The Advisor proposes. It never applies.
Every proposal goes to a named person. Accepting applies the action as that person, through your own systems. Rejecting changes nothing at all.
The constraint comes first.
Every proposal goes to a named person through the approval gate, which is the single path by which an action or a change to the layout takes effect.
Accepting applies the action as that person. The commit reaches your bed board, your workforce system, or your warehouse control system through your own adapter. The twin never writes into those systems privately.
Rejecting changes nothing. The clock, the plant, and the layout stay exactly as they were, and the rejected action rests for a cooldown so it is not proposed again immediately.
Every proposal, acceptance, and rejection is an audit record carrying the identity of the person who decided.
The path a proposal takes
- Exception
- Candidates
- Score
- Propose
- A person accepts
- Your system
- Audit
Reject at the fifth step and nothing downstream happens.
What it watches.
The Advisor reads the connected model's own forecast, a short future run from the current state on a fixed cadence. Its triggers are that forecast's exceptions, plus two that the plan and the line hand it earlier.
- Exceptions raised by the forecast itself.
- An order the plan already places past its due date, the moment a replan says so.
- A running order whose measured rate says it will be late, before the plan or the forecast has reacted.
On the reference line, that last trigger fired ten minutes after a fault, before anything else did.
How it decides.
It does not guess. Each candidate action is run as a branch of the same model, from the same moment, and scored as a weighted sum of the differences at the horizon against the do-nothing baseline.
Only actions that are legal on that plant at that moment are considered. Only actions that score better than doing nothing are proposed, best first, and never more than a small number are open at once.
Two evaluations of the same moment give the same proposals, at any playback speed.
Scored against doing nothing
Each candidate is judged on the difference it makes at the horizon, not on its own merits.
- Late orders
- Minutes late
- On-time percentage
- Throughput
- Rejects
- Missed truck slots
The catalogue today
Every one of these is an action an operator could have sent by hand. Nothing in the catalogue changes a product, a recipe target, or a safety setting.
- Repair the faulted heads
- Expedite an order
- Re-sequence by campaign
- Re-sequence by due date
- Run the line faster, within its design band
- Run the line slower, within its design band
- Dispatch the nearest agent
- Send the truck now
The evidence.
Every proposal carries its reason chain, read backwards from the log in the plant's own numbers.
- Which order, and how late it is
- The batch's slip
- The line's rate against nominal
- Which machines are faulted, and since when
- Jam minutes
- The rush order inserted ahead
- The pack-chain blocks
Nothing in the chain is inferred. If the log cannot back a number, the sentence says less rather than more.
What the Advisor is not.
- Not closed-loop control. It sets no setpoints automatically, and that is out of scope today.
- Not a model that predicts. The forecast is the simulation itself.
- Not a black box. What runs in production is heuristics plus branch runs of the same model.
- Not sold on its own. It requires the connected model and its divergence checks, because a forecast without a residual is theatre by week six.
In healthcare, the verbs the pack may never use.
- Change acuity
- Choose treatment
- Auto-discharge
- Move a patient without documentation
These are refusals built into the pack, not policy notes written around it.
One measured example.
On the reference line, with half the filler heads faulted, the accepted repair proposal brought a late order from 34 minutes late to 12.
That is one line, one fault, and one accepted proposal. Your operation's numbers will be your own.
One clock. One model. Many futures. One reality.
Model the operation. Rehearse the change. Connect it to reality.
Request a Demo