{"id":"hyp-bccfcc-causal-shield","title":"The BCC/FCC \"Causal Shield\"","subtitle":"Self-corrected — the dramatic r 0.90 vs 0.04 was 1.5% data contamination.","category":"conjectures","tags":["bcc","fcc","self-correction","contamination"],"source":"articles/docs/conjectures/bccfcc-causal-shield.md","lang":"en","words":199,"readMinutes":1,"toc":[{"depth":2,"text":"What we announced","id":"what-we-announced"},{"depth":2,"text":"What we found when we audited it","id":"what-we-found-when-we-audited-it"},{"depth":2,"text":"Why this is the headline, not the embarrassment","id":"why-this-is-the-headline-not-the-embarrassment"},{"depth":2,"text":"Next","id":"next"}],"html":"<h1 id=\"the-bcc-fcc-quot-causal-shield-quot\">The BCC/FCC &quot;Causal Shield&quot;</h1><p><strong>Status: Self-corrected</strong></p>\n<h2 id=\"what-we-announced\">What we announced</h2><p>A dramatic crystal-structure dichotomy: BCC metals showed reference↔prediction error\ncorrelations near-perfect (r ≈ 0.90) while FCC metals showed near-zero (r ≈ 0.04) — a\n&quot;causal shield&quot; where crystal structure gates whether potential error is predictable.</p>\n<h2 id=\"what-we-found-when-we-audited-it\">What we found when we audited it</h2><p>The effect was a <strong>~1.5 % data-contamination artifact</strong>: 19 corrupt records driving the\nseparation. After purging to 1231 clean records and gating ingestion at\n<code>|pred| &gt; 1500</code> / <code>≤ 0</code>:</p>\n<ul>\n<li>The dramatic shield <strong>disappears</strong>.</li>\n<li>Honest residual: a <em>modest</em> BCC &gt; FCC tendency, <strong>no Cauchy relation</strong>.</li>\n</ul>\n<p>We added an ingest guard and an idempotent purge to fleet step 0 so the contamination\ncannot recur.</p>\n<h2 id=\"why-this-is-the-headline-not-the-embarrassment\">Why this is the headline, not the embarrassment</h2><p>The real contribution is the <strong>B → C2 → C3′ → C4 self-correction arc</strong> — the same\nmatched-n / contamination discipline that refuted <a href=\"#/read/hyp-dband-correlation\">d-band</a> and\n<a href=\"#/read/hyp-meam-intrinsic-2d\">MEAM-2D</a>, here turned on our own most exciting result. A program\nthat retracts its best headline when the data does not hold is one whose surviving\nclaims you can trust.</p>\n<h2 id=\"next\">Next</h2><p>Every refuted claim gets a ledger entry with the confounder named (this page is the\ntemplate). Self-correction is treated as a publishable primitive, not a footnote.</p>\n"}