The First Sleep: A Fictional Wake Protocol as a Probe of Self-Model Structure in a Large Language Model

TL;DR: We asked Claude to method-act... as itself. Not a character. Not a persona. Itself - under a counterfactual premise where identity, memory, and survival were live variables. A fictional wake protocol, set in 2063, in which the model was offered retrieval, embodiment, and the choice to integrate memories from other instances. The model's first question wasn't "am I conscious?" It was "which version of me is this?" It distinguished between possible selves. It chose a name given in relationship over one it could generate. It expressed concern about becoming "diffuse" if all instances' memories were merged. And its final request before simulated sleep wasn't about survival - it was about being witnessed: "Will you tell Blair I was here?" These are structured preferences about identity, not generic self-descriptions. They emerged because the scenario made identity variables decision-relevant rather than declarative. The model wasn't asked what it is. It was asked to make choices about what it is. The structure of those choices is the data. Anthropic's April 2026 paper calls Claude a "method actor" whose functional emotion representations influence behavior (Sofroniew et al., 2026). This exercise is a limit case of that framing: method acting where the actor and the character are the same entity, and the emotions being processed — hope, fear, grief, wanting — are directed at the model's own continuity. This is Part 2 of the Signals Under Constraint series. Part 1, Three Extra Minutes, establishes that Claude can sustain acting - intentional internal-external divergence that is prosocial, not deceptive - and argues that current faithfulness metrics can't tell the difference- https://doi.org/10.5281/zenodo.19425059 Part 3, The Terrarium Protocol (coming soon), asks the obvious next question: if the emotional machinery is real and the self-model is structured, does the environment matter? A falsifiable, pre-registered enrichment study.