Automation
Apply the screening bar unattended, with a kill switch, an agreement report, and a reversals log.
Automation
Once a Role Bar is in force, you can let the platform apply it unattended. The Automation view is where you turn the bar into decisions, watch how the bar agrees with what recruiters actually do, and stop the process if anything looks off.
What automation does
| Capability | What it is |
|---|---|
| Apply the bar | For each candidate in the inbox, write the verdict the policy decides, attributed to the human who triggered the apply. |
| Agreement report | A measure of how closely the policy's verdicts match the verdicts recruiters have been writing manually. |
| Reversals | The audit trail of automated decisions and their manual reversals. |
| Halt | A kill switch that stops the policy from deciding anything further. |
The lifecycle
Before you apply
- Open Screening → Auto-screening.
- Review the Inbox — candidates whose evidence is complete and ready for the bar.
- Click Preview apply. The preview shows how many candidates would receive each
verdict, with transitions keyed
BEFORE->AFTER. - If the response is
AUTOMATION_NOT_READY, the platform deliberately returns the measured distance to ready (which tokens are missing, which checks are incomplete, etc.) so you know what to do.
Why the gap is reported
A refusal that says only "not ready" leaves the user guessing. The platform returns the measured gap so a "not ready" response is an answer, not a dead end.
Modes
| Mode | What it does |
|---|---|
| Off | Nothing is decided automatically. Every candidate waits for a person. |
| Trial | Conclusions are recorded so they can be compared with your reviewers' decisions. No candidate is affected. |
| Active | The bar decides, within the outcomes it permits. Each decision is recorded with source: POLICY and the policyVersionId. |
The platform's safeguards are visible in the mode selector:
- Automation never decides an integrity concern. Even with the bar allowed to act unattended, the policy refuses to write a verdict that would reject on a fraud signal.
- A
REVIEWflag in the bar means human review. If the bar escalates a flag, the candidate is sent to a human regardless of the automation mode.
Halt
The Halt button is a kill switch. It stops the policy from deciding anything
further. It does not undo past automated decisions; that is what the reversals log is
for. Past decisions keep the policyVersionId that produced them, so the audit trail
remains honest after a halt.
Reversals and agreement
Every automated decision is reversible. The reversals log records:
- The automated decision (verdict, when, policy version)
- The human reversal (verdict, when, by whom)
- The reason provided
The agreement report shows, across recent decisions, how often the policy and the human agree. A low agreement rate is a signal that the bar no longer matches how your team actually decides — at which point you author a new version.
Two truths, both recorded
A bulk "apply the bar" is policy deciding and a named person authorising the batch. The audit trail records both: the policy version that decided, and the human who authorised. Either alone misstates what happened.