pki.sgit.ai / documents / guess-the-agent

Guess The Agent: The Tree Asks Cheap Questions, Guesses Your Grant, And The Product Is The Gap Between What You Predicted And What It Found

TypeDev brief — eighth of 4 September; the question set is the part most likely to be wrong Versionv0.33.64 Date4 September 2026 AuthorDinis Cruz (project lead) and collaborators LicenceCC BY 4.0 Sourceraw markdown · view on GitHub

Summary

A guessing game whose real output is a measurement. It inverts the burden: instead of asking a person to enumerate what their agent can do, it asks cheap questions with obvious answers and infers the profile by ordinary decision-tree induction over a belief, choosing the question that splits the remaining space most evenly and never eliminating on one answer. The tree is public and forkable, because a wrong branch is a contribution and a tree nobody can inspect is a fortune teller. Two modes are two instruments: deterministic by default, a model as fallback, never a model in front of an arithmetic step. The player predicts before the reveal, and the gap between prediction and computation is the finding, the third instance in a week of a belief, a measurement and the distance between them. The naming collision is caught: the game measures a prediction gap and never a surprise. Built here as /guess/, steps 1 to 5.

Key concepts

Key ideas

On this site

Adds /guess/ (index.html, app.js, tree.json, selftest.json), admin/build/gen_guess.py with its self-test gate, and the bench entry.

Read the document

📄 Original document · v0.33.64 · 4 September 2026 · rendered from the raw markdown (the source of truth)