A resource planning system that carries a result rather than a task runs the same six stages continuously: observe, interpret, plan, act, reconcile, adapt. Most systems sold today run four of them and stop.
No stage here is new. Reporting observed, analytics interpreted, planning engines planned, and workflow acted, decades before any of this was called autonomous. What is new is the requirement that one system runs all six without a person carrying the result between them, on the clock the business actually moves at, not on a weekly batch.
That is why the stages are worth naming individually. A system is not judged on whether it has an observe capability. It is judged on where its loop breaks, and the break is usually in the same two places: between act and reconcile, where a system asserts an outcome it never proved, and between reconcile and adapt, where a system records what went wrong and changes nothing.
Stage 1
Take in transactions, events, resource states, constraints and outside signals as they happen.
A goods receipt, a bank line, a supplier's shipment notice and a change to a customer's forecast all land as events, not as a nightly file.
Stage 2
Work out what those signals mean for this business: its policies, its history, its commitments.
A short delivery is read against the contract's tolerance, the customer's promised date and the plant's current cover, not against a general rule about short deliveries.
Stage 3
Decide what should happen next, given dependencies, policies, objectives and constraints.
A projected shortage produces a plan that weighs existing cover, open orders, approved suppliers and the cash it will consume.
Stage 4
Execute the permitted actions, across every resource the plan touches.
The order is placed, the commitment recorded, the receiving expectation created and the cash forecast updated, by the system, under its granted limit.
Stage 5
Confirm the intended outcome actually occurred, record the consequence, and prove the record against the world outside the system.
Goods receipt, invoice and purchase order are matched, the variance is posted, the payment is scheduled, and the bank line is tied back to it.
Stage 6
Re-plan as conditions move, learn from what the reconciliation showed, and escalate what needs judgement.
A late shipment changes the plan for the affected orders within the hour, and the exception that cannot be absorbed reaches a named human with its evidence attached.
A line ends. A loop does not, and the difference is not a drawing convention. What reconciliation produces is evidence: the order landed short, the payment cleared four days late, the promised date is now at risk. That evidence has one consumer, which is the next plan. A system that reconciles into a report nobody reads has converted a fact about the world into a fact about a screen.
Adapt is where the evidence is spent. It re-plans the orders the late shipment touched, it moves the threshold that produced an avoidable exception, and it escalates what it cannot absorb to a named person with the evidence attached. The output of adapt is a changed position, which is what the next observation reads. The cycle closes because the last stage writes what the first stage sees.
Four of the six stages are the comfortable ones. Observing, interpreting, planning and acting can each be demonstrated, timed and sold, and a system that does all four is a capable piece of automation. It is also a system whose last recorded belief about the world is its own intention. The output is a recommendation queue that a person rekeys. Nothing has happened until they do, and the system's authority is imaginary.
The two stages after it are where responsibility for a result actually sits, and they are the ones a demonstration rarely reaches, because both are only visible when something has gone wrong. A task is finished when the action is taken. A result is finished when the consequence has been recorded, proved against the world, and fed back into what the system does next. That distinction is why the maturity model grades on who carries the consequence rather than on what the system can do.
Stage five could have been called verification, and calling it that would have cost the model most of its meaning. Verify is a quality-assurance word. It asks whether the thing the system did matches what the system was asked to do, and it is satisfied by a check the system runs against its own record.
Reconcile is the word an accountant, a controller and a materials planner have each used every month for thirty years, and they all mean the same operation by it: take what this system believes, take what something outside this system says, and account for every difference between them. It carries the two things verify omits. The consequence is recorded, in the ledger and the stock position, not in a log. And the record is proved against the world outside the system, by the bank statement, the supplier invoice and the count on the shelf.
The vocabulary was already correct, so the model borrows it instead of inventing a term. Confirm the intended outcome actually occurred, record the consequence, and prove the record against the world outside the system. A system that cannot do that has no basis for the claim that it owns any outcome at all, which is why the stage keeps the older and more demanding word.
Where a loop breaks is answerable about a system somebody can demonstrate.
Take the ARP test