{
 "id": "plate-xxiii",
 "number": "XXIII",
 "title": "The Borrowed Verdict",
 "act": {
  "number": "VI",
  "title": "The Higher Court"
 },
 "corpus_scope": "ICLR 2018–2026 · 8,439 papers · every meta-review unit matched to the reviewer unit it most resembles",
 "question": "Where does the area chair's language in a meta-review come from — near-copy, echo, or their own words — and whose voice gets echoed?",
 "page_url": "https://atlas-of-judgment.pages.dev/",
 "figures": [
  {
   "id": "fig-23a",
   "title": "What the meta-review is made of",
   "deck": "73% of meta-review reasoning is the area chair's own words; 0.5% is near-copy. But criticism is borrowed more than praise — at every cutoff.",
   "dom_host": "#acecho-tiers",
   "claims": [
    {
     "id": "23a-own-copy",
     "statement": "73% of meta-review reasoning is the area chair's own words; 0.5% is near-copy.",
     "value": {
      "quantity": [
       0.73,
       0.005
      ],
      "unit": "share_of_meta_review_units",
      "own": 0.73,
      "copy": 0.005
     },
     "source_island": "acecho-data.json",
     "source_path": "tiers[]",
     "derivation": "own = 0.7306 → 73%; copy = 0.0047 → 0.5%, share of meta-review units by cosine-similarity tier to their nearest reviewer unit",
     "recompute": "scripts/build_acecho_data.py",
     "dom_ref": "data-echo23a=\"own-copy\"",
     "verified": "checked against shipped JSON, 2026-08-25",
     "caveat_refs": []
    },
    {
     "id": "23a-crit-vs-praise",
     "statement": "Criticism's borrowed edge reaches 6.6 points past praise's, and the gap survives any cutoff (+7.0 / +6.6 / +3.3 at similarity thresholds of 0.70 / 0.75 / 0.80).",
     "value": {
      "quantity": [
       7.0,
       6.6,
       3.3
      ],
      "unit": "pp_gap_by_cutoff",
      "cutoffs": [
       0.7,
       0.75,
       0.8
      ]
     },
     "source_island": "acecho-data.json",
     "source_path": "tiers[]",
     "derivation": "borrowed-mass (copy+echo) share for criticism minus praise meta-units, recomputed at similarity cutoffs 0.70/0.75/0.80; at the 0.75 default, negative meta units echo reviewers at 28.8% against 22.4% for positive ones",
     "recompute": "scripts/build_acecho_data.py",
     "dom_ref": "data-echo23a=\"crit-vs-praise\"",
     "verified": "claim recomputed from source island, 2026-08-25",
     "caveat_refs": []
    }
   ]
  },
  {
   "id": "fig-23b",
   "title": "When the AC borrows, from whom",
   "deck": "Echoes concentrate on one reviewer (60% from a single voice), and in rejections they lean to the harsher side of the panel, 60/40.",
   "dom_host": "#acecho-who",
   "claims": [
    {
     "id": "23b-single-voice",
     "statement": "On papers with three or more echoed units, 60% of echoes come from a single reviewer.",
     "value": {
      "quantity": 0.6,
      "unit": "share_of_echoes_from_one_reviewer"
     },
     "source_island": "acecho-data.json",
     "source_path": "top_echo_share_hist[] / side[]",
     "derivation": "share of echo-tier meta units on multi-echo papers (≥3 echoed units) attributed to the single most-echoed reviewer",
     "recompute": "scripts/build_acecho_data.py",
     "dom_ref": null,
     "verified": "claim recomputed from source island; source_path corrected from who[], 2026-08-26",
     "caveat_refs": []
    },
    {
     "id": "23b-low-vs-high",
     "statement": "The most-echoed reviewer is the panel's uniquely lowest rater in 16% of cases against 13% for the uniquely highest.",
     "value": {
      "quantity": [
       0.16,
       0.13
      ],
      "unit": "share_of_panels",
      "lowest": 0.16,
      "highest": 0.13
     },
     "source_island": "acecho-data.json",
     "source_path": "echoed_position[]",
     "derivation": "ACECHO.echoed_position.lowest and .highest, among the 3,334 panels with full, non-unanimous rating panels; in half of panels the echoed reviewer's rating ties an extreme",
     "recompute": "scripts/build_acecho_data.py",
     "dom_ref": "data-echo23b=\"low-vs-high\"",
     "verified": "claim recomputed from source island, 2026-08-25",
     "caveat_refs": [
      "nearest-neighbor-not-causal"
     ]
    },
    {
     "id": "23b-rejection-lean",
     "statement": "In rejections, echoes lean to the harsher side of the panel, 60/40; accepts do not (51/49, interval crossing even).",
     "value": {
      "quantity": [
       0.6,
       0.51
      ],
      "unit": "share_toward_harsher_side",
      "rejections": 0.6,
      "accepts": 0.51
     },
     "source_island": "acecho-data.json",
     "source_path": "top_echo_share_hist[] / side[]",
     "derivation": "share of echoes attributed to the harsher-than-mean side of the panel, split by decision, among the 2,220 decided rejections and 752 accepts where the echo-weighted rating differs from the panel mean",
     "recompute": "scripts/build_acecho_data.py",
     "dom_ref": null,
     "verified": "claim recomputed from source island; source_path corrected from who[], 2026-08-26",
     "caveat_refs": [
      "nearest-neighbor-not-causal"
     ]
    }
   ]
  }
 ],
 "caveats": [
  {
   "id": "nearest-neighbor-not-causal",
   "text": "Nearest-neighbor attribution measures textual proximity, not causal influence: an area chair and a reviewer may both independently echo the paper itself rather than one echoing the other.",
   "scope": "Fig 23b who-is-echoed and side-of-panel findings",
   "verified_by": null
  }
 ],
 "links": {
  "recompute_scripts": [
   "scripts/build_acecho_data.py"
  ],
  "data": [
   "/api/v1/data/acecho-data.json"
  ],
  "related_plates": [
   "plate-xxii"
  ]
 }
}