Live exceptions
Ranked by carton-equivalent impact on weekly throughput and committed delivery. —Where finishing hours went, week by week
Six finishing work centers (302 · 310 · 316 · 317 · 321 · 325). Changeover is measured from booked Changeover events.
Run
Changeover (make-ready)
Downtime
Changeover share vs. weekly output
Measured changeover against weekly output, 51 weeks.On-time delivery exposure
Items whose committed (allocated) quantity exceeds on-hand + on-order.Source: FGIemExport_2026-09-09 · Qty Allocated, Qty On-hand, Qty On Ord, Qty Ordered YTD
Make-ready burden by work center
Mean measured changeover per operation, 12 months, across all four areas.Rate and changeover burden
Good cartons per run hour against changeover share of occupied time.All work centers
9 Sep 2025 – 9 Sep 2026, plant historian export — 156,755 event rows.Line availability — gluer 325 entered service part way through the window
Its twelve-month averages are not comparable with the other five lines. correctionShift schedule and planned availability
The plant's own shift definitions. Every capacity figure in this application is computed against these hours.Preventive maintenance and wear
Weekly PM by line, and what the event log can and cannot tell us about it.Machine calibration parameters
Rail positions and set-up parameters — what exists today, and what the model would need.Failure pathway – how changeover time becomes a missed shipment
Click any node for the evidence behind it. Edge labels are fitted effects, not opinions. GlassboxAttributed changeover hours — finishing
Fitted model applied to the measured changeover events in the attributable window.Fitted model
OLS on make-ready hours per transition.Product-family transition matrix — finishing
Mean measured changeover hours, from-family (row) to to-family (column). Blank = never sequenced that way in 12 months.Diagonal = staying inside a family. Every off-diagonal cell is a campaign break.
Observed contrasts
Raw means, before the model separates overlapping causes.How an order moves through the plant
Every job number is order‑release‑operation. One order is worked in up to four areas, in this order, and the release suffix counts repeat runs.Source: 156,755 event rows in the plant historian export, grouped into 1,846 work-center operations.
Where the queue actually is
Capacity against the open commitment, by area.Route mix
How the year's release-runs were routed.Machine compatibility — how much routing freedom actually exists
From Item Run Speeds and observed history: an item can only run where a rate has been established.Finishing coverage by product family
Open commitment, cartons, by family and qualified line.What the data cannot see
Tier 2 and Tier 3 knowledge that must be captured rather than inferred.Captured in the ISF Orchestrator — the scheduler agent there merges these rules with the historical model.
Data tiers
What the engine consumes today, and what remains to be captured or connected.Can the plant make the year?
A refined version of the finishing capacity projection, driven by measured rates and the crew that is actually on the floor. liveDemand
What the year has to produce.52
0.285
Both are the plant’s own planning figures, not ISF assumptions.Crew on the floor
Gluers staffed per shift. This is the lever that moves capacity.Demand against capacity, line by line
Required run hours use the measured rate by default. A site planning rate typically sits below it; the toggle shows both.Crew scenarios
Weekly cartons against demand at each staffing level.Where the week goes
Opportunity hours per line, and what is actually claimed.Upstream — two parallel routes, and one of them is day-shift only
All work goes either through the sheeter 600 into the press 900, or through the press 917. Never both.Units — sheets upstream, cartons in finishing
One sheet converts to about 4.3 cartons. Every cross-area figure in this application applies that conversion.Sleeve work distorts two of the line averages
Sleeves run far faster than the general mix and need almost no changeover between one another. correctionWhat closes the gap
Each lever priced against the same demand, so they can be compared rather than argued about.Optimisation engine — route and sequence the open commitment
Work centers are grouped by area. An order is scheduled through every area on its route in turn; a stage cannot start before the stage in front of it finishes. idle
Finishing alone is the default. Adding an upstream area makes the engine respect the full route, and the flow time becomes a measured output rather than an assumption.
Capacity: —
Changeover60
On-time delivery40
Throughput45
Load balance25
Weights are relative; the engine normalises them.By area
Each area prices its own transitions from its own fitted model.Search frontier
Every candidate weighting the engine evaluated.Work center loading
Baseline vs optimised, against each line's own capacity.Optimised sequence
Every transition carries the reason the engine paid for it.| Run the optimisation to build a sequence. |
Committed schedule
The workflow saved from the What-If Simulator, ready for the floor. nothing savedLine schedule
Teal = running, orange = make-ready. Hover a block for the job.Run
Changeover
Day boundary
No committed schedule yet.
Job cards
Replaces the index-card board. Scan-ready sequence per line.| Save a scenario from the simulator to populate the board. |
Scheduler agent — changeover & delivery workflow
Tier 1, Tier 2 and Tier 3 sources are assembled, submitted to the Causal Control Tower, and returned as a committed schedule. idle
Last run: never
Tier 2 variable inputs
The plant conditions the schedule depends on. Change one and the workflow re-solves against it. in syncCausal Control Tower — current status
Response returned to the orchestrator by the backend service call. no responseRun the workflow to query the control tower.
Agent mesh log
Every call the scheduler agent made, in order.--:--:--IDLEAwaiting run request.
Committed schedule to execute
Released to the floor as scan-ready job cards, one sequence per folder-gluer.No schedule committed in this session.
Tier 2 rule register
Tribal knowledge captured as explicit rules and applied as adjustments to the fitted model.Rules are authored by plant staff and versioned. Each carries the model term it adjusts, so its effect stays auditable.