RGM — Relational Space / H1a

Can derived context improve retrieval?

Can contexts derived from record memberships improve evidence retrieval over fixed contexts? This RGM entry covers the Relational Space / ERM H1a experiment line. The ERM prototype combining patterns and statements has a separate entry.

A minimal question within a broader architecture

H1a is designed to test the minimal model: source records, anonymous memberships and derived contexts. Role ports, paths, facets and the broader architecture depend on separate hypotheses. A minimal retrieval result cannot validate them all.

The primary comparison is offline-derived context against fixed-schema context. Null-model and neighbourhood controls, stability, interpretability and cost also matter. One retrieval score does not fulfil the whole research claim.

Calibration and decision are separate

The reviewed v0.5 amendment changes the null-model mixing adequacy rule, not the primary hypothesis or datasets. Calibration reports remain distinct from decision-run outcomes.

It also records a cross-platform reproducibility risk: some selected E_off structures on Mac differed from the reference Linux structures. A reference-machine gate is therefore a substantive experimental control.

The scope of this entry

This page describes design and calibration, not a final H1a superiority result. Deriving contexts from anonymous memberships alone does not establish learned meaning or completion of the broader architecture.

Source note

Sources reviewed on 14 October 2026: Relational Space constitution and reference architecture v0.4, H1a preregistration v0.4, v0.5 amendment and decision record. No experiment was run, and no raw conversations or decision-split inputs were opened for this entry.