Bridging physical steel with critical path scheduling

In heavy industrial manufacturing, scheduling tools frequently operate in corporate fiction. Traditional Gantt software places tasks based solely on typed calendar dates and task dependencies, assuming raw materials and hardware exist out of thin air.

On the hangar floor, that assumption created daily operational friction.

I redesigned the scheduling engine to make physical warehouse deliveries a first-class scheduling predecessor, synchronising dock receipts directly with the active project timeline.

Before rolling the design system across the entire OS, the Tasking module served as the end-to-end pilot migration, 31 production files in all. It was the right proving ground precisely because it is the densest interface in the platform: a thousand rows of real project data will find a spacing rule that a static frame of six rows never will.

Testing on live datasets surfaced physical constraints that no amount of review in a design file could have caught.

To guarantee that work cannot be scheduled before physical parts arrive, a task’s earliest start is bounded by four operational constraints: its planned dates, its preceding tasks, its vendor delivery dates, and the shop-floor anchor. The latest of the four wins, because a constraint that can be overridden is not a constraint.

The shop-floor anchor. If a task’s planned start has already passed while it sat waiting on delayed materials, an arrival today cannot revive an expired date in the past. The system snaps earliest fabrication forward to the active work shift, which is the only date a fitter can actually act on.

Plant working calendar awareness. The anchor respects the facility’s four-day schedule, Monday through Thursday, so a Friday delivery schedules the earliest weld for Monday morning rather than for a day nobody is in the bay. The timeline says so on the bar itself rather than leaving the coordinator to work out why a date moved.

Status purple isolation. Purple is quarantined strictly as a material dependency indicator, powering the package badge, the purchase order tag and the stalled timeline bar. It is never allowed on a standard interface button, so a coordinator scanning a wall-mounted timeline can find every material blocker without reading a word.

A purchase order line is a finite pool, but the legacy system let every task claim against it independently. Two coordinators could each allocate the same eight units and both believe they were covered until the truck arrived. I implemented an available-to-promise model that calculates unallocated stock in real time, deducting every other task’s claim before it will accept a new one.

The arithmetic was the easy half. The harder half was showing it without turning every row into a spreadsheet, so the material link modal was rebuilt around progressive disclosure: fast to scan by default, complete on demand.

Clean primary scannability. Each line shows only what an operator needs to decide: the part title, the shipping status such as In-transit · ETA Aug 28, the allocated ratio 1 / 3, and a single link trigger. Everything else is one interaction away.

The over-allocation lock. When a line is fully committed elsewhere, the stepper locks to 1 / 0 and the row says so in plain words, fully allocated, nothing left to link, with the trigger disabled. A disabled control that does not explain itself is just a bug the user cannot report.

Claimant popovers on demand. Clicking an allocation ratio opens the breakdown: total ordered, what has actually landed at the dock, the unallocated balance, and the specific tasks currently holding units. That last part is what turns a refusal into a negotiation, because the coordinator who is blocked can now see exactly who to go and ask.

In legacy tasking systems, adding a subtask to an active job converted the row into a parent container and permanently destroyed its material links and dependencies. Which meant the safest thing a coordinator could do with a live task was leave it alone.

Live constraint re-pointing. When a task becomes a parent group, its material links and predecessor dependencies transfer automatically to the first subtask, preserving allocated quantities and history rather than recreating them. Nothing is dropped, so nothing has to be re-entered.

Dynamic rollup envelopes. The parent derives its timeline from its children, spanning the earliest child start to the latest child finish, so a summary bar can never contradict the work underneath it.

Inside the task inspection drawer, material items render as structured state cards rather than flat text tags. A tag reading “pending” tells a coordinator nothing they can act on. A count and a date do.

Traditional project management algorithms calculate float relative to whichever task happens to finish last. In aerospace fabrication that produces a dangerous blind spot: the algorithm moves the goalposts as the work slips, masking the upstream delays that threaten a fixed customer delivery date.

I redesigned the backward pass to anchor directly to the workstream target delivery milestone. The deadline is fixed by contract, so the arithmetic should treat it as fixed.

True negative float. When an assembly chain slips past its contract delivery date, float drops below zero and the entire delayed sequence renders in crimson, making schedule pressure obvious at a glance. The value is not clamped, because a chain four days late and a chain one day late need different responses.

Honest slack representation. When work is comfortably ahead, the timeline displays zero critical path pressure rather than arbitrarily colouring the final task red. An indicator that is always populated teaches people to ignore it.

Automatic graph recalculation. Any delivery date change logged by purchasing updates downstream floats across the active project graph in real time. A vendor email at four in the afternoon moves the timeline before anyone has opened it.

The core insight of Engine 05 was proving that scheduling software cannot live in isolation from the shop floor and the warehouse. Every failure in the legacy system traced back to the same root: the timeline had no channel through which physical reality could reach it.

When vendor delivery slips, dock receipts and piecework progress are embedded directly into the scheduling graph, the Gantt chart stops being a decorative presentation of the plan and becomes an authoritative industrial operating tool. The same instinct runs through the materials and procurement ledger ↗, where the constraint was the receipt rather than the schedule.