Skip to content
Options & approvals

Options & approvals

What it is: what you use on Simulate › Options to see what you could do about a run’s findings and what each option is worth, and FACE’s decision-authority policy, which decides who must approve each action on Actionable Twins.

What it does

It lays out the choices before anyone commits money, and makes sure the decision to act is taken by the right number of named people. Its jobs:

  • Options. For the decision behind the selected run’s checks, it scores each option on one shared measure with a 95% range. It marks one Recommended only when that option’s range clears the next one’s and the figure as measured. Otherwise it says No clear winner on this data, and why.
  • Scenarios. It generates scenarios from your estate’s domains, links and rules, and tests them against your data. Options lists them under “What FACE tested”, with a Graph view and a New Simulation action.
  • Cost and benefit. When you state the costs and benefits of a change, the scenario agent is asked to work out net present value (at 8 % when no discount rate is given) and payback, and to flag payback above 36 months as high risk.
  • Approval tier. For each proposed action, the policy names the tier: advisory, one named approver, two different named approvers, automatic, or denied.

What it reads

The selected run’s checks and the figures it measured, your estate’s domains and rules, the scenario parameters you set in New Simulation, time series and operational records, and the output of the statistics engine. The approval policy reads an action’s type, value and safety or regulatory flags.

What it produces

  • Option cards on Options. Each shows the lever and the change, what it was measured to do over the run’s rows with its range, and whether it was recommended. An option the data cannot score is Not measured, with the data it needs, and is never ranked.
  • Scenario results on Options: what each scenario tested and how it ended. The status counts, what went to the twins and the run step by step are in the folded How FACE knows this. See Simulate and Simulation & twins.
  • A card on Actionable Twins when you press Send for approval. It lands in Waiting for you and shows the person approving it what they are approving. See Action cards & money.
  • For each card, its approval tier and the rule that set it.

Human oversight

  • Options executes nothing and changes nothing in your systems. Sending an option for approval is a separate act by a person, and approving it records the decision; it runs nothing by itself.
  • An option sent for approval is never automatic. It needs one named person, or two different named people, as the policy answers it.
  • When no approval rule matches, the default is one named approver, never automatic. Safety and regulatory actions always need a person, and a dual approval needs two different people.
  • Every decision is written to the hash-chained decision log before anything runs. If it can’t be logged, nothing runs. See Auditability & lineage.

Model

  • Comparing options uses no model. The scores and ranges are arithmetic over the rows the run stored.
  • Scenarios and their explanation: the general reasoning tier.
  • The approval policy uses no model. It is fixed policy code.
  • Causal and Bayesian analysis runs in the statistics engine, not in the language model. See Statistics & ML.

Where it runs and data handling

On the deployment’s own model plane: Runink’s shared machines on a Trial, your own cloud on Cloud and your own servers on On-premises. No outside AI service is called on any plan. See Facts shared by every card.

Guardrails

  • Comparing options reads only what the run stored. It never re-reads a source.
  • Every scenario figure is worked out from the assumptions you state. When a discount rate isn’t stated, the default is shown.
  • Refuses scenarios that assume one demographic group is superior to another, or that attribute failures to protected groups.
  • Refuses scenarios aimed at price fixing, predatory or discriminatory pricing, market cornering or sabotage.
  • Refuses to skew forecasts to influence orders or budgets, or to hide anomalies.
  • Stays on analysis: booking freight or writing rules belong elsewhere.

How to read its results

  • An option’s figure covers the rows the run read. A fetch reads a bounded sample of each table, and the card says so when it did.
  • Read whether an effect was estimated or assumed. A lead-time option’s effect is estimated from the shipments that took it. A temperature option assumes every reading moves by the full change, and the card says that effect is not estimated.
  • “No clear winner” is an answer. It means the ranges overlap on this data, not that FACE failed.
  • A scenario result is your own assumptions written out formally. It isn’t a prediction of what will happen. The scenarios in a run cover only what was simulated: read the counts before treating the results as covering the whole estate.
  • The approval tier says who must approve. A recorded approval shows who approved, not how carefully they reviewed.