Next Research Decision

Status: proposal

Candidate A — Human Baseline Procedural Extraction

• expected information gain: high
• hypotheses affected: H002, H003, H015, H017, H018
• falsification value: high
• effort: high
• dependency value: high
• publication value: very_high
• urgency: high

Candidate B — Negative Non-Isomorphic Controls

• expected information gain: very_high
• hypotheses affected: H002, H013, H016, H017, H018
• falsification value: very_high
• effort: medium
• dependency value: very_high
• publication value: very_high
• urgency: very_high

Candidate C — Model-Version Stability

• expected information gain: medium
• hypotheses affected: H003, H015
• falsification value: medium
• effort: medium
• dependency value: medium
• publication value: medium
• urgency: medium

Candidate D — Additional Procedural Families

• expected information gain: high
• hypotheses affected: H003, H017, H018
• falsification value: high
• effort: high
• dependency value: medium
• publication value: high
• urgency: medium

Candidate E — Recognition-Control Replication

• expected information gain: high
• hypotheses affected: H013, H016
• falsification value: high
• effort: medium
• dependency value: high
• publication value: high
• urgency: high

Recommended Candidate

Candidate B — Negative Non-Isomorphic Controls

Rationale

• The largest current interpretive risk is false structural equivalence.
• EXP-003 produced the strongest candidate theoretical signal, so the next test should be adversarial rather than confirmatory.
• Negative controls would directly test whether the instrument incorrectly reports stable structure when only superficial similarity is present.
• This decision also informs H013, H016, H017, and H018 at once.

Alternatives Considered

• Human baseline remains the strongest follow-on for publication quality, but it is more expensive and should benefit from a sharper adversarial model-side test first.
• Recognition-control replication remains important, but it narrows one threat rather than addressing false equivalence directly.

Not Authorized Automatically

• Do not create ECR-000004 from this proposal alone.