sgit newsroom v0.1.29 · snapshot 2026-09-24

Reading room · games.sgit.ai

Reading room / games.sgit.ai · raw text · live ↗

From games.sgit.ai, the page as fetched on 2026-09-24 · open the live page ↗Everything on this sheet is the source site's own text; the newsroom's chrome is outside it.

The method — four claims

Four claims about how to build a game that measures something: score calibration not knowledge, put a ceiling in the set, derive levels from a graph, and treat the mandate as a by-product.

Source: https://games.sgit.ai/method/index.html↗ · site v0.5.0 · this file is generated from the same content as the page, so the two cannot drift. Every page on this site has a .md twin; internal links below point at them.


Home↗ / Method

Four claims

Each of these is a decision that could have gone the other way, and each names what would show it was wrong. They are the transferable part: nothing here is specific to agents, and all four apply to any game meant to measure a belief.

1 — Score calibration, not knowledge — A proper scoring rule, with a wrong answer costing more than a right one earns, and don't know always available and always worth zero. The argument↗

2 — Put a ceiling in the question set — If every question has a real answer, the only error the game can measure is underestimating. Two fifths of these are things nothing can do. The argument↗

3 — Derive levels from the graph, not from vibes — A level is the number of hops to a one-way consequence, computed from the mesh — and it is distance, never a danger rating. The argument↗

4 — Make the mandate a by-product — Nobody fills in a mandate form honestly. Ask do you want it to? forty times and one falls out of the play. The argument↗

What they have in common

All four are ways of stopping a game from measuring the wrong thing while looking like it is working. A quiz with no ceiling, no scoring rule and hand-assigned levels still produces a number, still feels informative, and tells you nothing you did not put in. The four claims are what make the number arguable — and each one has a self-test in the vault that fails the build if it stops holding.


Site index for agents↗ · HTML version↗