Healthcare

In care delivery the question is never whether a model produced a plausible recommendation. It is whether a licensed clinician's authority stood behind the action that reached the patient record, and whether that can still be demonstrated a year later when a payer, a board, or a plaintiff asks.

Health systems have spent two decades building attribution into everything that touches a patient. Every order has an ordering provider. Every note has an attesting author. Every amendment has a timestamp and a reason. Agentic systems arrived without any of that. A model call leaves a log line, not a signature, and a log line is not evidence of authority — it is evidence that software ran.

Orcher closes that gap without asking clinical operations to change how medicine is practiced. A directive is issued in plain language by a person whose licensure, specialty, and privileging are already known to the identity fabric. Every downstream agent inherits that scope and nothing wider. Before any write reaches the chart, the claim is reconciled against the systems of record that govern it — the encounter, the problem list, the coverage policy, the formulary. What commits is hashed with the clinician, the directive, and the verification result attached.

Human-in-the-loop assumed a reviewer who reads carefully, disagrees regularly, and is given the time to do it. Clinical reality supplies none of those conditions. Volume rises, review windows shrink, and confirmation becomes the default. The result is oversight theater: a signature that satisfies a policy document while proving nothing about whether the content was ever examined.

Human-in-the-Role inverts the arrangement. Authority is bound before execution rather than confirmed after it. A nurse's directive cannot produce a physician-level order regardless of how the model was prompted, because the identity fabric never issues a credential wide enough to permit it. The control is structural, and structural controls do not fatigue.

For every action that reached a patient record, Orcher can produce six facts without a forensic exercise: who held authority, what the directive asked for in plain language, which systems of record were checked, what the verification returned, which provider processed the data and under what terms, and what the cycle cost. Those six facts are the whole of what a regulator, a malpractice carrier, or an internal quality committee actually needs.

That record is written at execution time, not reconstructed afterward. Reconstruction is where health systems lose cases — not because the care was wrong, but because the evidence of who decided was never captured in a form that survives review.