CLAIM AND PROOF LEDGER | VERSION 1.0 | 9 NOVEMBER 2026

Contract common to C1-C9: initial Pr(theta=1)=1/4; independent binary read noise 1/4; read then independent flip h for each executed probe; at most two probes; no final transition/delay; history-based private bit P; independent disclosure flag probability lambda; action theta XOR E*P; expected mismatch plus c times probe count. Domain h in [0,1/2], lambda in [0,1], c >= 0. No support outside this contract is claimed.

C1 | Filtering, manuscript (3).
Basis: Bayes formula and known transition kernel, established machinery. Local proof gives exact B_0, B_1 and T_h. Prior antecedents: [2] Section II, [4] posterior process, [5] Sections 3-4. Timing order explicitly differs from [2] eq (10), without claiming generic POMDP exclusion.

C2 | Terminal risk, Lemma 1 and (4).
Proof: sealed branch MAP; disclosed branch mismatch iff theta=1. A point forecast is chosen before the independent reveal coin. This is a specialization of the stipulated evaluator, not a universal disclosure theorem.

C3 | Branch-0 dominance, Theorem 1 proof.
Certificate: one-probe reading-time error min(b,1/4) for b<=1/2; propagated sealed risk h+(1-2h)min(b,1/4). With b=b_0, sealed and disclosed continuation risks separately >= b_0; c>=0. Two subregions split at h=3/16; weak inequalities preserve endpoint ties.

C4 | Global frontier, (6)-(7).
Proof: five stopping shapes reduced by C3 and one-probe dominance; conditional minimization of terminal reports; random strategies are mixtures of contingent deterministic plans; compare remaining two risks. Counts n=11/8, R_A given by (10). Exact parameter domain includes boundary ties. Novelty status: exact formula derived here; historical priority unestablished.

C5 | Endpoints.
Local algebra for h=0, h=1/2, lambda=1, c=0. Existence of an optimal representative is claimed, not uniqueness; redundant zero-cost probes may tie. At lambda=1,h=0,c=0 arbitrary forecasts need not be MAP but have equal disclosed error.

C6 | Harmful first check, helpful contingent plan.
Fixture h=1/16,lambda=1/4,c=1/128. First-only gross error 9/32; adaptive gross 225/1024 plus 11/1024 cost gives 59/256<1/4. Exact computation and analytic substitution. Classical nonmyopic-sensing antecedents [3],[7]; no historical novelty claim for the general phenomenon.

C7 | Mandatory continuation harms.
Conditional after 0 at h=1/16:3/20 ->31/160 (all lambda,c=0). Mandatory two at h=1/4,lambda=0,c=0:3/8 versus immediate 1/4. Different execution policies stop at different times. Available optional information cannot worsen the optimized objective because stopping remains feasible.

C8 | Policy/outcome certificate.
S_0 has 2 terminal bits; S_1 has 2+2^2=6; S_2 has 2+6^2=38. There are 64 product outcomes from 6 binary random variables. Unused probe variables are marginalized, not applied or revealed. check_frontier.py uses exact Fractions and 160 finite cases; proof_check.py is independently written. Tests corroborate, not prove a continuum theorem.

C9 | Action law and compiled matching.
Local conditional probability identity following (1). Explicit and compiled policies make identical choices under coupled random bits. No performance/resource separation claimed. [5] already supplies the conceptual equivalence; no new architecture claim.

C10 | Research and release scope.
Full historical priority not established. Monahan full text unavailable; [3] PDF format unsupported. Source comparisons specify exact reading scope. AI reviews are not human peer review. No data from a physical system, consciousness conclusion, generic self-model advantage, or general theorem beyond the finite contract is claimed. Release recommended only at technical-addendum scope, subject to recorded review dispositions.
