Build your operation
Everything at once: mixed stakes, environments, information, mission changes and finite specialists. Only capabilities you own are available. There is no single correct policy.
Upgrades for this level
Capital —Calibrated probability
300 crShow a calibrated probability next to the raw model score.
Basic automationRequired
1,500 crWrite rules on the recommendation and calibrated probability, and run whole shifts at once.
Uncertainty intervals
500 crShow the lower and upper bound around every calibrated probability.
Uncertainty-aware rules
1,000 crRules can use the lower bound, upper bound and interval width.
Information-aware rules
1,500 crRules can buy a scan, forecast or diagnostic before deciding.
WHY? factual explanations
700 crSee which conditions push the calibrated probability up or down, each with its own uncertainty range.
Alternative explanations
800 crSee how the probability and its range would change under other conditions, and carry out the ones that are real actions.
Alternative-action planner
1,500 crRules can carry out the most promising real change before deciding.
Reject-aware routing
1,000 crSee which missions the assistant rejects for routine automation, and send missions to specialists.
Extra specialists
1,500 cr+15 specialist slots per sandbox shift.
Capacity- and stakes-aware rules
800 crRules can see specialist slots left and the mission type.
Ensured explanations
700 crMark alternatives whose uncertainty is no wider than the current prediction's, and filter to those.
Guarded explanations
700 crFilter explanation candidates whose hypothetical situation is poorly supported by the reference data.
Per-environment calibration
1,000 crCalibrate within each operating environment (Mondrian) and show the environment. Rules can use it too.