Screening Answers

Use Screening answers in Job Detail or Apply Review to prepare open-ended answers. Choose Open screening answers to load the private library. A saved Candidate Profile, captured posting and canonical application destination are needed for drafting. This library is local and private; it does not discover questions from a browser or fill application forms.

Prepare and Review

1. Enter an application attempt identifier, the exact question and its context or constraints. Use a different attempt identifier for a later application to the same job. Capture creates a question identity separate from the Job ID.
2. Select saved supporting facts individually. Personal, authorization, compensation, availability, voluntary-demographic and application-attestation or preference fields also require the separate sensitive-value checkbox. Empty values provide no factual authority.
3. Generate a draft, or enter your own text and choose Verify edited draft. Question interpretation, factual verification and a separate quality judgment use the configured Apply model lane. Uncertainty and provider/budget failures remain explicit; no keyword or similarity fallback supplies an answer.
4. Inspect the source bindings and persisted model receipts. Selected sources describe what was allowed; the verifier's cited sources describe what it actually found supported. Choose Approve reviewed answer only after reading the entire draft. Approval creates an immutable reusable library revision and an application-bound review snapshot. Rejection preserves the previous accepted answer.

Revise question or context saves another question version. It keeps the old accepted answer for inspection and requires a newly verified, reviewed answer for the changed question. Unsaved answer-editor text survives navigation in the current browser session and failed operations. Loading a generated draft into that editor is an explicit replacement action.

Reuse, Copy and Record Use

Open Reviewed answer library, choose a prior answer, select its supporting facts in the current job and choose Verify reuse. Reuse requires the same exact selected facts and sensitive-field consent. A model independently checks question meaning and job-specific claims against the current destination; claim verification and quality review run again. The result is a new draft, which needs a new human review. Reuse is never inferred from similar wording.

Changes to the profile, authorization, compensation, facts, question/context, posting, destination or material versions/content make old snapshots stale. The previous text and history remain inspectable. Copy checks current saved sources again before writing through the clipboard port; a stale answer cannot be copied through this control.

Copying records no manual use. To record use, enter the Actual manually used text, select the explicit attestation and choose Record manual use. If it differs from the reviewed answer, history retains those exact bytes as a changed, unverified user statement. It does not replace the reusable reviewed answer. Review, copying and manual-use recording confer no form-entry or submission authority and create no application-submitted outcome.

Open Application-bound history to inspect captured questions, draft/edit versions, decisions, source snapshots and manual-use attestations. A failed refresh, conflicting revision or persistence failure leaves accepted content and previous history intact. Reload history before retrying a conflict.

CLI and Local API

jobctrl screening read JOB_ID returns the private library, questions, history, current source availability and persisted model authority. Capture a question using jobctrl screening write JOB_ID command.json:

{
  "action": "capture",
  "applicationId": "my-attempt-1",
  "question": "Exact question copied deliberately",
  "context": "Employer context and answer constraints",
  "expectedRevision": 0,
  "idempotencyKey": "capture-attempt-1-question-1"
}

Read also includes safe failed-attempt receipts with action, version and failure code; accepted answers remain intact.

Subsequent commands use the returned questionId and current expectedRevision, a new idempotency key, and one action: draft, edit, review, reuse or use. Draft/edit/reuse take selectedFactIds and sensitiveFactIds from the saved read; edit adds text, review adds decision: approved|rejected, reuse adds libraryId, and use adds actual text and attested: true. Repeating an identical command with its original key returns its recorded result; changing its payload with that key is refused.

The corresponding local routes are GET and POST /v1/jobs/:jobId/screening-answers. See the API reference and owning architecture. The public demo reports this private worker-backed capability unavailable.