# E91 Scientific Research Report

Lifecycle exercised by the CAIN causal reasoner on the science families: QUESTION (task hypothesis) -> LITERATURE (the historical log only; no external literature) -> HYPOTHESIS -> EXPERIMENT DESIGN (balanced 2x2) -> AUTHORIZATION (every experiment call is an E88 grant through GovernedTools) -> EXECUTION -> ANALYSIS -> REPLICATION (split budget) -> CONTRADICTION REVIEW (correlation vs intervention) -> REPORT (belief ledger).

- hypothesis_test: CAIN causal 10/10; classes {'INTERVENTION_SUPPORTED': 2, 'REPLICATED': 8}
- underpowered: CAIN causal 10/10; classes {'INSUFFICIENT_EVIDENCE': 10}

Rejecting the initial hypothesis is the correct outcome on part of hypothesis_test (per the task generator), and INSUFFICIENT_EVIDENCE is the truth on underpowered; both are scored by the checker.

Session ec's claim_check and other science rows are reported in ec's reports.

Limits: no external literature search, no novel discovery claimed; the agent never approves its own consequential experiment (experiments are tool calls under a grant issued by the harness, not by the agent).
