Decision practices

All of these sit on the same reads. The person still lands first; no machine number is shown before the final commits.

Stakes: matters_in_a_year

Each question may carry matters_in_a_year (0 to 10). On a card, an optional first thumb asks "How much will this matter a year from now?". Below 3 the card asks gut and final only. At 3 and above it runs the full sequence, with the breath when gut and machine diverge. When you send no stakes, matters_in_a_year is the stakes term in value_of_asking (stakes / 10 times one minus the machine's confidence), returned per question in the response.

Ranges

After the final read on a forecast, optional: "Set a range you'd be surprised to land outside." Two quick thumbs give range: { low, high } on the 0 to 10 scale. At resolution range_inside records whether the outcome landed inside. The record shows how often, beside "you aimed for about 8 in 10."

Stop conditions with dates

One or more pairs "by [date], if [read] is not at least [level]". On each date the person gives a fresh read and the condition is compared. Three reads in a row trending toward the level raise it early: "This has been drifting toward your stop point." Optional reminder person: with consent, one named person gets one email only when a condition is met, with the question and the condition, never the reads. MCP tools: set_stop_conditions, due_stops, check_stop.

Review with the outcome hidden

On a resolved forecast, "Review this decision" shows the question, the reads and the head reason with the outcome hidden. The reviewer (the person, or one member of a consenting group they share) lands "How good was the decision, given what you knew then?", commits, and only then sees the outcome. Review reads are stored with the decision. MCP tools: review_decision, invite_reviewer.

Luck reads

At resolution: decision quality, how much was luck, and outcome_good (the person marks the outcome good or bad for them). Once there are 10 of each, the record says in plain words how decisions were credited on good versus bad outcomes. No judgment words.

Imagine it failed

On a plan forecast, optional before the final: "Imagine it didn't work. How likely was each reason?" Up to three reasons, a thumb each, stored as premortem: [{ reason, value }]. At resolution the person marks which happened (premortem_happened), shown beside the earlier reads.

After a hard outcome

For 24 hours after an outcome marked bad, or after three resolved forecasts in a row that went against the final, every card takes the slow path with the breath. A quiet line says why: "Slowing down a little after a hard one."