Triangulated Best-Judgement Assessment as Constrained Estimation: Conservation Structure, Corridor Operators, and the Statistics of Evidentiary Standing

This paper supplies the mathematical foundations of the two-pathway indirect audit framework of MWP-2026-02 and derives, from those foundations, the forensic properties that Commonwealth evidence law demands of a best-judgement assessment. The audited project is modelled as a control volume whose internal capital accumulation is bound by conservation to fluxes across instrumented boundaries — customs, banking, withholding, payroll and trade surfaces — so that each evidence stream is a boundary measurement inducing an upper bound on true cost. The supportable-cost determination is formalised as a corridor operator on these bounds and shown to be antitone in state-side evidence, isotone in taxpayer substantiation, idempotent, and 1-Lipschitz: an error in any single input propagates to the determination at most one-for-one. The stochastic analysis quantifies the downward (anti-taxpayer) bias of the minimum of noisy ceilings via order statistics, and converts the framework's graduated stringency into an explicit Bonferroni coverage guarantee and an asymmetric-loss quantile choice, giving the "known error rate" that Daubert-style admissibility analysis requests. Escalation is treated information-theoretically: corridor tightening is a monotone submodular set function of the evidence streams, so the platform's greedy request-next protocol is near-optimal with the classical (1 − 1/e) guarantee. Digit-distribution evidence is derived from scale invariance, and its forensic role is confined, by construction, to reliability and culpability. A final section maps each legal standard — the Trautwein rationality requirement, the civil balance of probabilities, the Makita exposure rule, the Daubert factors — onto a specific theorem, statistic or disclosed parameter of the framework, closing the distance between the mathematics and the courtroom. All numerical illustrations use the synthetic reference case; no real taxpayer data is used.