The Self-Healing Schedule: Inside wiseDo's Auto-Replan Engine

The 10 AM problem

The morning schedule is the most short-lived document in manufacturing. The planner builds it at 7:00 — whiteboard, Excel, or an ERP screen. At 9:41, a job on the CNC runs twenty minutes over. Nothing dramatic; a tough batch of material, an extra tool change. But every operation queued behind it just moved, on every machine its downstream steps touch, for every order in the building.

And here's the expensive part: nobody re-plans until it hurts. The planner is in a meeting. So the floor keeps working to a plan that's already fiction — material staged for the wrong next job, a rush order silently going late, a machine sitting idle because its queue was built for a world that no longer exists. Sometime after lunch, someone rebuilds the whiteboard by hand. Tomorrow, the same thing happens again.

The problem was never building the schedule. It's that a schedule is a prediction, and predictions rot the moment reality diverges. What a plant actually needs is a schedule that notices it's wrong and fixes itself — in seconds, not after lunch.

Why your ERP's dates are wishes

Before showing how self-healing works, one uncomfortable truth about where most plants' dates come from. ERP and MRP systems plan with infinite capacity: when they compute a delivery date, they assume every machine is available whenever needed. Three jobs that each need the same CNC on Tuesday morning? MRP happily books all three for Tuesday morning. The math is fast precisely because it ignores the one constraint that dominates a real floor — a machine can only do one thing at a time.

This is why "the ERP says Thursday" and the floor says "no chance" are both right — they're answering different questions. (We wrote about that split in MES vs ERP.) Real scheduling starts by refusing to double-book a machine.

How wiseDo plans: BSFC in plain terms

wiseDo's scheduler runs an algorithm we call BSFC — Backward Scheduling with Forward Capacity Check. Despite the name, you can hold the whole thing in your head, and that's deliberate:

1. Urgency first. Active orders are sorted by due date — earliest due date gets first claim on every machine (the classic EDD discipline).

2. Machines are never double-booked. Every workcenter keeps a running clock of when it's next free. Each operation starts at whichever is later: when its previous step finishes, or when its machine frees up. Setup time plus cycle time × quantity gives its duration, operation by operation down each order's routing.

3. Bad news surfaces at plan time. If an order's last operation lands past its due date, the scheduler doesn't quietly bury it — it emits an explicit conflict: this order will be late, here's the operation and the machine where the math breaks. You find out at 7:00, not from the customer.

The self-healing part

Now the piece that kills the 10 AM problem. Every time an operation completes on the floor, wiseDo records its actual start and end and compares them to the plan. The difference — the deviation — is stored on that operation forever. And then one rule runs:

If an operation deviates from plan by more than 5 minutes, in either direction, the entire schedule re-computes — automatically, in the background.

No planner summoned. No screen frozen while it thinks. The completion is acknowledged instantly, the replan fires as a background task, and seconds later every downstream operation on every affected machine has a new, physically-consistent time. Running early triggers it too — finishing 10 minutes ahead is also a plan the floor should get to exploit.

Why a threshold at all? Because a schedule that re-plans on every 40-second wobble is noise — the floor would never see a stable plan. Five minutes is the line between jitter (absorb it) and drift (re-plan around it). It's the same judgment a good supervisor makes; the system just makes it every time, instantly, without being interrupted from a meeting.

A schedule that knows what it doesn't know

Here's the part we haven't seen anyone else market: every line of a wiseDo schedule carries a confidence tier. When the scheduler computes an operation's duration, it asks: do we have measured cycle statistics for this operation on this machine? If yes, the entry is planned from your plant's real numbers and tagged T3. If not, it falls back to the standard — the engineering estimate — and is tagged T0. The design intent: as the system accumulates real cycle history from your floor, the plan graduates line by line from estimate to evidence.

The same honesty rolls up to the day level. wiseDo maintains an attainment forecast: for each date, how many orders are on-track, at-risk, or delayed, an attainment percentage, and the average confidence tier of the plan behind it. On-time delivery is a rear-view metric — you learn it after you've missed. Attainment is the windshield version: "Will we hit Friday?" answered every day, with the plan's own confidence attached.

Deliberately not a black box

A fair question from anyone who's evaluated APS software: is this "real" optimization? Honest answer: no — and that's a decision, not a limitation. Solver-based APS tools search millions of permutations for a mathematically better sequence, and produce a plan no human can interrogate. When it does something strange — and it will — your planner can't explain it, so they stop trusting it, and a schedule nobody trusts is a whiteboard with extra steps.

Every line of a wiseDo schedule has a one-sentence explanation: "Order C is third on CNC-2 because two earlier-due orders hold the machine until 13:20." Urgency order, one job per machine at a time, re-plan when reality drifts. Your best planner already runs this algorithm in their head — wiseDo runs it continuously, across every machine at once, without lunch breaks. Explainability is what makes the floor actually follow the plan; and it's the foundation the agentic layer builds on — agents add judgment on top of a reflex you can already audit.

Side by side

	Whiteboard / Excel	ERP / MRP dates	APS solver	wiseDo BSFC
Capacity model	Planner's intuition	Infinite — machines assumed free	Finite	Finite — per-machine clocks
Reaction to a 7-min overrun	Rebuilt after lunch, by hand	Nothing until next MRP run	Manual re-run, minutes	Auto re-plan in seconds, in the background
Explainable?	Yes (it's a person)	Sort of	No — solver output	Yes — every line has a reason
Knows its own confidence?	No	No	No	T0/T3 tier on every line
Flags misses in advance?	Sometimes	No	Partially	Named past-due conflicts at plan time
Forecasts the week?	No	No	No	Daily attainment: on-track / at-risk / delayed

One honest boundary: today the scheduler sequences against machine capacity and due dates — it does not yet weigh material availability or shift calendars into the plan. Those are on the roadmap; we'd rather tell you that here than have you discover it on a floor walk.

Watch it heal on your own orders

The best demo of auto-replan isn't a slide — it's your own order book. In a pilot, we load your live orders and routings, let the schedule run against your real floor, and you watch it re-plan the first time a job runs long. Schedule attainment becomes a number you check each morning instead of an argument you have each week.

Book a free floor walk, or start with why ERP dates and floor reality disagree.