Verdicts
The five verdict stages, what each means, who can write them, and how the audit trail records the difference between a human and a policy decision.
Verdicts
A screening verdict is a recorded decision about a candidate. Every change is an event in the candidate's history.
The five stages
| Stage | Meaning | Writable? |
|---|---|---|
New NEW | No decision recorded yet — "Pending" | Yes, by resetting (NEW button) |
Screening SCREENING | Evidence is still arriving | No — reflects state, not a decision |
Clear CLEAR | A reviewer advanced this candidate | Yes (Pass) |
Review REVIEW | A reviewer flagged this for a second look | Yes (Review) |
Reject REJECT | A reviewer ruled this candidate out | Yes (Reject) |
There is also a INSUFFICIENT_DATA value used as a machine suggestion only; it is
never a storable verdict.
Buttons are imperative, not stative
The buttons in the screening console use imperative labels:
| Button | What it does |
|---|---|
| Pass | Sets the verdict to CLEAR |
| Review | Sets the verdict to REVIEW |
| Reject | Sets the verdict to REJECT |
| Reset | Clears the decision and returns the candidate to NEW |
Status names ("Passed Screen", "Needs Review") describe a state. Action names ("Pass", "Review") offer to create one. The two are different parts of speech, and a button should offer to do something, not describe what is already done.
Who can write
| Role | Can write a verdict? |
|---|---|
| Admin | Yes, on any candidate in the workspace |
| Recruiter | Yes, on candidates from their own assessments |
See Roles and permissions for the full matrix.
Recording a verdict
- Open the Screening console.
- Find the candidate (search, filter, or pick a tab).
- Click Pass, Review, or Reject. Optionally add a reason.
- The row is removed from the active tab and the count updates. A toast appears with an Undo button that restores the previous verdict.
Undoing is a refetch rather than a patch: the row has left the active tab, so the only correct way to bring it back is to ask the server again.
The audit trail
Every verdict change is recorded as a SCREENING_VERDICT event with:
| Field | What it is |
|---|---|
from | The previous verdict (null if there was none) |
to | The new verdict |
reason | The optional reason provided |
actor | The user who acted |
source | HUMAN, SYSTEM, or POLICY |
when | The timestamp |
The distinction between HUMAN and POLICY is load-bearing. A bulk "apply the bar"
is policy deciding and a named person authorising the batch, and both facts are
recorded because either alone misstates it. Collapsing them would make an automated
employment decision indistinguishable from a data copy.
Override vs verdict
Two different writes can both be called "verdict":
| Write | Scope | What it does |
|---|---|---|
| Screening verdict | The candidate, across all their assessments | A reviewer's decision on a candidate |
| Recommendation override | A single attempt on a single assessment | A reviewer's manual override of the CETS recommendation on that one attempt |
The two are deliberately separate. One corrects the machine's read on one attempt; the other records the person's call on the candidate.
Recommendation vs verdict
The recommendation is the model's read. The verdict is the person's call. They are shown side by side in the queue so you can reconcile them, but they are stored separately so the audit trail is honest.