After the machine

This record publishes the locked design specification for the AI Interaction Outcome Profile (AIOP) — After the Machine, the sixth instrument (A6) in the Applied Series of the Heartbeat Framework — ahead of the full instrument paper. Publishing the specification serves the function of a protocol registration: the design is timestamped before the instrument exists, so that what changes between specification and instrument is itself part of the record. AIOP is a dual-lane psychosocial outcome measure for adults in 24-hour care settings who use conversational AI. Self-report is one lane; keyworker observation the second; a change-over-time log makes both auditable. Six bipolar domains — displacement versus supplementation, agency, epistemic stance, regulation in the tool's absence, human alliance, disclosure pattern — are scored as a profile, never a single number. Four commitments distinguish the design. Every axis runs from a harm pole to a benefit pole, so supplementation can be evidenced as readily as displacement. A one-time baseline module establishes the person's own pre-AI normal — trajectory is read against the person, not a population, and the closest people are treated as the pre-existing ledger of who the person was before the machine. A rapid-triage red route interrupts the measurement machinery wherever harm to self, harm to others, or harm from others is indicated, handing off to the statutory frameworks that already govern the setting (Mental Capacity Act 2005; Mental Health Act 1983 as amended; Care Act 2014 s42) — risk is absolute, not relative. And administration inherits the review cadence already binding the setting, at national and provider level, rather than imposing one. AIOP is the person-level sibling of The Damping Audit (10.5281/zenodo.21455540), which asks the same question at ward level; it is designed to supply the ongoing human-outcome evidence a DCB0160 clinical safety case requires but currently has no instrument for. The review architecture is published in full: four orthogonal gates, one held, three deliberately open, with reviewer naming governed by an executed redaction rule pending written consent. A worked example (Fernlea House, pseudonymised composite) runs the machinery end to end, including one flag and its correction. Evidence Tier C: internally coherent, unvalidated; the validation pathway and its confounds are named. The full instrument paper follows. CC BY 4.0.