full-audit-v2/poster_embed.html

<div style="font-family:Inter,system-ui,sans-serif;background:#071827;color:#eef7ff;padding:32px;border-radius:24px;line-height:1.35">
  <div style="color:#69d2ff;font-size:14px;font-weight:800;letter-spacing:.12em">ICML 2026 · INDEPENDENT REPRODUCTION</div>
  <h1 style="font-size:34px;line-height:1.08;margin:10px 0 8px">Causal Modeling of Selection in Evolution</h1>
  <p style="font-size:18px;color:#b9cee0;margin:0 0 24px">Exact graph audit + paper-native PC, GES, CD-NOD, and Census PUMS evaluation</p>
  <div style="display:grid;grid-template-columns:repeat(3,minmax(0,1fr));gap:14px">
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#73efb4">0 / 40,800</div><div>clique-augmentation mismatches</div></div>
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#73efb4">0 / 3,843</div><div>oracle CPDAG adjacency errors</div></div>
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#73efb4">+252</div><div>sound oracle multi-domain orientations</div></div>
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#ffd166">150</div><div>finite-sample PC/GES runs, N=5,000</div></div>
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#ff8f8f">−0.0113</div><div>PC oriented-vs-all adjacency precision</div></div>
    <div style="background:#102c40;padding:18px;border-radius:16px"><div style="font-size:28px;font-weight:800;color:#69d2ff">155,618</div><div>complete official PUMS rows</div></div>
  </div>
  <div style="margin-top:20px;padding:18px;border-left:5px solid #ffd166;background:#0d2436;border-radius:10px">
    <b>Outcome:</b> the exact graphical representation and oracle-identifiability claims reproduce. The finite-sample precision improvement does not reproduce in this clean-room implementation; one of seven real-data studies was rerun.
  </div>
</div>