7 input prompts in the form
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Battle-tested complete workflow on entech: ABC Manufacturing + Anodize
recipe (id=136) cloned to part-variant (id=1775) → SO S00276 confirmed →
fp.job 1234 with 17 steps → recorded 56 measurement values exercising all
13 input types (incl. all 4 new types) → CoC chronological report renders
69KB with all values incl. photo thumbnails.
Bugs found and fixed:
1. fp.process.node.input_ids missing copy=True — when a master recipe
was cloned per-part (the standard variant pattern), the operator
prompts on each step did NOT get copied to the variant. Result: jobs
built from variants ran with zero prompts even though the master had
them. Fixed: input_ids now copy=True so cloning auto-duplicates.
2. CoC chronological template read dest.input_ids where dest is
fp.job.step. Steps don't carry input_ids — that field lives on the
recipe node. Result: AttributeError aborted the entire CoC render.
Fixed: walk via dest.recipe_node_id.input_ids; preserves the existing
collect=True filter.
3. CoC chronological template used hasattr() in a t-value expression.
QWeb's expression engine doesn't expose Python builtins, raised
KeyError: 'hasattr'. Fixed: use 'collect' in i._fields instead.
Also enhanced photo rendering in CoC: was just "[Attachment]" placeholder;
now renders an actual <img> thumbnail (max 80px tall) plus the filename.
Battle-test script saved to fusion_plating/scripts/bt_e2e_anodize_v2.py
for re-runs / regression testing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Implements 2026-04-29-step-library-audit-design.md. Bumps fusion_plating
to 19.0.18.7.0, fusion_plating_jobs to 19.0.8.12.0, fusion_plating_reports
to 19.0.10.2.0.
LIBRARY EXPANSION
- 8 new Step Kinds: Receiving, Electroclean, Strike, Salt Spray,
Adhesion Test, Hardness Test, Packaging, Tank Replenishment
- 4 new input types: photo, multi_point_thickness, bath_chemistry_panel, ph
- DEFAULT_INPUTS_BY_KIND rewritten to seed audit-grade prompts on every
kind (bath IDs, photos, multi-point thickness, signatures, etc.)
- + Common Audit Fields one-click button on the library template form
- Default Operator Instructions relabel + alert callout
PER-RECIPE CONFIGURABILITY
- collect (Boolean) per recipe-step input prompt — opt out without delete
- collect_measurements (Boolean) master switch on recipe step — when off,
wizard skips entirely
- template_input_id (Many2one) traceability link from recipe to library
- Recipe-step backend form view exposes the new fields with handle drag,
toggle, target range, and library-source column
RUNTIME WIRING
- Step input wizard filters node.input_ids to step_input AND collect=True;
short-circuits on collect_measurements=False
- New input types: photo (image widget + ir.attachment), multi-point
thickness (5 readings + auto avg, skips empty cells), bath chemistry
panel (pH/conc/temp/bath bundle), pH (0-14 numeric)
- Composite values JSON-serialized into value_text; photo via attachment
CoC REPORT
- Filters captured prompts to collect=True only
- Renders new input types with appropriate format
MIGRATION (post-migrate.py for 19.0.18.7.0)
- Backfills collect=True on recipe-step inputs
- Backfills collect_measurements=True on recipe steps
- Re-runs action_seed_default_inputs on every existing template
(idempotent, preserves user edits)
- Backfills template_input_id by name-matching against source library
template (handles JSONB vs varchar name columns)
SEED DATA
- 8 example templates (one per new kind) in fp_step_template_data.xml
with noupdate=1
BATTLE TEST
- bt_step_library_audit.py: 29 assertions all PASS on entech
OWL EDITOR EXTENSION DEFERRED
- The simple recipe editor's per-step Instructions/Measurements
expansions were not implemented in this pass; users configure via the
backend recipe-step form. Track follow-up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds 8 new Step Kinds (Receiving, Electroclean, Strike, Salt Spray,
Adhesion Test, Hardness Test, Packaging, Replenishment) with industry-
standard default measurements. Adds 4 new input types (photo,
multi_point_thickness, bath_chemistry_panel, ph). Beefs up existing
kinds (cleaning, etch, plate, bake, ship, etc.) with bath ID, photos,
multi-point thickness, signatures.
Per-recipe configurability: each recipe step can disable, rename,
retarget, reorder prompts; add custom prompts; toggle entire-step
data collection. Library is the smart default; recipe is final say.
Office-to-operator instructions: relabel as Default Operator
Instructions in the library; per-recipe override surfaced in the
simple recipe editor; falls back to library default at runtime when
recipe override is empty.
Battle test plan covers 18 assertions end-to-end on entech.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Resolution chain: explicit override → days offset → part lead time → order
commitment. Adds x_fc_default_lead_time_days on part catalog; per-line
effective_part_deadline + effective_internal_deadline computes; order-level
completion_date rollup + is_late_forecast warning.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
