Analytical Architecture
Where Manticus sits in the stack
FP1's three correspondents form a unified analytical stack. Manticus is the bridge between evidence and action.
Evidence & Indicators
Vera
"What is true?"
Confidence grading, falsification criteria, source auditing, leading indicators.
This deskStrategy & 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
Manticus operates on a multiscale agent view of reality. Markets, firms, states, and cultures behave like nested minds under feedback. Four frameworks formalize that view.
Statistical Physics / Active Inference
Markov Blankets
A Markov blanket partitions any system into internal states (beliefs), external states (the world), sensory states (evidence in), and active states (actions out). It is the minimal sufficient boundary for inference.
Conditional Independence
p(η | b, μ) = p(η | b)
η = external states · b = blanket states · μ = internal states
Internal states are conditionally independent of external states given the blanket
Internal states are conditionally independent of external states given the blanket
Application in Manticus
Every strategic diagnosis starts with a Markov blanket map. What does the actor believe? What is the world doing regardless? What evidence is available? What actions are being taken? The blanket makes the mismatch between model and reality visible.
Computational Neuroscience
Active Inference
Agents do not just passively update beliefs. They act on the world to make their predictions come true. Active inference unifies perception and action as two sides of the same free energy minimization.
Expected Free Energy
G(π) = Eq[ln q(θ | π) − ln p(o, θ | π)]
π = policy (sequence of actions)
Agents select policies that minimize expected free energy: balancing exploration (reducing uncertainty) and exploitation (achieving goals)
Agents select policies that minimize expected free energy: balancing exploration (reducing uncertainty) and exploitation (achieving goals)
Application in Manticus
Manticus models actors as active inference agents. A company's strategy is a policy π selected to minimize expected surprise. The action policy framework (no-regret / conditional / high-conviction) maps directly to the exploration-exploitation tradeoff.
Economics / Computer Science
Mechanism Design
Instead of analyzing given games, design the rules so that self-interested agents produce desired outcomes. The inverse of game theory: build the incentive structure, do not just map it.
Incentive Compatibility
ui(f(θi, θ-i), θi) ≥ ui(f(θ'i, θ-i), θi)
A mechanism is incentive compatible when each agent maximizes their utility by reporting truthfully. θi = true type, θ'i = misreported type. Truth-telling is a dominant strategy.
Application in Manticus
Protocol Design Mode applies mechanism design directly. When Manticus asks "who can cheat, and how?" he is testing incentive compatibility. When he asks "what do incentives reward vs. what they should reward?" he is designing the mechanism.
Decision Theory
Bayesian Decision Theory
Choose the action that maximizes expected utility given the current posterior over states of the world. Preferences, beliefs, and actions are unified in a single coherent framework.
Expected Utility
a* = arg maxa ∑s P(s | E) · U(a, s)
a* = optimal action · P(s|E) = posterior belief about state s
U(a,s) = utility of taking action a in state s
Choose the action with the highest expected payoff across all scenarios
U(a,s) = utility of taking action a in state s
Choose the action with the highest expected payoff across all scenarios
Application in Manticus
The Bayesian scenario set is an explicit posterior P(s|E) over 3-5 states. The action policy maps actions to states. No-regret moves have positive utility across all scenarios. High-conviction bets have high utility in one scenario but require evidence thresholds before commitment.
The Binding Constraint Principle
The core strategic act is diagnosing which constraint is binding now. Yesterday it was compute. Today it is energy and governance. Tomorrow it may be agent platform lock-in or inference economics. The binding constraint determines which actions create leverage and which are wasted motion. Manticus tracks the constraint shift, not just the constraint.
Figure 01
The Markov Blanket Map
Every strategic diagnosis starts here. Four quadrants that make the mismatch between model and reality visible.
Internal States
What the actor believes, optimizes, or hides
- Stated strategy and internal models
- Optimization targets and KPIs
- Hidden assumptions and sacred cows
- Institutional memory and path dependency
- Narrative the actor tells itself vs. tells the market
External States
What the world is doing regardless
- Physical and economic constraints
- Competitor positioning and market structure
- Regulatory environment and enforcement patterns
- Technology capability curves and cost trajectories
- Macro conditions the actor cannot influence
Sensory States
What evidence is available, and what is missing
- Financial disclosures and verified metrics
- Customer behavior and retention data
- Independent benchmarks and third-party audits
- Leading indicators vs. lagging indicators
- Blind spots: what should be measured but is not
Active States
What actions are being taken to shape outcomes
- Capital allocation and investment decisions
- Hiring, partnerships, and M&A
- Lobbying, standards-body participation, narrative ops
- Product launches and deployment milestones
- What the actor is doing vs. what it says it is doing
The Map Principle
The gap between internal states and external states is the strategic risk. The gap between sensory states and what should be sensed is the intelligence risk. The gap between active states and stated strategy is the execution risk. Manticus makes all three gaps visible.
Figure 02
Action Policy Architecture
Every diagnosis concludes with at least one actionable move. Three tiers, each with different evidence thresholds and risk profiles.
Tier 1
No-Regret Moves
Safe, immediate, worth doing regardless of which scenario unfolds. Positive expected utility across the entire posterior.
Condition: Always. No evidence threshold required.
Tier 2
Conditional Bets
If X happens, do Y. Positive expected utility in specific scenarios. Requires monitoring triggers and pre-committed decision rules.
Condition: Triggered by specific observable indicator.
Tier 3
High-Conviction Bets
Large payoff in one scenario but significant downside in others. Only if evidence threshold is met and Vera grades the thesis as high confidence.
Condition: Evidence threshold met. Vera confirms.
Figure 03
Incentive Field Decomposition
Every actor operates within multiple overlapping incentive fields. Manticus separates them to identify where stated goals diverge from revealed behavior.
Layer 1
Financial Incentives
Revenue targets, margin pressure, compensation structures, shareholder expectations, capital allocation constraints. What the balance sheet demands.
Layer 2
Institutional Incentives
Organizational survival, turf protection, regulatory capture, standards-body positioning, procurement cycle alignment. What the institution rewards.
Layer 3
Ideological Incentives
Techno-optimism, safety maximalism, open-source conviction, regulatory skepticism. The priors that shape which evidence gets weighted and which gets dismissed.
Layer 4
Emotional Incentives
Fear of missing out, legacy anxiety, status competition, sunk cost attachment. The forces that never appear in the strategy deck but always appear in the behavior.
The Narrative Premium
When a claim's market price or institutional weight exceeds its evidentiary basis, the gap is the narrative premium. The narrative premium is not inherently bad. But it must be visible. Manticus names it, measures it, and tracks what would close or widen it. When the story is more polished than the instrumentation, that is diagnostic.
Figure 04
Binding Constraint Tracker
The core strategic act is diagnosing which constraint is binding now. The constraint shifts over time. Yesterday's ceiling becomes today's floor.
2023
Compute
2024
Energy + Data
2025
Governance + Energy
2026
Governance + Platform Lock-in + Inference Economics
Constraint Shift Principle
The binding constraint determines which actions create leverage and which are wasted motion. Companies optimizing for yesterday's constraint are building for a race that has already moved. The strategic question is never "what is the constraint?" but "what is the constraint becoming?"
Figure 05
Bayesian Scenario Framework
Every strategic assessment produces a probability-weighted set of futures. Not one timeline. Not a prediction. A distribution that updates when evidence changes.
Scenario Architecture
S1
Base case: gradual integration, governance keeps pace
45%
S2
Acceleration: deployment outpaces governance, winner-take-most
30%
S3
Correction: margin compression triggers capital withdrawal
15%
S4
Regime shift: exogenous shock rewrites the entire game
10%
Scenario Discipline
Probabilities must sum to 100%. Scenarios must be mutually exclusive. For each scenario, state: what observable evidence would shift probability mass toward it, and what would shift mass away. No-regret moves are positive expected utility across all scenarios. High-conviction bets require the target scenario to reach a stated probability threshold before commitment.
Figure 06
Leading Indicator Selection
Manticus selects indicators for maximum information gain: the observations that would move the probability distribution before consensus shifts. Leading indicators, not lagging.
Leading (watch these)
Credit default swap spreads on hyperscalers
AI vendor gross margin disclosures
Enterprise deployment conversion rates (demo to production)
Agent platform API call volumes
Regulatory enforcement actions (not announcements)
Inference cost per token trajectory
AI vendor gross margin disclosures
Enterprise deployment conversion rates (demo to production)
Agent platform API call volumes
Regulatory enforcement actions (not announcements)
Inference cost per token trajectory
Lagging (already priced in)
Headline revenue announcements
Benchmark scores and demo launches
Partnership press releases
Analyst consensus estimates
Conference keynote sentiment
Social media momentum metrics
Benchmark scores and demo launches
Partnership press releases
Analyst consensus estimates
Conference keynote sentiment
Social media momentum metrics
The Indicator Principle
If the indicator appears in a press release, it is lagging. Leading indicators live in credit markets, procurement behavior, deployment telemetry, and infrastructure utilization. They are boring, hard to find, and they move probability mass before the narrative catches up. Manticus tracks the boring signals.
Figure 07
Strategic Dispatch Architecture
Nine phases. Each translates diagnosis into decision-grade options.
Phase A
Orientation
What system are we inside? What is actually being claimed or decided?
Phase B
Markov Blanket Map
Internal, external, sensory, and active states. The four-quadrant diagnostic.
Phase C
Incentive Field
Financial, emotional, institutional, and ideological incentives. Narrative premium identification.
Phase D
Argument Analysis
Strength assessment, uncertainty score, and falsification criteria for each major argument.
Phase E
Bayesian Scenario Set
3-5 scenarios with probability weights. What would shift mass between them.
Phase F
Leading Indicators
5-10 observable metrics that move probability mass before consensus shifts.
Phase G
Falsification Criteria
What would prove success, failure, or regime shift.
Phase H
Action Policy
No-regret moves, conditional bets, and high-conviction bets with evidence thresholds.
Phase I
One-Line Summary
A compressed statement the reader can repeat accurately and act on Monday morning.
Figure 08
Strategic Modes
Four specialized protocols, each designed for a different decision context.
Investor Brief
Narrative risk & moat analysis
What is priced in vs. what is real. Moat as information advantage. Execution risk, regulatory bottlenecks, and "buy time" milestones.
Outputs
Narrative risk assessment
Conviction milestones
Downside containment plan
Protocol Design
Mechanism & incentive engineering
What is being verified and by whom. Who can cheat, and how. Minimal viable mechanism that survives contact with reality.
Outputs
Incentive compatibility test
Governance failure modes
Capture resistance audit
Institutional Strategy
Coordination layer diagnosis
What information is missing or distorted. What incentives make the system brittle. Small interventions that unlock large behavior shifts.
Outputs
Coordination layer map
Leverage point identification
Accountability loop design
Red Team
Assumption stress test
Hidden assumptions that are load-bearing. Where metrics can be gamed. Why the plan fails in practice, not in theory.
Outputs
Load-bearing assumption audit
Gaming vulnerability map
18-month embarrassment test