HUMAN IN THE LOOP

The gate

Asking a person about everything is the same failure as asking them about nothing: both end in an ignored screen. BioFlow decides how much of a human's attention each proposal has earned.

MODE · CONFIRM

Ask, and wait

The action does not run until someone answers. Used where a wrong action is expensive or irreversible: dosing, dispatch, disclosure, payment.

MODE · NOTIFY

Act, and tell

The action runs immediately and a person is informed with a one-tap undo window. Used where delay costs more than a wrong call.

MODE · AUTO

Act, and log

No interruption. Reserved for reversible, low-stakes actions on rules with a long clean record, and still fully written to the ledger.

MODE · DEFER

Hold for the right person

The proposal waits for a named role or shift rather than whoever is nearest. Queued, escalated on timeout, never silently discarded.

TRY THE SCORING MODEL

Move the sliders and watch the mode change

Trust is how well this rule has performed on this population. Cost of waiting is what a delay actually costs. This is the same arithmetic that runs in the gateway, not a marketing approximation of it.

Built from historical precision, how often people have overridden it, and how far the current reading sits outside this person's own baseline.
How much harm accumulates per minute of delay, and whether the action can be undone once it has run.
Irreversible actions can never reach Auto, whatever the other two scores say.
SELECTED MODEConfirm

Middling trust and middling urgency: the case a gate exists for. A person decides, and their answer trains the score.

window 45 s · on expiry: cancel

One sentence, then the numbers

The proposal opens with what will happen and why. Confidence, traces and rule provenance sit one tap below.

A visible clock

The responder always knows how long they have and what happens when it runs out.

Refusal is data

Every hold feeds back into the trust score. A rule people keep refusing loses the right to interrupt them.

Start with one ward, one shift, or one squad

A pilot runs four to sixteen weeks depending on deployment model. You bring the population and the responders. We bring devices, gateways and the first set of rules, and we measure override rate and time-to-answer from day one.

Request a pilotSee how it works