[amaybaum]

Source-correction unit for #946 (gate) and #918 (Introduction). It comes from the full-corpus audit of #918 at 60c976fe: R-32 / A-B3, the card-Bell sentence.

• Base: main at 0d917665.
• Head: 0e235c3c.
• Review: the independent review returned ACCEPT on this exact head after five rounds. The owner required that:
  • the correction cover every post-deal operation, rule-driven shuffles included;
  • the reviewer's CHSH = 4 counterexample be shown to be excluded;
  • every admitted local strategy be shown to have CHSH ≤ 2.

Finding

Book ch01:286 and ch03:188, and their FULL mirrors, said that no arrangement of cards violates a Bell inequality, with no conditions. Several card games reach CHSH = 4:

• setting-dependent reversible dynamics, or a dealer that reads both settings (a PR box);
• a rule-prescribed shuffle that reads the other player's setting;
• a schedule chosen by the other player's setting;
• two coins that are each setting-independent but jointly encode ab;
• a readout rule selected by the other player's setting.

Correction

Both sites now state the boundary at Main §3.3's hypotheses. Such a game violates no Bell inequality when all of the following hold:

• once the deal is down, every operation (each shuffle the rules prescribe, each move and each setting choice) reads and writes one player's own cards only;
• which operation is applied to a player's cards, and how that player's ±1 outcome is read from them, depends only on those cards, that player's own setting and on chance that, taken together with the deal, is independent of the settings;
• an outcome is recorded in every run, and no run is set aside;
• the settings are chosen independently of the deal.

The ch03 gloss states the joint condition. Each chapter line is byte-identical to its FULL mirror.

Evidence

1. General claim: a paper reduction for any deck and any finite sequence, in the track README:
   • admitted operations are products f_A × id or id × g_B, closed under composition;
   • private or shared chance is dealt in advance as part of Ω′ = (deal, all chance), jointly independent of the settings;
   • responses are q_a(Ω′), r_b(Ω′);
   • Main §3.3 then gives |CHSH| ≤ 2.
2. Finite instance: card_bell_check.py (25 checks, all exact), an exhaustive four-card instance, not the proof.
   • Countercontrols that must reach 4, and do: X1, X1′ (the reviewer's counterexample), X2, X3, L1, L2, P2, C1, C2, C3.
   • Admit controls that must stay ≤ 2 or admit, and do: P1, C4 (shared setting-independent chance), ADM2, E3, O3a.
   • A PIN check that both sentences appear in their chapters and in FULL.

Checks

These were re-run by the reviewer under python3 -I:

• card_bell_check.py: 25 PASS, output byte-identical to the committed .out.
• coverage_check and edge_rigidity_probe pass.
• release_gate passes every step except lean-axioms, which runs in CI.
• Book: 559 pages. The extra page is reflow from the longer sentences; the median characters per page are steady.

Cross-unit

• Trial merges with SC-33: Bell placement in Main and the book (ontic parameter dependence; ceiling hypotheses) #968 (SC-33) and SC-17 (Explainer): Bell at Main §3.3's scope, the dynamics' form fixed up to one coefficient, chirality without the memory claim #963 (SC-17) are clean on every .md. FULL .tex/.pdf must be rebuilt after either lands.
• For r32: the Introduction's card sentence (current L188) must carry the same five clauses.

The per-claim record is in verification/audits/manuscript/intro-corpus-audit-20261026/tracks/SC39-card-bell/README.md.

No merge is authorized. Merging needs owner approval at an exact SHA, as a normal merge commit.

---

Generated by Claude Code

[amaybaum]

Lean kernel check failed on 0e235c3c (run 40636786901) in its install step, before any check ran. elan could not download lean-4.35.0-linux.tar.zst from releases.lean-lang.org because of an SSL connect error. This PR changes no Lean file and no workflow file.

The same download error has now hit #974 (04:45 UTC), #976 (05:09) and this PR (05:14). The Lean checks on #973 (04:42) and #975 (04:52) passed. So the failure is intermittent and comes from the toolchain download, not from any PR content.

Once this run finishes I will re-run the failed job once. If it fails again, I'll report it as an infrastructure blocker; it won't be treated as passing.

---

Generated by Claude Code

[amaybaum]

Closing without merge: fully absorbed into main. The owner approved this closure in the Workstream D survival audit, checked against main c596904a.

Both SC-39 card-game corrections are on main verbatim:

• ch01:286 = FULL:630.
• ch03:188 = FULL:1062.

No unrestricted card-game Bell claim remains in the corpus. The branch and history are kept.

---

Generated by Claude Code