← Switch to Classic Register (Heuristic Rubric) Redmont Empirical Legal Methods · Research Monograph

Redmont Counsel Adversarial Latent Index

Bayesian Gaussian Graph Process with Laplace posterior estimation, pro se anchor baselines, and conservative lower-bound rankings (R = μ − 1.5σ).

Ranked Counsel Evaluated

Graph Match Edges Analyzed

Direct Adversarial Contests

Pro Se Anchor Baselines

k = 1.5

Conservative Credible Penalty

#	Counsel	Conservative	Mean μ	Uncertainty σ	90% Credible Range	Decided

Select any counsel row to view their empirical distribution.

Adversarial Matchup Simulator · Probit Predictive Distribution

Select any two counsel to compute their exact posterior win probability P(A ≻ B) = Φ((μ_A − μ_B) / σ_diff) accounting for covariance in the match graph.

Technical Monograph · The Bayesian Graph Laplacian Architecture

An adversarial network formulation of Redmont litigation: latent skill parameters, graph Laplacian precision, and conservative risk-adjusted standing.

§ 1 · The Directed Match Graph & The Pro Se Anchor

Litigation is not a sequence of isolated trials against an abstract benchmark; it is a sparse directed graph of paired adversarial contests. Let counsel be nodes i ∈ {1, …, M}. When two counsel meet on opposing sides of case m, they generate an edge with relative performance y_m = (v_A − v_B)/2 ∈ [−2, +2] and weight w_m = 0.5 + d_m/10.

y_m = (v_A - v_B)/2 in [-2, +2], w_m = 1/2 + d_m/10

The Pro Se Anchor Baseline: Roughly 40% of Redmont litigation involves defaults, unrepresented defendants, or abandoned complaints. Dropping them discards valuable evidence; treating them as real players distorts the graph. The system constructs an artificial anchor entity θ_anchor ~ N(−1.5, 10⁻⁴) pinned firmly at the lower bound. Beating an empty chair provides nearly zero Bayesian update (P ≈ 96% → Δθ ≈ 0), while losing to an unrepresented party triggers an immediate, catastrophic downward revision.

theta_anchor ~ N(-1.5, 10^-4), theta_i ~ N(0, 1.5)

§ 2 · Exact Posterior Closed-Form: The Graph Laplacian

Because both the observational likelihood and prior are Gaussian, the joint posterior distribution over the entire bar is analytically exact. No Markov Chain Monte Carlo approximation or convergence heuristics are required. The posterior precision matrix Q is the graph Laplacian weighted by case difficulty, regularized by the prior covariance Σ₀:

Q = X^T W X + Sigma_0^-1, Sigma = Q^-1, mu = Sigma(X^T W y + Sigma_0^-1 mu_0)

The diagonal entries σ_i = √Σ_ii yield the exact posterior uncertainty for each counsel. Notice the geometric property: counsel who face active, highly-connected opponents burn down their uncertainty σ_i rapidly, while counsel who only litigate isolated defaults remain saddled with high variance.

§ 3 · Risk-Adjusted Standing · The Uncertainty Discount

A single appearance establishes only that counsel prevailed on that specific afternoon. Under Bayesian estimation, unobserved skill is a probability density, not a settled rank. The register prices sample volatility directly into standing via a 93.3% lower-bound discount (k = 1.5):

R_i = 50 + 20(mu_i - 1.5*sigma_i) in (10, 90)

The Uncertainty Discount Doctrine:

The Volatility Haircut: A counsel winning their sole appearance against an absent player earns a high raw mean skill μ ≈ 0.8 (nominal 66). But with only one data point, the standard error is massive (σ ≈ 0.9). Their conservative index μ − 1.5σ ≈ −0.55 anchors standing at 37.2. The register assigns no laurels to unverified variance.
Evidence Accumulation: A veteran who has litigated contested trials against established counsel burns their standard error down to σ ≈ 0.35. Their conservative standing tracks their true skill tightly.
The Continuous Arena: Counsel cannot park an unblemished record on the shelf. Sustained rank cannot be stored; it must be defended across multiple adversarial trials.

§ 4 · Calibrated Predictive Probit Link

When Counsel A and Counsel B face each other, their expected win probability integrates over the joint posterior distribution, incorporating the off-diagonal covariance Σ_AB established by common opponents:

P(A beats B) = Phi((mu_A - mu_B) / sqrt(sigma_A^2 + sigma_B^2 - 2 Sigma_AB + 1))