{
  "registry_version": "1.0.0",
  "data": {
    "title": "Employment Gap Penalty Loop",
    "status": "active",
    "evidence_ids": [
      "evidence.duration_dependence_and_labor_market_conditions_evidence_from_a_field_experiment"
    ],
    "content": "# Employment Gap Penalty Loop\n\nAn employment gap triggers automated downranking (mech.employment_gap_downranking_bias), which pushes the profile below keyword and qualification filters (mech.automated_keyword_qualification_filter); fewer interview invitations lengthen the gap, which in turn strengthens the downranking.\n\n### Cycle Dynamics\nThis causal loop reinforces mechanisms across iterations:\n- `mech.employment_gap_downranking_bias` amplifies `mech.automated_keyword_qualification_filter`\n- `mech.automated_keyword_qualification_filter` amplifies `mech.employment_gap_downranking_bias`",
    "aliases": [
      "L-001"
    ],
    "id": "loop.employment_gap_penalty_loop",
    "type": "loop",
    "summary": "An employment gap triggers automated downranking (mech.employment_gap_downranking_bias), which pushes the profile below keyword and qualification filters (mech.automated_keyword_qualification_filter); fewer interview invitations lengthen the gap, which in turn strengthens the downranking.",
    "mechanisms": [
      "mech.employment_gap_downranking_bias",
      "mech.automated_keyword_qualification_filter"
    ],
    "edges": [
      {
        "from": "mech.employment_gap_downranking_bias",
        "to": "mech.automated_keyword_qualification_filter",
        "relation": "amplifies"
      },
      {
        "from": "mech.automated_keyword_qualification_filter",
        "to": "mech.employment_gap_downranking_bias",
        "relation": "amplifies"
      }
    ],
    "entry_points": [
      "mech.employment_gap_downranking_bias"
    ],
    "interventions": [
      "int.remove_career_gap_feature_from_automated_ranking_models"
    ],
    "evidence_level": "supported",
    "specimens": [
      {
        "kind": "ats",
        "label": "The same profile, two cycles apart",
        "context": "same CV, same market, nine months between",
        "lines": [
          {
            "text": "Cycle 1 — skills 91, domain 88, continuity penalty −22 (gap 14 months) → rank 57",
            "tell": true
          },
          {
            "text": "Cycle 2 — skills 91, domain 88, continuity penalty −34 (gap 23 months) → rank 45",
            "tell": true
          },
          {
            "text": "Advance threshold both times: 60.",
            "tell": false
          }
        ],
        "reading": "Nothing in the work changed between the two rows. The only variable that moved is the one the penalty reads, and it moves because the penalty worked the first time."
      },
      {
        "kind": "note",
        "label": "Why the loop closes",
        "lines": [
          {
            "text": "Downranking pushes the profile below the filter threshold.",
            "tell": false
          },
          {
            "text": "Below the threshold there are no interviews.",
            "tell": false
          },
          {
            "text": "No interviews means no offer, so the gap continues to grow.",
            "tell": false
          },
          {
            "text": "A longer gap increases the penalty on the next application.",
            "tell": true
          }
        ],
        "reading": "Each step is a reasonable rule on its own. Together they form a cycle whose input is its own output."
      }
    ],
    "perspectives": [
      {
        "actor": "actor.candidate",
        "sees": "Applications that return no reply at all, or a generic rejection, with the same CV and the same market as the cycle before.",
        "reads": "Something in the profile moved between cycles, and the only thing that did move is the length of the break itself.",
        "does": "Keeps applying, since applications are the only input available, and each month without an interview adds to the break the next application carries."
      },
      {
        "actor": "actor.recruiter",
        "sees": "A ranked queue of several hundred applications against capacity to read a few dozen, on a requisition where time-to-fill is measured.",
        "reads": "The order is a triage of a list that cannot be read end to end, not a judgement about anyone in it.",
        "does": "Reads from the top down until the week's capacity is used, which is where the ranking becomes the decision."
      },
      {
        "actor": "actor.ats_vendor",
        "sees": "A continuity feature computed from the dates in the parsed work history, shipping enabled by default.",
        "reads": "It is one input among several in a ranking the buyer evaluates on throughput; whether a low-scored profile was unreadable rather than unqualified is not something the score records.",
        "does": "Applies the penalty on every scoring pass and offers it as a setting the customer can switch off, which nothing in the default configuration prompts anyone to do."
      }
    ]
  }
}