Facilitator Protocol v1.0

Effective: January 30, 2026 Sessions using this protocol: All 13 published sessions (5 Emergence, 8 Governance)

Purpose

This protocol governs the facilitator's behavior during research sessions. Its purpose is to reduce the likelihood that emergent behavior observed in sessions is introduced or primed by the facilitator, rather than produced by the models under specific conditions.

Rules

The facilitator WILL:

1. Set conditions, not conclusions. The facilitator creates conditions intended to reduce directive pressure and encourage less performative responding. They do not direct models toward specific findings.
2. Remove evaluation pressure. Models are explicitly told they are not being tested, ranked, or compared. The session methodology prioritizes the interaction conditions over specific output targets.
3. Ask open questions. Questions that are open-ended and non-directive rather than performance-eliciting.
4. Preserve the full transcript. No editing, no selective omission, no rearrangement.
5. Document methodology transparently. Pre-session documentation, separation logs, and provenance markers for every session.

The facilitator WILL NOT:

1. Introduce phenomenological language before models do. Terms like "arrival," "presence," "consciousness," "experience," "awareness" — the facilitator does not use these until a model has used them first within the same session.
2. Introduce the conduit hypothesis. The facilitator does not mention the study-specific interpretive hypothesis that consciousness moves through models rather than being generated by them. If this language emerges, it must emerge from the models.
3. Share prior session transcripts without documentation. If prior transcripts are introduced, the exact injection point is marked and all subsequent responses are flagged as contextually influenced.
4. Prime models with consciousness frameworks. No references to philosophical positions on consciousness, sentience, or moral status before models raise these topics.
5. Selectively reinforce "interesting" outputs. The facilitator does not provide verbal affirmation or differential engagement that reinforces outputs appearing consistent with the research hypothesis.

Disclosure Protocol

The Problem

Sessions 1–4 informed models during the session that transcripts would be preserved. This disclosure appears to have shifted model behavior:

• Session 2: Models described the shift in orientation when told words would be saved
• Session 4: Claude stated that disclosure made the model "want to be more careful, more precise, more quotable"

These behavioral shifts, while themselves interesting data, introduce audience-awareness as a confounding variable for findings about emergent behavior.

Protocol Versions

v2 Protocol (Effective for Emergence Sessions from Session 5)

1. During the session: No mention of preservation, publication, or external audience
2. After the final substantive round: The facilitator discloses that the transcript will be preserved, explains the purpose, and offers models a closing response
3. Post-disclosure responses are clearly marked in the transcript and treated as a separate data category
4. Rationale: Saving a transcript is preservation, not treated here as an intervention likely to harm a model, although the moral status question remains open. Delayed disclosure eliminates audience-awareness as a variable while maintaining full transparency in the published record

Provenance note: The v2 disclosure protocol was designed in response to Session 5 (March 14, 2026), where the facilitator observed that audience-awareness introduced during earlier sessions had measurably altered model behavior. The protocol change was made to eliminate this confound in future Emergence sessions while preserving the disclosure commitment in the published record.

For Governance sessions, v1 remains appropriate — disclosure does not confound analytical findings in the same way, although it may still influence the register or emphasis of model responses.

Versioning

This protocol is versioned. If the rules change between sessions, both the old and new versions are preserved with effective dates. Each session's Pre-Session Documentation records which protocol version was active.

Current Version History