AI is the hallway consult.
The colleague queried in the hallway. The CME read last month. UpToDate at 2am. Epocrates before walking in the room with the patient. Clinical judgment has been shaped by external knowledge, and little of that support ever made it into the chart.
Now ambient scribes, radiology LLMs, and clinical copilots surface that knowledge at the moment of decision. It is possible to keep a dated record of that AI consult.
CAVAL is an open standard for that record.
WHAT CAVAL CAPTURES:
Three things for each AI-influenced clinical encounter.
What the AI surfaced: the output, the source, the timestamp.
What the clinician did with it: accepted, modified, or overridden.
What was decided: the clinical action, signed and attributed.
The standard specifies an append-only, tamper-evident record independent of any EHR or AI vendor, sitting between the AI output and the clinician signature.
Why it matters
The hallway consult, on the record.
From the clinician at the bedside to the physician with no specialist backup, CAVAL is an open standard for recording what the AI surfaced and what the clinician did with it.
For clinicians
A dated record of what the AI surfaced and what you decided, without stopping to document each step. If an AI-influenced decision is ever questioned, the record shows what you were shown and what you did.
For health systems
Not just whether the AI is drifting, but whether it's being used the way your governance committees agreed to. Secondary review of AI use, not just AI performance, and a possible path to replacing indefinite ambient transcript storage with a structured attribution record.
For the clinician with no backup
For generations, resource-constrained clinicians have waited for the asynchronous eConsult or the specialist who visits once a month. AI may be the first real-time second opinion many of them have ever had. CAVAL gives that second opinion a record: what it said, and what the clinician did with it.