Observable review
Retrospective measurement of recorded review behavior, intervention and escalation. Missing evidence remains unknown.
VALO product · experimental prototype
Human Control Barometer
Measure the evidence of human control in AI-assisted decisions.
HCB is a standalone retrospective measurement product. It examines review timing, intervention, escalation and independent challenge so teams can inspect what a human review actually involved.
Inspect the code · Read the research · Discuss a research pilot
The product
Review should leave measurable evidence
An approval records a decision. HCB examines the observable review around that decision: time available, interventions, challenges, escalation and independent judgment. Authority evidence can be inspected separately from review behavior.
Separate control and authority
Behavioral evidence describes the review. HEIMEL resolves current handlingsrett at the consequence boundary. A barometer reading does not grant permission to act.
Human Control Research / Nishita track
Did the right person understand the proposal?
The research programme tests comprehension, independent challenge and reviewer expertise. Teachback asks reviewers to explain a proposal, predict its consequences and identify errors. Code explanations, rationale and provenance can support review; whether they improve understanding must be measured.
Understanding and task fit
Relevant expertise and performance on the actual task are separate research variables. A confident approval alone establishes neither understanding nor competence.
Physiological markers
Gaze, pupil response, HR/HRV, EDA and EEG are candidate research channels. Their added value must be tested against behavior and individual baselines. Physiology is not ground truth.
Product status
Experimental. Retrospective validation
The current baseline is a software measurement prototype with behavioral HCM integration. It is not presented as a validated clinical system or a finished physiological monitoring product.
No validated physiological inference, prospective control-failure prediction, clinical safety claim or sensor-created authority or fitness decision is claimed. Teachback and reviewer-fit methods remain research work.
Inspect and collaborate
Start with a bounded review study
The immediate application is retrospective or simulated review of AI-assisted decisions. Clinical simulation and machine-generated code review are research contexts; results must be tied to a defined task, baseline and observed evidence.
Public repository · HCM, HCB and HCS research · Contact VALO