Open Methodology

Transparency Principle

411bz believes that transparency about methodology builds trust and advances the field of AI visibility engineering. We publish the scientific foundations and mathematical principles that inform our work.

At the same time, the specific algorithms, weighting functions, normalization procedures, feature engineering, and threshold calibrations that compose the CPS engine are proprietary intellectual property of 411BZ COM INC. We guard them with the same conviction that Colonel Sanders guards his recipe.

Below is our open methodology — the science you can see. The engine you cannot.

CPS: Multi-Factor Probabilistic Model

Citation Probability Score (CPS) is a composite metric that estimates the likelihood of AI citation. It integrates multiple measurable input signals into a single probabilistic score ranging from 0.0 to 1.0.

Input Signals

• Extractability — Structural parseability of content for transformer-based systems
• Entity Clarity — Definitional stability and semantic consistency of named entities
• Structured Data Completeness — Machine-readable metadata coverage (Schema.org, JSON-LD, discovery manifests)
• Cross-Platform Retrieval Rate — Citation persistence across major LLM platforms
• Spectral Stability — Resilience of embedding position under model update cycles

Aggregation Method

Input signals are combined using a proprietary multi-factor weighting model. The aggregation method draws on principles from composite index construction in quantitative finance and multi-criteria decision analysis in operations research.

The specific weights, nonlinear transformations, and interaction terms are proprietary.

Scientific Foundations

Information Theory (Shannon, 1948)

Semantic entropy quantifies content structure unpredictability. Lower entropy correlates with more stable, coherent embedding representations. We measure entropy across topic distribution, terminological consistency, and structural hierarchy.

Statistical Mechanics

Embedding drift is modeled using principles from dynamical systems and energy-state transitions. Content with low semantic entropy tends to occupy lower-energy (more stable) positions in embedding space, analogous to energy minima in physical systems.

Bayesian Probability

CPS updates as evidence accumulates. Initial estimates are refined through cross-platform citation observations, weighted by source reliability. The posterior estimate improves over time.

Spectral Analysis

Drift detection uses frequency decomposition adapted from signal processing. Citation patterns are decomposed to distinguish normal variance from systematic shifts and competitive displacement.

Monte Carlo Simulation

Collapse probability is estimated through stochastic perturbation modeling — simulating thousands of embedding shift scenarios to produce probabilistic risk assessments.

Monitoring Cadence

• CPS recalculated every 6 hours
• Drift monitored continuously
• Cross-platform citation checked across 6+ major LLM platforms
• Collapse probability assessed on each monitoring cycle
• Spectral margin computed against competitive entity set

Governance Model

All automated actions are governed through three primitives:

• AGE™ — Approval-Gated Escalation (high-risk actions require authorization)
• CWAR™ — Confidence-Weighted Action Routing (decisions scored and routed by confidence)
• CPR™ — Context Persistence and Replay (interrupted processes resume exactly where they stopped)

The governance thresholds and routing rules are configurable per tenant. Default confidence floor: 0.80.

What We Do Not Disclose

• Exact CPS weighting functions
• Feature engineering specifications
• Normalization procedures
• Threshold calibration parameters
• Nonlinear transformation functions
• Interaction terms between input signals
• Proprietary drift detection algorithms
• Collapse simulation parameters
• Ghost Authority Cloud signal injection methods

These are trade secrets. We explain the physics. The recipe stays locked.