Human review workspace

Signals become questions, not verdicts.

This workspace brings independent channels together, shows what is known and missing, and helps reviewers choose a proportionate next step.

Independent channels present0
No demo inputs yet
Review routeRoutine
Human decision required
Context completeness—
Not danger probability
Automation boundaryHuman
No employment action generated

Signal convergence map

What the system must not infer

No diagnosis

Stress, mental illness, personality disorder or clinical condition.

No ideology inference

Religion, political belief, radicalisation or motive from indirect proxies.

No guilt or employment recommendation

No automated suspension, termination, hiring, promotion or disciplinary decision.

Reviewer checklist

Verify specificity

Are dates, words, actions and context sufficiently concrete to examine?

Check independence

Are multiple reports truly independent, or are they repetitions of the same source?

Consider benign explanations

What ordinary operational, interpersonal or wellbeing explanations should be tested?

Choose the least intrusive adequate response

Support, manager conversation, operational review, HR, security or emergency action should remain distinct.

Proportionate response

Review saved locally with a timestamp for the prototype audit trail.
Signal convergence increases the need to examine context; it does not increase a person’s “danger score.” Production Sentinel should track calibration against defined review outcomes, false positives, overrides and subgroup disparities.