A field can hold one value. When one agent wants to raise it and another wants to lower it, arbitration takes both cases, measures the distance between them and settles the matter inside the band your team set.
Five steps: claim, assemble, measure, rank or refer, record.
Each run produces recommendations per agent. Before anything is carried out, the engine groups them by the field they would change: the safety stock of one material at one plant, the date on one schedule line, the open quantity on one order line.
Where one field carries two proposals, arbitration assembles both cases whole: the objective each agent serves, the conditions that matched, the confidence, the value at stake and the weight the agent carries for that field.
It then measures the gap. The gap is the distance between the two proposed values, read against the band your team set around the current value. A small gap is a matter of degree; a wide gap is a matter of judgement.
Inside the band the weights rank the two, the higher one is applied and the other is recorded as set aside. Beyond the band both cases go to the named person who owns that field, side by side on one screen.
Two agents can both be right; the field still holds one value.
| Safety stock agent | Excess stock agent | |
|---|---|---|
| Objective it serves | Hold the agreed service level on this material | Release working capital held as excess cover |
| What it reads | Demand variability, lead time variability, service level | Stock on hand, stock on order, forward demand |
| What it proposes | Raise safety stock from 1,100 to 1,400 pieces | Lower safety stock from 1,100 to 900 pieces |
| Its basis | Demand variability rose 34% since July; two late deliveries in August | Cover runs at 11.4 weeks against a policy of 8 |
| Its confidence | 0.88, from 14 months of demand history | 0.91, from stock and demand read this morning |
| Its weight on this field | 0.62, set by the planning lead for class A materials | 0.55, set by the planning lead for class A materials |
| What it is worth | Service level held at 98% on a line worth $2.1M a year | $46,000 of working capital released |
One contested field in the showcase estate, between two designed roles from the agent blueprint, shown as they would argue it. Showcase estate. blueprint
Both cases stay whole. A person reading this table sees two sound arguments from two different objectives, which is what a contested field usually holds. Arbitration settles which one is applied today, and the record keeps the other so the next review can read it.
Four settings your team owns: the band, the weights, the floor and the cooling period.
| Setting | What it controls | Who sets it | In the showcase estate |
|---|---|---|---|
| The band | How far apart two proposals may be before a person decides | The owner of the field, per object and class | 15% of the current value for class A materials |
| The weight | Which objective ranks higher inside the band | The owner of the field, per agent and class | Safety stock 0.62 · excess stock 0.55 for class A |
| The floor | The lowest value a proposal may reach, whichever agent wins | The planning lead, per material or class | 800 pieces for 7741-X, set from the service agreement |
| The cooling period | How long a field stays settled after a decision | The owner of the field | 14 days, so a value settles before it is revisited |
A band is a statement about judgement. Inside it the two agents differ by an amount your team is content to settle by rule; beyond it the difference is large enough that a person should see both cases.
Weights say which objective leads when the difference is small. A team protecting service sets the safety stock weight higher; a team releasing cash sets the excess weight higher.
Both settings carry a version history, so a change to a weight is as visible as a change to a rule. An auditor reading a decision sees the band and the weights that were in force that day.
A field that reaches arbitration repeatedly is worth a look. The manager report counts contested fields per agent and per period, which usually points at a policy that wants an edit.
Take one case, take the other, or set a value of your own; every route records a reason.
Beyond the band, the case goes to the person named on that field, with both proposals, both bases and both values at stake on one screen. The clock and the escalation route work as they do for any approval.
That person has three choices: take one of the two proposals, set a value of their own, or leave the current value in place. Each choice records a reason.
A value of their own is common, and useful. J. Ridley set 1,250 pieces, between the two proposals, and the reason went on the record beside both cases.
The learning loop reads that decision. Where a person settles the same contest the same way three times, Juno proposes a weight change or a rule edit for approval.
Showcase estate. The safety stock and excess stock roles are designed, yet to be built; the scene shows how arbitration would settle them.
On the Monday run both agents reached the safety stock of 7741-X at plant 1210. The safety stock agent asked for 1,400 pieces after a rise in demand variability; the excess stock agent asked for 900, because cover was running at 11.4 weeks.
The gap was 500 pieces on a current value of 1,100, which is beyond the 15% band, so both cases went to J. Ridley. At 09:40 the value was set at 1,250 with the reason recorded. The cancellation agent carried on with the open lines, because that field was claimed once.
The five routes an approved action can travel, and how Juno hands work to the system that will carry it out.
They are ranked together. The widest gap is measured against the band, so a case that holds one outlying proposal reaches a person while the others settle by weight.
It is kept whole on the record, with its basis and its value at stake. The outcome check reads it later, and the manager report counts how often each agent is set aside.
The person who owns that field, usually the planning lead for a material class. Both settings carry a version history, and a change is visible to an auditor.
It runs between evaluation and action on the fields that carry two proposals, which is a small share of a run. Everything claimed once carries on as usual.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.