TOKNITH
Platform

What the model guarantees.

Everything on this page runs in the product today and can be tested in a demo. Where a guarantee has a limit, the limit is stated.

Replay to the identical state.

Every input is recorded. Rewind restores a saved state, replays the recorded inputs, and lands on the same fingerprint of the entire model state, bit for bit, on the same build.

Same inputs, same run.

Same build, same seed, same inputs, same result. Evaluating the same moment twice gives the same forecast and the same advice.

The why chain.

Open a late or at-risk order and read why in one sentence, in the plant's own numbers: heads faulted and since when, the line's rate against nominal, jam minutes, changeovers and cleaning, what was inserted ahead, and pack-chain blocks. Each number says whether it was measured, nominal, or planned.

The promise line.

Load the shift's orders and the model plans them. The first plan is kept as the promise, and every order shows what was promised, what is planned now, and what actually ran. Replans and moved orders are counted against it.

Three early warnings.

The plan says which orders it cannot fit the moment it replans. A running order's measured rate says it will be late minutes after a fault. The forecast says what the plant will do to the horizon. Each warning says which of the three it came from.

Fixes are work.

The Advisor proposes a repair. Once a person accepts it, an operator with the right skill walks to the equipment, works the repair time, and clears the fault. The record says who fixed what, and when.

Changeovers cost what they cost.

Changeover times come from a matrix of product pairs. A change away from an allergen product runs the clean-in-place, and the line waits until it is done. Liquid volume is conserved through every phase, and the tests check it.

The plant is a versioned document.

Diff any two versions of a model and restore either one. Every accepted layout change becomes a new version with a name on it.

Change the plant while it runs.

Edit a parameter or add a unit, validate it against a trial build, and apply it live. Everything that still exists keeps its state.

Nothing from a feed is guessed at.

Duplicates are dropped. Anything the model does not understand is set aside with a reason. Late arrivals are refused and logged.

Events at the microsecond.

Time is an integer count of microseconds. A crossing, like a bottle reaching its fill level, is found in closed form rather than at the end of a step, and only the units near a crossing run at the fine rate.

Scrub to any second.

One timeline drives every view. Scrub to any second of an eight-hour run and the 2D, 3D, schedule, and inspect views agree.

Measured on our build machine

MeasureResult
Eight hours of the reference cell422 seconds of wall time, 68 times real time
Events recorded over those eight hours191,622
Restore points written285,945
Storage for the run560 MB
Time spent at the fine step rate0.07 %
Forecasting and the Advisor in four worker processes50.8 times real time, over a ten-minute window

Replay identity holds on the same build and platform. Identity across different platforms is not claimed.

One clock. One model. Many futures. One reality.

Model the operation. Rehearse the change. Connect it to reality.

Request a Demo