Docs

Every page is also Markdown: add .md to the address. An index for AI assistants is at /llms.txt.

Introduction

Introduction
What Human Handoff is and the 60-second start.
API reference
POST /v1/handoff, GET /v1/handoff/:ticket, GET /v1/versions, errors and rate limits.

Calibrated human judgment
Why a person's number is useful once you know how well it tracks what happens.
The gut-first order
Gut, then head, then heart, then your number.
The biological check
When the gut agrees with the machine, skip the slow path.
Confidence for a human answer
1 minus the spread of gut, head and final, divided by 10.
Updating after the reveal
The pre-reveal final is the answer; one optional Land again sweep is kept apart as updated_read and moved.
Question class and the 30-resolved gate
Every question has a class; a person's reads combine only after 30 resolved forecasts in that class.

Answer types
likelihood, levels, pick, and how Jev's shape maps.
Say it your way: POST /v1/felt
Map a typed felt-sense phrase to a proposed value and range; commit only what the person confirms.

One card with several questions
Fan-out: several questions, one card, answered independently.
Confidence bands
Act, confirm, or hand to a person.
Combining several levels answers
Normalise each, weight in code, keep the raw answers.
Routing between machine and person
One code path for machine and human answers.
Decision practices: stakes, ranges, stop conditions, reviews
matters_in_a_year, surprise ranges, stop conditions with dates, reviews with the outcome hidden, luck reads and imagined failure.

Limits
Where people's judgment is unreliable, and why the machine answer is hidden.