Skip to content
AvelinLabs

Decision Scoring API Use Case

Problem

Section titled “Problem”

A match score alone is not enough to decide whether a workflow should automate, review, or reject a result.

How AvelinLabs Helps

Section titled “How AvelinLabs Helps”

POST /api/v1/job/analyze and POST /api/v1/job/classify return decision-support fields that help applications interpret result strength. They share the same ranking and root-confidence pipeline; their response shapes differ.

Why It Is Different

Section titled “Why It Is Different”

Avelin separates the proposed occupation from the strength and limits of its support. Review and ambiguity labels make human attention an explicit workflow outcome rather than hiding uncertainty inside one score.

Input Signals

Section titled “Input Signals”

Use the fields according to their contract:

  • top-level confidence represents calibrated winner strength for the response;
  • results[].confidence represents relative probability within the returned ranked set;
  • trust_score adds top-result quality support to raw winner strength;
  • uncertainty, is_ambiguous, domain_is_ambiguous, weak-signal, and quality fields expose different limitations;
  • decision.reason and decision.applied_rule explain the routing label that was selected.

Decision Output

Section titled “Decision Output”

The current job-intelligence responses use these decision labels:

  • AUTO_ACCEPT
  • REVIEW
  • REJECT
  • AMBIGUOUS

These labels route an API result; they do not approve a hiring decision. Combine them with product-specific review and risk policies.

AUTO_ACCEPT

Section titled “AUTO_ACCEPT”
  • Meaning: Confidence satisfies the runtime rule and epistemic uncertainty stays within its configured boundary.
  • Typical action: Use as a display or low-risk enrichment default only where your own policy permits it.

REVIEW

Section titled “REVIEW”
  • Meaning: Usable support that does not meet the automatic-accept path.
  • Typical action: Send to a reviewer or consultant.

REJECT

Section titled “REJECT”
  • Meaning: The configured minimum path for review or acceptance was not met.
  • Typical action: Do not use for automation.

AMBIGUOUS

Section titled “AMBIGUOUS”
  • Meaning: Epistemic uncertainty is high enough to prefer review over commitment.
  • Typical action: Ask for more input or review manually.

Product Fit

Section titled “Product Fit”

Decision scoring can support analyst review queues, workflow routing, quality control, dashboards, and human-in-the-loop automation.

Implement and verify

Section titled “Implement and verify”

Use the Response Field Reference and Output Interpretation before defining thresholds. Then test representative strong, weak, ambiguous, vague, and noisy inputs with the official executable examples; do not infer production accuracy from confidence alone.