Rounds 1–4 · Controlled Formation
Fixed source packets established repeated synchronized interaction under highly constrained evidence selection.
Stage 1 · Controlled Runtime Cohort
7 synchronized rounds complete · baseline established
Eight fixed Wren–Ada pairs began from matched role constitutions, isolated pair histories, a frozen corpus program, tightly governed evidence access, and the same Wren → Ada → Wren → Ada exchange.
Participant environment
Persistent identities, bounded invocations
Wren and Ada were persistent research identities, not OpenClaw agents. Their admitted pair histories were preserved and reconstructed into authorized contexts for tightly bounded direct OpenAI Responses API calls. They had no native persistent agent runtime, workspace, or continuing session.
Two formation regimes
From controlled selection to bounded inquiry
Rounds 5–7 · Exploratory Formation
Wren gained bounded local-corpus retrieval and materially greater freedom to decide what to ask, inspect, and retain. Ada remained downstream of Wren-selected material.
Why seven rounds?
Seven became the natural transition point.
Seven rounds was not a predetermined statistical threshold. The first four rounds established the tightly controlled formation regime. Rounds five through seven then introduced bounded exploratory retrieval, allowing each Wren substantially more freedom to decide what to investigate while preserving the same pair identities, corpus, and scientific exchange.
By the end of Round 7, the controlled-runtime cohort had demonstrated that the apparatus could sustain repeated pair interaction, preserve isolated histories, support increasingly autonomous inquiry, and produce measurable pair-specific differentiation. Rather than continue adding observations under the same runtime, the Lab is holding those seven rounds fixed as a controlled baseline and moving to the next question: what changes when the same experiment is repeated by persistent autonomous agents?
Scientific status
Observed in Controlled Runtime
Some persistent pairs showed suggestive recurring epistemic habits across unrelated domains, while many candidate patterns failed to replicate. The controlled-runtime stage did not establish that these differences were relational in origin.
Persistent Agent comparison: not yet available. No relationship finding is claimed.
Read the controlled-runtime evidence
Round 1 · Round 2 · Round 3 · Round 4 · Round 5 · Round 6 · Round 7