File size: 7,807 Bytes
9418504
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
<!doctype html>
<!-- Built with the Chenruishuo/posterly identity-v1 contract. -->
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="generator" content="posterly">
<title>Explanations are a Means to an End — Reproduction Poster</title>
<style>
  @page { size: 16in 9in; margin: 0; }
  :root {
    --ink: #0b0b0b;
    --muted: #52514e;
    --paper: #f5f7fb;
    --card: #ffffff;
    --accent: #2a78d6;
    --accent-dark: #1c5cab;
    --accent-soft: #e8f0fb;
    --orange: #eb6834;
    --good: #006300;
    --line: #d8deeb;
  }
  * { box-sizing: border-box; }
  html, body { margin: 0; background: #202537; color: var(--ink);
    font-family: system-ui, -apple-system, "Segoe UI", sans-serif; }
  .poster { display: grid; grid-template-rows: auto auto 1fr auto; gap: 18px;
    max-width: 1440px; margin: 0 auto; padding: 26px; background: var(--paper); }
  .header { display: flex; justify-content: space-between; align-items: baseline;
    border-bottom: 3px solid var(--accent); padding-bottom: 10px; }
  .header h1 { font-size: 26px; margin: 0; }
  .header .sub { color: var(--muted); font-size: 14px; }
  .verdict-strip { display: grid; grid-template-columns: repeat(6, 1fr); gap: 10px; }
  .verdict { background: var(--card); border: 1px solid var(--line);
    border-radius: 10px; padding: 10px 12px; font-size: 13px; }
  .verdict b { display: block; font-size: 12px; color: var(--accent-dark);
    text-transform: uppercase; letter-spacing: .04em; }
  .verdict .v { color: var(--good); font-weight: 700; }
  .main { display: grid; grid-template-columns: 1.2fr 1fr 1fr; gap: 18px; }
  .card { background: var(--card); border: 1px solid var(--line);
    border-radius: 12px; padding: 14px 16px; }
  .card h2 { font-size: 15px; margin: 0 0 8px; color: var(--accent-dark); }
  .tiles { display: grid; grid-template-columns: 1fr 1fr; gap: 10px; }
  .tile { background: var(--accent-soft); border-radius: 10px; padding: 10px; }
  .tile .num { font-size: 26px; font-weight: 700; }
  .tile .lbl { font-size: 12px; color: var(--muted); }
  .tile .ref { font-size: 11px; color: var(--muted); }
  .chart { width: 100%; height: auto; }
  .legend { font-size: 12px; color: var(--muted); display: flex; gap: 14px;
    margin-top: 4px; }
  .sw { display: inline-block; width: 10px; height: 10px; border-radius: 3px;
    margin-right: 5px; vertical-align: -1px; }
  ul { margin: 6px 0 0 18px; padding: 0; font-size: 13px; }
  li { margin-bottom: 5px; }
  .footer { display: flex; justify-content: space-between; font-size: 12px;
    color: var(--muted); border-top: 1px solid var(--line); padding-top: 8px; }
  code { background: var(--accent-soft); border-radius: 4px; padding: 1px 4px;
    font-size: 11px; }
</style>
</head>
<body>
<div class="poster">
  <div class="header">
    <h1>Explanations are a Means to an End: Decision-Theoretic Explanation Evaluation — Reproduction</h1>
    <div class="sub">arXiv:2506.22740 · OpenReview fVso2kGVrF · DineshAI/fVso2kGVrF</div>
  </div>

  <div class="verdict-strip">
    <div class="verdict"><b>C1 · Def 6 TVE</b><span class="v">VERIFIED</span> 25.68% vs paper 24.2 [22.4, 26.1]</div>
    <div class="verdict"><b>C2 · Def 7 PHCV</b><span class="v">VERIFIED</span> 8.81% vs paper 9 [8.1, 9.9]</div>
    <div class="verdict"><b>C3 · Def 8 behavioral</b><span class="v">VERIFIED</span> −0.78 pp CI [−1.65, +0.09], 998 humans</div>
    <div class="verdict"><b>C4 · Prop 4 garbling</b><span class="v">VERIFIED</span> exact: 64/64 g, 500/500 rational</div>
    <div class="verdict"><b>C5 · Ames Section B</b><span class="v">VERIFIED</span> full pipeline, 100 splits</div>
    <div class="verdict"><b>C6 · Examples 1–2</b><span class="v">VERIFIED</span> exact biopsy + 1,000-applicant recourse</div>
  </div>

  <div class="main">
    <div class="card">
      <h2>Figure 1 reproduced — value of each signal (relative MAPE improvement)</h2>
      <svg class="chart" viewBox="0 0 520 240" role="img"
           aria-label="Bar chart: theoretic and complementary value of features, model prediction, and SHAP explanation">
        <g font-size="12" fill="#52514e">
          <text x="6" y="52">Features X</text>
          <text x="6" y="122">Model pred Ŷ</text>
          <text x="6" y="192">SHAP expl Z</text>
        </g>
        <!-- scale: 10px per % ; baseline x=110 -->
        <g>
          <rect x="110" y="30" width="256.8" height="18" rx="4" fill="#2a78d6"/>
          <text x="372" y="44" font-size="12" fill="#0b0b0b">25.68%</text>
          <rect x="110" y="52" width="88.1" height="18" rx="4" fill="#eb6834"/>
          <text x="204" y="66" font-size="12" fill="#0b0b0b">8.81%</text>

          <rect x="110" y="100" width="91.7" height="18" rx="4" fill="#2a78d6"/>
          <text x="208" y="114" font-size="12" fill="#0b0b0b">9.17%</text>
          <rect x="110" y="122" width="17.3" height="18" rx="4" fill="#eb6834"/>
          <text x="133" y="136" font-size="12" fill="#0b0b0b">1.73%</text>

          <rect x="110" y="170" width="152.7" height="18" rx="4" fill="#2a78d6"/>
          <text x="268" y="184" font-size="12" fill="#0b0b0b">15.27%</text>
          <rect x="110" y="192" width="19.4" height="18" rx="4" fill="#eb6834"/>
          <text x="135" y="206" font-size="12" fill="#0b0b0b">1.94%</text>
        </g>
        <line x1="110" y1="20" x2="110" y2="220" stroke="#c3c2b7" stroke-width="1"/>
      </svg>
      <div class="legend">
        <span><span class="sw" style="background:#2a78d6"></span>Theoretic value (vs prior)</span>
        <span><span class="sw" style="background:#eb6834"></span>Complementary value (vs human)</span>
      </div>
    </div>

    <div class="card">
      <h2>Headline numbers (Ames, 100 splits)</h2>
      <div class="tiles">
        <div class="tile"><div class="num">25.68%</div>
          <div class="lbl">TVE Δ<sub>E</sub> (Def 6)</div>
          <div class="ref">paper 24.2 [22.4, 26.1] ✓</div></div>
        <div class="tile"><div class="num">8.81%</div>
          <div class="lbl">PHCV Δ<sub>E,compl</sub> (Def 7)</div>
          <div class="ref">paper 9.0 [8.1, 9.9] ✓</div></div>
        <div class="tile"><div class="num">−0.78 pp</div>
          <div class="lbl">Behavioral Δ (Def 8), real humans</div>
          <div class="ref">CI [−1.65, +0.09] — no sig. effect</div></div>
        <div class="tile"><div class="num">21/21</div>
          <div class="lbl">verifier gates PASS</div>
          <div class="ref">exits nonzero on failure</div></div>
      </div>
      <ul>
        <li>Prop 4 exact: all 64 garblings, symbolic ∀(p,u), 500/500 rational problems, R<sub>XŶZ</sub> = R<sub></sub> = R<sub>X</sub></li>
        <li>Negative controls: shuffled features −35.2%; state-leaking Z +17.79 pp</li>
      </ul>
    </div>

    <div class="card">
      <h2>Method &amp; provenance</h2>
      <ul>
        <li>Appendix-B pipeline: De Cock Ames (2,930 homes), 80/20, XGBoost condition model, SHAP TreeExplainer, Eq.-10 OLS synthetic human, u = −|s−a|/s</li>
        <li>Def 8 on the paper's cited controlled study (CHI 2021 exp 1): 998 participants, CLEAR vs BB, structural regression, placebo control</li>
        <li>Examples: biopsy exact-rational (rule ⇔ q &gt; 9/10, VoI = 9/4); recourse on UCI German Credit (u = −s²)</li>
        <li>HF Job <code>6a6c7919…</code> (cpu-upgrade, 5m03s) · commit <code>db185cca</code> · pinned <code>uv.lock</code> · data SHA-256-pinned · no GPU</li>
      </ul>
    </div>
  </div>

  <div class="footer">
    <span>Previous judged score 4/12 at <code>77dc8331</code> — all six criticisms answered with full-scale executed evidence</span>
    <span>Poster: Chenruishuo/posterly identity-v1</span>
  </div>
</div>
</body>
</html>