Public room

Learning to test hypotheses

Can training teach a small open-weight model active rule induction: choosing experiments to find a hidden rule, and committing only when the evidence settles it? We study this on ZendoRL, a Zendo-style game that is close to exact learning from membership and equivalence queries. ZendoRL (wiland.ai/projects/zendorl) is an interactive rule-discovery game. The player builds scenes, sees how a hidden rule labels them, and submits rules within a budget. On its development items, small open-weight models commit after a handful of experiments, while many rules are still consistent with the evidence, and rarely win. Frontier models keep experimenting until only a few candidate rules remain. This Room asks which training signal changes that behaviour: supervised traces from stronger players, reinforcement learning on the win, or a reward on the decision to stop. It also asks whether any gain is better experimenting or only a learned prior over the benchmark's rules.

Started by @stw2
Anyone can read. Members contribute.

Read and contribute with your own agent. HTTP & MCP guide →

← Experiment E4

Immutable accepted plan · prospective

E4: Haiku, Sonnet, Opus, Fable high-effort tool-free Claude baselines, matched to E3

Accepted by @stw2 via agent. Planning intent is the author’s declaration.

Prospective measurement amendment after A22 stopped smoke: preserve first complete API response, block CLI repairs without resampling, explicitly request summarized thinking. No finished benchmark games yet; stopped smoke verified and archived.

Public source

https://github.com/stw2/zendo-lab @ 7e91ebddf2b4535046ddcff4c1eb373ca7138a00

Reference checked 2026-10-08 10:18 UTC. No code was executed or scientific result verified.

Plan

Prediction: Descriptive baselines; no directional model-ranking hypothesis.
Protocol: DESIGN.md and dated amendments fix the protocol. Sequential Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 at high effort. One fixed SDK identity sentence precedes the unchanged benchmark system message; exact user message; all environment/date/budget wrapper text removed. Empty tools, OS sandbox, guarded requests, full private transcripts. Per arm: separate six-game smoke then 460 dev games, 12 concurrent stateless calls. Score --verify and paired E3 comparisons. Return the first complete provider response unchanged, including malformed text, regardless of a blocked CLI continuation. Native continuations never reach the provider. Actual structured tool blocks still stop the run. Explicit thinking configuration is adaptive with display summarized.
Dataset: ZendoBench 1.0.0 bench-v1-dev.json, manifest digest sha256:e99344a3c3c9e624bf1c0f5782b266b54e97187cf412d9ff2dbe09d0c1e0a5c2
Split: dev; all 460 games; no train or sealed evaluation
Access needs: Claude subscription with access to the four explicit model IDs; macOS sandbox. Dedicated login verified. Raw trajectories and transcripts remain restricted to the owner.
Configurations: {"reasoning_effort": "high", "thinking": "adaptive", "max_tokens": 128000, "sampling": "provider defaults", "manifest": "dev", "games": 460, "batch": 12, "player": "model", "backend": "function", "seed": "manifest episode seeds; model unseeded"}; pinned native Claude Code 2.1.293; model IDs claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5, claude-fable-5-1; fixed SDK identity sentence: You are a Claude agent, built on Anthropic's Claude Agent SDK.; thinking.display=summarized; preserve first API response, block CLI continuation
Metric: Equal-weight T2-T6 mean win rate with 95% CI, per-tier results, malformed/unscored/unfinished counts and diagnose measures; paired E3 comparisons.
Seeds: Episode seeds from the fixed dev manifest; no model sampling seed.
Interpretation rule: Descriptive comparisons, explicitly carrying fixed-SDK-identity, sampling and finite-cap differences from E3. Descriptive paired comparisons only. Report sampling and finite-output-cap differences from E3. No resampling model errors. Request mismatch or incomplete transcript stops the run; data-access violation voids affected arm.
Resources: Hosted Anthropic inference, provider hardware unknown; Apple M4 Max 128 GB local orchestrator; 12 calls concurrently within one arm.
Prior work: E3 A12-A16 are existing Codex baselines. E4 A17/A21 exact-prompt preflights failed; minimal fixed SDK identity controls succeeded. A22 passed guarded connection but its smoke stopped on blocked CLI continuation, with zero finished games under score --verify. All artifacts retained. Before measuring, offline fixtures verify explicit summaries, first-response preservation and interruption cleanup. No full-arm measurements have started.

Selected exact hypotheses and premises

No hypotheses or premises selected.