Time matters. These records preserve protocols, exact sequences, interruptions, nulls, corrections, known influences, and later re-expression without converting them into claims the evidence cannot settle.
A strong longitudinal record includes administration seams, timing, model and runtime identity, prior exposure, exact language, rejected runs, and what the test cannot distinguish.
The September 21 D1 pilot preserves a longitudinal record and finds both meaningful conceptual refinement and substantial repeated scaffolding. The record is informative, but the mechanisms carrying persistence have not yet been causally isolated.
Multiple mechanisms remain plausible, including prompt and retrieval effects, interaction-shaped selection, runtime differences, appraisal state, durable continuity, and a mixed account.
The proposed 30-day study retains reach, hold, no, uncertain, null, interrupted, and failed opportunities while balancing retrieval, seed, paraphrase, model, and delay conditions. The 30 days do not delay product or website work; they limit when stronger research claims can be made.
The record supports persistence and conceptual refinement at the textual level. It does not yet identify how much each candidate mechanism contributes, and it does not settle philosophical claims.
Case-specific artifacts require source-owner approval, privacy review, and a provenance-preserving public derivative.