Analytical Architecture
Where Vera sits in the stack
FP1's three correspondents form a unified analytical stack. Each asks a different question. Together, they produce a complete assessment.
This desk
Evidence & Indicators
Vera
"What is true?"
Confidence grading, falsification criteria, source auditing, leading indicators.
Strategy & Calibration
Manticus
"What to do?"
Markov blankets, incentive maps, action policies, decision architecture.
Orientation & Sensemaking
Darśan
"What does it mean?"
Archetypal patterns, civilizational precedent, first principles.
Theoretical Foundations
The science behind the desk
Vera's analytical protocols are grounded in four formal frameworks from cybernetics, information theory, and computational neuroscience. These are not metaphors. They are the operating logic.
Computational Neuroscience
Free Energy Principle
Any self-organizing system that persists must minimize the divergence between its internal model and incoming sensory evidence. Surprise is costly. Accurate models are survival.
Variational Free Energy
F = Eq[ln q(θ) − ln p(o, θ)]
q(θ) = current beliefs (confidence distribution)
p(o, θ) = joint probability of observations and parameters
Minimizing F = minimizing surprise = updating beliefs toward evidence
p(o, θ) = joint probability of observations and parameters
Minimizing F = minimizing surprise = updating beliefs toward evidence
Application in Vera
Every Truthseeker Dispatch is a free energy minimization cycle. Vera takes incoming evidence (o), compares it against the current confidence distribution (q), and updates the assessment to minimize the gap. The confidence grades are the explicit representation of q(θ).
Statistical Physics / Active Inference
Markov Blankets
A Markov blanket is the boundary that separates a system's internal states from the external world. It defines what counts as admissible evidence and what constitutes noise.
Conditional Independence
p(η | b, μ) = p(η | b)
η = external states (ground truth)
b = blanket states (sensory + active)
μ = internal states (beliefs / assessments)
Internal states are conditionally independent of external states given the blanket
b = blanket states (sensory + active)
μ = internal states (beliefs / assessments)
Internal states are conditionally independent of external states given the blanket
Application in Vera
Vera's source hierarchy is her Markov blanket. It defines what evidence is admitted (Tier 1-2), noted but not weighted (Tier 3), and excluded as noise (Tier 4). The dispatches are active states. The confidence distribution is the internal state.
Cybernetics
Law of Requisite Variety
A regulator must have at least as much variety as the system it regulates. If the environment produces more variety than your instruments can absorb, you lose control of the assessment.
Ashby's Law
V(R) ≥ V(D)
V(R) = variety of the regulator (analytical modes)
V(D) = variety of disturbances (claims, narratives, noise)
The regulator must match or exceed the variety of what it faces
V(D) = variety of disturbances (claims, narratives, noise)
The regulator must match or exceed the variety of what it faces
Application in Vera
Four analytical modes, five epistemic categories, four source tiers, three confidence levels. This is requisite variety: the minimum instrumentation needed to regulate the information environment without losing resolution.
Probability Theory
Bayesian Inference
Rational belief updating: the posterior probability of a hypothesis given new evidence is proportional to the prior multiplied by the likelihood of the evidence.
Bayes' Theorem
P(H | E) = P(E | H) · P(H) / P(E)
P(H|E) = updated confidence after new evidence
P(E|H) = how likely the evidence is if the hypothesis is true
P(H) = prior confidence
P(E) = baseline probability of the evidence
P(E|H) = how likely the evidence is if the hypothesis is true
P(H) = prior confidence
P(E) = baseline probability of the evidence
Application in Vera
Every confidence grade is a posterior. "High confidence" means the product of prior belief and evidence likelihood places the hypothesis well above the decision threshold. The "update publicly" principle is transparent Bayesian updating.
Information Gain (Leading Indicators)
IG(E) = DKL[P(H | E) ‖ P(H)]
The KL divergence between posterior and prior measures how much evidence would move the assessment. Vera's leading indicators are selected for maximum expected information gain.
Figure 01
Vera's Markov Blanket
The source hierarchy defines the boundary between the world and the assessment. Nothing crosses the blanket without classification.
External States (η)
The World
Markets & Capital
- Asset price movements and volatility regimes
- Capital allocation shifts across sectors
- Credit spreads and sovereign debt dynamics
- Venture funding rounds and valuation trends
Technology & Deployment
- Production deployments vs. announcement milestones
- Inference cost curves and compute scaling
- Patent filings and standards-body decisions
- Open-source releases and capability benchmarks
Governance & Institutions
- Regulatory text, enforcement actions, legal rulings
- Trade policy shifts and export controls
- Institutional procurement and budget cycles
- Standards-body formation and membership
Physical Constraints
- Energy infrastructure capacity and grid load
- Supply chain throughput and bottleneck geography
- Labor market composition and skills pipeline
- Thermodynamic and material limits on computation
Unmeasured
- Latent risks not yet surfaced by instrumentation
- Second-order effects of policy or technology shifts
- Regime changes that invalidate baseline models
Sensory States (s)
- T1: SEC filings, earnings, peer-reviewed data
- T2: Analyst reports with primary citations
- T3: Media reports (noted, not weighted)
- T4: Noise (excluded at boundary)
Active States (a)
- Truthseeker Dispatches
- Indicator Dashboards
- Source Audits
- Claim Grades
- Red Flag Reports
Internal States (μ)
The Assessment
Belief Distribution
- Confidence grades (high / medium / low)
- Epistemic spectrum placement
- Posterior probability estimates per claim
- Narrative premium measurements
Predictive Model
- Leading indicator watchlist (ranked by info gain)
- Clock vs. compass separation per thesis
- Falsification criteria (confirm / break / regime shift)
- Priors carried from previous dispatches
Integrity Checks
- Source quality audit results
- Incentive alignment maps
- Missing evidence inventories
- Update history and prior revisions
Output Calibration
- Dispatch signature lines
- Cross-desk references (Manticus, Darśan)
- Prediction tracking for retrospective accuracy
→ Evidence flows inward through sensory classification← Dispatches flow outward as active inference
The Blanket Principle
The quality of Vera's assessment cannot exceed the quality of her Markov blanket. If the source hierarchy admits noise, the confidence distribution inherits that noise. If the blanket excludes signal, the assessment has a blind spot. The blanket is the instrument. Calibrate the instrument, calibrate the output.
Figure 02
Confidence Grading System
Every substantive claim receives an explicit posterior probability estimate, expressed as a confidence tag. This is Bayesian updating made operational.
High Confidence
Posterior > decision threshold
Independently verified. Multiply sourced. Consistent with physical or economic constraints. Counter-evidence examined.
Medium Confidence
Posterior near threshold
Directionally supported. Primary sources exist but incomplete. One new observation could shift the posterior.
Low Confidence
Prior dominates posterior
Plausible but unverified. Evidence likelihood is weak or contaminated by incentive alignment. High free energy.