{
  "registry_version": "1.0.0",
  "data": {
    "title": "Remove Career Gap Feature from Automated Ranking Models",
    "status": "active",
    "evidence_ids": [
      "evidence.duration_dependence_and_labor_market_conditions_evidence_from_a_field_experiment"
    ],
    "content": "# Remove Career Gap Feature from Automated Ranking Models\n\nDisable employment duration gap penalties in recruitment AI algorithms and screening models.\n\n### Expected Effects\n- mech.employment_gap_downranking_bias stops operating in the ranking model: continuity is no longer an input\n- loop.employment_gap_penalty_loop loses the edge that closed it, because a gap can no longer widen the next gap\n\n### Measurements\n- `gap_candidate_screen_rate`\n- `long_term_placement_success`",
    "aliases": [
      "I-004"
    ],
    "id": "int.remove_career_gap_feature_from_automated_ranking_models",
    "type": "intervention",
    "summary": "Disable employment duration gap penalties in recruitment AI algorithms and screening models.",
    "targets": [
      "mech.employment_gap_downranking_bias",
      "loop.employment_gap_penalty_loop"
    ],
    "actor": "ats-vendor",
    "scope": "ecosystem",
    "cost": "low",
    "evidence_level": "supported",
    "expected_effects": [
      "mech.employment_gap_downranking_bias stops operating in the ranking model: continuity is no longer an input",
      "loop.employment_gap_penalty_loop loses the edge that closed it, because a gap can no longer widen the next gap"
    ],
    "measurements": [
      "gap_candidate_screen_rate",
      "long_term_placement_success"
    ],
    "specimens": [
      {
        "kind": "ats",
        "label": "The same profile, with the feature removed",
        "lines": [
          {
            "text": "Skills match: 91/100 · Domain match: 88/100",
            "tell": false
          },
          {
            "text": "Continuity penalty: feature disabled",
            "tell": true
          },
          {
            "text": "Final rank: 89/100 — advanced to recruiter queue",
            "tell": false
          },
          {
            "text": "Model card updated: employment_gap removed from feature set, 14 March.",
            "tell": false
          }
        ],
        "reading": "The arithmetic is the whole intervention. The same person, the same work, above the threshold instead of below it."
      }
    ],
    "perspectives": [
      {
        "actor": "actor.ats_vendor",
        "sees": "A feature column in the training set, and the ranked order that shifts for every customer once it is removed.",
        "reads": "The removal itself is one line in a feature list and a re-scoring run; the effect it is made for happens outside the platform, where there is no telemetry.",
        "does": "Ships the change with the model card updated, and carries the support load from customers whose thresholds were tuned against the old distribution."
      },
      {
        "actor": "actor.recruiter",
        "sees": "Profiles in the queue that the previous ranking placed below the review threshold.",
        "reads": "The queue is longer and the screening capacity is the same; ranking decides reading order, not how many get read.",
        "does": "Reads down the new order until capacity runs out, and the profiles below that line stay unread — for a reason that is no longer the gap."
      },
      {
        "actor": "actor.public_policy_and_industry_standards",
        "sees": "A published model card naming the feature that was removed, in a market where the same feature ships enabled by default elsewhere.",
        "reads": "One vendor removing it changes one vendor's ranking; what would hold it across the market is a rule about what an automated screen may read at all.",
        "does": "Writes the requirement against what a vendor must publish and explain; what a customer switches on inside a configuration screen sits outside what any disclosure rule inspects."
      }
    ]
  }
}