design-research-experiment

name	design-research-experiment
description	Design or preregister a Zugzwang experiment with causal hypotheses, controls, protocol coordinates, datasets, metrics, power, budgets, promotion gates, and claim limits.

Design research experiment

Workflow

1. Formulate one primary causal question.
2. Identify prior work and the exact unresolved gap.
3. Decompose capability: perception, tracking, legal action, policy, value, search, explanation or system reliability.
4. Define conditions with R/H/K, observation profile and budget.
5. Build token/compute-matched controls and negative/placebo conditions.
6. Freeze sampling unit, pairing, randomization, exclusions and dataset hashes.
7. Define primary/secondary outcomes and evaluator provenance.
8. Simulate variance/power or state why pilot is exploratory.
9. Set hard budget, stopping and promotion rules before outcomes.
10. Write claim boundaries and preregistration amendment policy.

Mandatory anti-confounds

• same model snapshot/date where comparison requires it;
• parsing error separate from illegal action and legal blunder;
• engine assistance separated from post-hoc evaluation;
• no transposition-near knowledge leakage;
• SAN/semantic legal list leakage documented;
• full-game promotion based on predeclared local criteria;
• paired openings/colors for match outcomes.

Use EXPERIMENT_CARD_TEMPLATE.md; do not launch paid execution from an informal prompt.