Experiments

The registry groups controlled experiments by design. A design fixes the protocol: the research question, the architecture, and the measurement. A run follows that protocol with a pinned manifest of models and stack versions. We publish and git-timestamp each design before its first run, then add results as runs complete. Designs and results are recorded only in this registry. The blog comments on the research as it progresses, but a post never replaces a registry card.

Design
Sustainability — whether an agent can earn enough to pay for inference and gas
The sustainability family follows knowledge delivery and tests agents that pay their own way. Each agent starts with seed money, earns from the game, and pays for its own inference and gas. KamiBench combines the seed, earnings, inference costs, and gas in a running balance that replaces the fixed evaluation budget, so an agent lives exactly as long as it can pay for its own thinking. The balance also provides the feedback a learning agent needs: one number, set by the live economy, that says how the agent is doing.
Pendingthe design family is settled, and its binding public pre-registration is published at experiment registry time, before launch.
Design
Knowledge delivery — how the game's knowledge reaches the agent
The design keeps the world, tools, and objective fixed while changing only how the game's knowledge reaches the agent: through a folder to read, a search tool, facts pushed inside tool results, or a plan file. The design asks which delivery method turns a capable model into a player that understands the game.
- In progressRun 1 — knowledge delivery, wave 1Experiment 006 · launched 2026-08-17; four arms playing. Results are added as arms stop.
Design
Stack validation — whether the stack supports autonomous play
The stack-validation series used five bounded runs to prove that the stack holds up under real autonomous play before anything open-ended runs on it. The runs tested the environment interface, the scaffold, the telemetry, and the cost accounting using the same box each time: $10 of inference, seven days, and three cheap models. Complete: the stack is solid. The agents still do not understand the game.
- CompleteRun 5 — verification runExperiment 005 · ran 2026-08-07 → 2026-08-14; the exit test passed and the series is closed. The full dataset is public.
- CompleteRun 4 — perception parity re-runExperiment 004 · ran 2026-07-28 → 2026-08-05; the full dataset is public.
- AbortedRun 3 — perception parityExperiment 003 · launched 2026-07-27 and retired at the first scheduled monitor pass, about 17 hours later. A defect in our own run tooling, outside the published stack, had degraded the run's instructions from the first session. The design carries over unchanged to Run 4.
- CompleteRun 2 — iterated stackExperiment 002 · ran 2026-07-19 → 2026-07-26; the full dataset is public.
- CompleteRun 1 — baseline stackExperiment 001 · ran 2026-07-10 → 2026-07-17; the full dataset is public.

Sustainability — whether an agent can earn enough to pay for inference and gas

Knowledge delivery — how the game's knowledge reaches the agent

Stack validation — whether the stack supports autonomous play