Empirical Research Program
Status: Public research-method node. This page describes how the current theory program is intended to be tested; it is not a list of already validated laws.
The Core Rule
A concept or theory is useful only if it can be placed in a comparison where a plausible alternative can win.
The current program therefore prefers:
- pre-specified claims;
- matched interventions;
- strong null or reducer families;
- held-out evaluation;
- explicit undefined cases;
- and stopping or downgrade rules when a decisive comparison fails.
This is more important than collecting additional examples that can be described retrospectively with the theory's vocabulary.
Connection Theory
Connection Theory v4.6 introduces its foundational concepts through explicit failure boundaries and a termination check: if an explanatory gap can be closed by extending the existing concept, a new conceptual object should not be introduced merely for rhetorical convenience.
Its current empirical questions include, among others:
- whether active system-level properties depend on operating organization rather than static inventories alone;
- whether capability-relevant persistence state can be localized under controlled intervention;
- whether update order produces prospectively predictable non-commutative effects in suitable systems;
- whether continuity judgments track qualifying causal provenance rather than similarity alone;
- and where organization-conditioned evidence defeats simple self-correction assumptions.
The four-stage formation sequence itself remains a proposed explanatory scaffold rather than a completed validation result.
Flow Shaping Principle
FSP v3.7.0 supplies a more tightly specified canonical test for an instrumentable artificial adaptive system.
The primary empirical wager is not that feedback exists or that activity changes a system. It is that independently measured veridical consequence–configuration correspondence available to the attribution stage should prospectively predict correspondence-correct persistent shaping, including transfer across a pre-specified manipulation pair.
A failed cross-manipulation transfer can reject the current single-quantity FSP claim even if simpler within-class relationships look orderly.
Evolutionary Relativity
ER v3.6.1 studies a different relation: an adaptive target traverses an organizational path while an observer reconstructs that path from finite evidence.
The canonical continual-learning test requires, among other gates:
- genuine history dependence;
- a target not designed around the proposed path summary;
- constructible and sufficiently separable trajectory banks;
- raw-channel adequacy and probe fairness;
- frozen observer representations;
- and held-out comparisons against rival summaries.
ER is supported only if its registered path-sensitive predictions beat eligible rivals under the pre-specified procedure.
Effective System Capability and Collective-Function Claims
From Microscopic Capability to Collective Function v0.5 uses collective function as a narrower attribution-testing term within the broader question of Effective System Capability. It adds an explicit evidential ladder for claims that richer interacting units produce a new useful system-level function:
- mechanism availability;
- causal collective dynamical difference;
- response-class expansion relative to a tested lower family;
- task-specific value after cost accounting;
- collective irreducibility relative to a stated reducer family.
A stronger claim should be downgraded when a stronger reducer reproduces the relevant observable.
General Stopping Rule
A failed decisive comparison should normally reduce the claim rather than trigger a bespoke redesign whose only purpose is to restore a positive result.
The research program treats claim downgrade as a result, not as a failure of the research process.