Alpha Capture

A Companion to Alpha by Design: Translating the Framework into Tools, Dashboards, and AI

The Alpha Process: Where Dashboards Fit

Before examining what dashboards can and cannot do, we must ground them in the alpha-generation process they serve. The complete framework is developed in Alpha by Design; here we summarize the lifecycle to establish context.

The Alpha Lifecycle

Alpha generation is not a single event but a process with distinct phases:

Where Dashboards Serve the Process

Dashboards do not participate equally in all phases. Their value concentrates in specific areas:

The Benchmark Reality: Why Active Risk Monitoring Matters

There is a second, equally important reason why dashboards are essential — one rooted in the business model of asset management itself.

Consider the math:

The implication is stark: alpha alone cannot meet client expectations. A PM who fully hedges systematic risk is left with only alpha — perhaps 2-3%. Clients paying fees expect absolute returns, not just benchmark-relative outperformance.

The Structural Consequence

This creates an unavoidable situation:

• PMs must intentionally carry systematic risk (beta) to generate absolute returns
• This systematic risk is exposure they have no edge on — they can't predict market crashes, sector rotations, or style shifts
• They're choosing to carry risk for return, not because they have insight

What This Means for Functions 4-5

This reframes Hedge Identification and Hedge Execution:

The practical questions become:

• "What systematic exposures do I have?" (Hedge Identification)
• "Are they larger than I intended?" (Monitoring)
• "How are those factors performing?" (Factor Performance)
• "Is factor drag corrupting my alpha?" (Alpha Corruption Monitor)
• "Is now a good time to tactically hedge?" (Hedge Execution)

Concrete Example: Alpha Corruption

Consider a PM whose stock picks are generating 3% alpha, but who has unintended Value factor exposure:

This is the core use case for Functions 4-5: not eliminating all systematic risk, but monitoring the systematic risk you're carrying and taking tactical action when it corrupts your alpha.

The Implication for This Document

Understanding where dashboards fit in the alpha lifecycle leads directly to the claims we'll make:

• Dashboards don't generate edge — because edge generation happens in Phase 1 (Discovery), where dashboards have minimal role
• Dashboards support edge management — because Phases 4-6 (Optimization, Monitoring, Harvest) are where dashboards excel
• The boundary matters — because conflating "shows useful data" with "generates alpha" leads to misallocated resources and false expectations

With this context established, we can now state the foundational principles precisely.

The Foundational Principle

Two Sources of Human Edge

Alpha originates from two distinct types of human judgment:

Red nodes generate alpha through belief mispricing. Gate nodes generate alpha through bet selection — filtering for situations where beliefs transmit reliably to outcomes.

Policy Constraints vs Mandate Constraints

Not all constraints are equal:

"I don't invest in poor management" is a policy constraint — learned judgment that filters for better outcomes. "I can't invest in tobacco" is a mandate constraint — an operating parameter. Both constrain the portfolio; only one generates alpha.

Claim 1: Real-Time Dashboard Data is Generally Unedgeable

Real-time data is largely symmetric — most participants see it simultaneously. Symmetric information is already priced. Therefore:

• Dashboards displaying real-time data rarely generate edge
• Derived data (calculations on symmetric inputs) rarely generate edge on their own
• AI operating on symmetric data alone faces steep headwinds

This is largely true but rarely acknowledged — because it challenges the value proposition of most data products.

Caveat: Speed advantages, superior processing, or unique data combinations can create edge even from "symmetric" inputs. But these are specific, defensible advantages — not the general case. Most users of dashboards and derived data do not have them.

Claim 2: Legitimate Dashboard Functions

If dashboardable data rarely generates edge, dashboards provide their clearest value for:

1. Understanding consensus — what does the market believe? (so you know what to diverge from)
2. Tracking diffusion — is your thesis being recognized? (where are you on the S-curve?)
3. Assisting hedging — what systematic exposures need neutralizing?
4. Self-calibration — how accurate have you been? (Ω measurement)

None of these generate edge. They support edge that exists elsewhere (in human beliefs).

The Boundary Condition

The boundary separates where edge is created from where edge is managed. This is both an architectural and epistemological distinction.

↓ BELIEFS + CONSTRAINTS ENTER HERE ↓
(PTA, scenarios, probabilities, red nodes, gate nodes, confidence)

↓ POSITIONS EXIT HERE ↓
(optimal weights, trade instructions)

The Conservation Law

The system transforms, measures, and deploys edge. It cannot create edge.

The app does two things with edge:

How the App Optimizes Edge (Without Creating It)

A Common Misconception: Does OPS Create Alpha?

It's tempting to claim that "Return at OPS - Actual Portfolio Return = Alpha Forgone." This framing implies that OPS itself generates alpha. It doesn't.

If an analyst has zero edge (Ω = 0, meaning their conviction adds no predictive value), then no amount of position sizing optimization will generate returns above market. OPS would just be optimizing noise.

The "alpha forgone" metric is real, but it measures capture efficiency, not edge generation:

Think of it this way: a perfectly calibrated scale doesn't make you heavier — it just tells you your true weight. OPS doesn't make your beliefs more accurate — it just ensures you're not leaving money on the table when your beliefs are accurate.

The Value Proposition

What the App Promises

• Capture your beliefs and constraint policies in structured form
• Filter opportunities through gate nodes (enforce admissibility)
• Optimize deployment (size, timing, hedging)
• Measure edge state (forward sharpe, α)
• Preserve edge (don't leak through bad sizing)
• Retire edge (sell discipline when exhausted)
• Score outcomes (close the learning loop)

What the App Does NOT Promise

• Find stocks for you
• Generate trading signals
• Replace your judgment
• Create alpha from data

The Stack: From Data to Decision

The app architecture is organized as a layered stack. Each layer builds on the one below, transforming raw inputs into actionable outputs. Understanding this stack clarifies where value is created and where it is not.

The Pre-AI Model: Deterministic Analytics

Before AI integration, the value chain was simpler:

In this model:

• Derived data was the intelligence layer — math-based, deterministic calculations that extracted insights from raw data
• Dashboards were the primary interface — humans consumed derived data visually
• No AI — all computation was formula-based, predictable, auditable

The AI-Enabled Model: Query and Synthesis

With AI integration, a new layer emerges:

In this model:

• AI sits between data and dashboards — can query across all data, synthesize insights, respond to natural language
• Dashboards become one interface among several — AI chat is another
• Derived data remains essential — AI queries it, surfaces it, but doesn't replace deterministic calculation

The Strategic Implication

This stack model frames the detailed layer descriptions that follow. Each layer is examined for: what it contains, what value it provides, and what it cannot do.

User Personas: Who Uses What

The app serves distinct user types, each interacting with different layers and requiring different capabilities. Understanding these personas drives interface design and feature prioritization.

Primary Personas

Analyst Workflow

Analysts are the source of edge. Their workflow spans the boundary:

Portfolio Manager Workflow

PMs deploy edge that analysts provide. Their workflow is entirely inside the boundary:

Risk Manager Workflow

Risk managers provide independent oversight. They consume data but don't create beliefs:

Persona-Specific AI Interfaces

Each persona interacts with AI differently based on their role:

Layer 1: Raw Data

Raw Data Requirements by Dashboard Function

Each dashboard function requires specific raw data inputs. This mapping clarifies what data must be captured, sourced, or integrated:

Raw Data for AI Tools (Outside Boundary)

AI tools that assist belief formation require additional data — primarily unstructured text and external signals:

Layer 2: Derived Data

Derived Data Requirements by Dashboard Function

Each dashboard function consumes specific derived metrics. This mapping shows where derived data creates value:

The Pre-AI Analytics Pipeline

Before AI, all intelligence came from deterministic transformations of raw data. This remains the foundation:

The derived data layer performs the "math" that extracts structured insight from raw inputs:

Layer 3: Dashboards

Why These Dashboards Exist: The Benchmark Conflict

Before examining specific dashboards, we must understand why managers need them. The answer lies in a structural choice most managers make — and the consequences that follow.

This creates a direct implication for dashboards:

With this context, the seven dashboard functions divide naturally:

• Functions 1-3: Support α discovery and diffusion monitoring (where edge lives)
• Functions 4-5: Support β identification and hedging (for the rare manager who hedges)
• Functions 6-7: Support scorekeeping and core analytics (table stakes)

For most managers, Functions 4-5 become monitoring tools rather than hedging tools — they watch systematic exposure and decide when to act, rather than hedging it away structurally.

Aspirational: The PM-Level Consensus BBN Generator

In Appendix C of Alpha by Design, we describe a Consensus BBN Elicitation Generator for analysts — a tool that provides company-specific node enumeration, consensus PDFs, and Jacobians as a starting point for edge discovery. The same architecture applies at the PM level for econometric/macro factors:

For the PM who doesn't hedge (most managers), this transforms monitoring:

Even without macro edge, this enables pre-computed reaction plans:

For the rare PM who does have macro edge, the generator provides the same workflow as the analyst: consensus baseline → identify divergence → articulate why → size the expression.

Purpose: Know precisely what the market believes, so you know what to diverge from.

Purpose: Track how consensus is evolving. Is it moving toward you (diffusion) or away (thesis risk)?

Purpose: Track whether YOUR specific thesis is being recognized and priced in.

Purpose: Identify systematic exposures where you have NO edge, monitor how they're impacting performance, and decide when to hedge.

Purpose: Once unintended exposures are identified, show how to neutralize them.

Hedging Instrument Menu

Purpose: Track actual performance against stated beliefs. Close the learning loop. Allocate capital to skill.

Analyst Scorekeeping

PM Scorekeeping

Organizational Scorekeeping

The Power of Cross-Firm Benchmarking

Internal scorekeeping tells analysts and PMs where they stand within their firm. But a more powerful question is: where do they stand relative to peers across the industry?

Cross-firm benchmarking — comparing analyst and PM metrics across multiple organizations — unlocks a distinct category of value. The logic follows directly from the framework's core distinction between what is trainable and what is not.

The limitation of internal benchmarks:

What Cross-Firm Comparison Can Improve

What Cross-Firm Comparison Cannot Improve

Critically, cross-firm benchmarks cannot:

• Tell you what to believe about a specific stock
• Create edge where none exists
• Make your insights correct
• Generate the alpha-producing idea itself

This is the same boundary we've established throughout the framework: infrastructure is trainable; edge is not.

The analogy to consensus is exact: you need to know what others believe to know if you're different. Cross-firm benchmarking is the consensus on process quality — the baseline to improve from.

Purpose: Standard measurement infrastructure. Not differentiating, but essential.

Portfolio Returns

Returns Attribution

Risk Measurement

Analyst Win/Loss

Layer 4: AI

ML's Legitimate Domain: State Detection, Not Belief Formation

Before discussing AI inside vs. outside the app, a clarification on what machine learning can and cannot do in this framework:

This distinction matters because it defines where ML creates value:

• Pre-research: ML flags where to look (Diffusion Radar, anomaly detection)
• During research: ML accelerates information processing (summarization, extraction)
• Post-belief: ML monitors state (is your thesis being priced in?)

At no point does ML form the belief that constitutes edge. The human analyst remains the source of causal reasoning about company-specific outcomes.

For treatment of factor strategies and systematic approaches within this framework, see Sections XIV-XV of Alpha by Design.

AI Outside the App: Thought Partner

This AI helps you FORM beliefs before they enter the app.

Alpha Theory Tools for Edge Discovery

Because edge is created outside the app boundary, Alpha Theory can provide tools, algorithms, and trained ML models that assist analysts in the belief formation process — without violating the framework's principles. These tools don't generate edge; they accelerate and discipline the human process that creates edge.

Diffusion Radar: A Special Case

The Diffusion Radar deserves special attention because it uses ML to detect that someone's thesis is being priced in — without knowing what that thesis is.

What it detects:

• Abnormal drift without public news
• Cross-asset inconsistencies (equity vs credit vs options)
• Microstructure changes (order flow, volume patterns)
• Estimate revision clustering
• Option skew changes before spot fully adjusts

What it outputs: "Diffusion signature detected in [TICKER] — worth investigating."

Why it's legitimate:

• ML can learn diffusion (it's a market property, not company-specific)
• Detection is mid-cycle — some alpha typically remains
• Human still has to discover the underlying thesis
• It's a radar, not a signal — tells you where to look, not what to believe

The analyst's job is to investigate why diffusion is occurring. If they can identify the underlying thesis and form their own view, they may capture residual alpha. If they can't identify a valid thesis, they pass.

The Edge Discovery Workflow

A complete pre-app workflow might look like:

1. Opportunity Flagging: Diffusion Radar or Consensus Mapper surfaces candidates
2. Research Intake: Synthesizer processes latest filings, calls, and research
3. Consensus Mapping: Analyst sees where street stands on key drivers
4. Red Node Elicitation: Interactive session surfaces analyst's divergent views
5. Causal Chain Construction: Analyst articulates why their view leads to different outcomes
6. Assumption Challenge: Tool stress-tests the thesis; analyst refines or abandons
7. Evidence Documentation: Supporting evidence linked to each node in the chain
8. Belief Capture: Structured output (PTA, scenarios, red nodes) enters the app

The output of this workflow is the belief input that crosses the app boundary — now well-articulated, stress-tested, and evidence-backed.

AI Inside the App: Infrastructure

What AI Inside CANNOT Answer

The Dashboard-AI Relationship: Defining the Search Space

A critical insight: dashboards define what AI can query. The "search space" for AI inside the app is exactly the data that dashboards display — raw data, derived data, and the relationships between them.

The Evolving Role: Dashboards Demoted, AI Promoted

As AI capabilities mature, the relative importance of dashboards declines. This is not deprecation — it's role clarification.

What Dashboards Retain

Even as AI becomes the primary query interface, dashboards remain valuable for:

• Ambient monitoring — wall screens showing portfolio state at a glance
• Structured workflows — belief capture forms, scenario builders, approval queues
• Regulatory reporting — standard formats required by compliance
• Audit trails — visual documentation of decisions and states

What AI Takes Over

AI becomes the primary interface for:

• Ad-hoc questions — "Why did my alpha drop last week?"
• Cross-functional queries — "Which positions have high forward Sharpe AND declining consensus?"
• Synthesis — "Summarize my portfolio's thesis status"
• Natural language capture — "Add a bull case for NVDA at $180 with 30% probability"
• Contextual recommendations — "Given my exposures, what should I consider hedging?"

Implications for Development

Layer 5: Optimization Engine

Differentiation vs Table Stakes

Not all features create competitive advantage. The 7 dashboard functions fall into two categories:

Deliverability Assessment

Implementation Phases

Build on existing data to deliver core framework metrics:

• Forward Sharpe calculation and dashboard
• Factor-adjusted α by position and portfolio
• Empirical Ω (analyst accuracy calibration)
• B-L optimization with empirically-calibrated confidence
• Consensus comparison (analyst PT vs street)
• Core performance and risk dashboards

Add logic and polish to partial-delivery items:

• Hedge recommendations engine
• Position sizing workflow (OPS → ROPS)
• Thesis status classification (on track / at risk / played out)
• Backtesting infrastructure
• Implied expectations from price

Introduce new data capture for full framework alignment:

• Thesis documentation (red nodes, causal chain)
• Evidence milestones and falsification conditions
• PM systematic view capture
• Structured post-mortems
• S-curve position estimation
• Option-implied distributions

The Monitoring Philosophy

This blueprint encodes a fundamental shift in what investment professionals monitor:

IP and Defensibility: Where the Moat Lives

Not all components of this architecture are equally defensible. Understanding which elements constitute intellectual property — and which are commodity — shapes strategic investment and competitive positioning.

For the complete treatment of moat architecture and defensibility analysis, see the companion document Alpha by Design: Moat Analysis.

The Moat Stack

Defensibility increases as you move down the stack:

Proprietary Tools: What's Missing from the Marketplace

Several framework-specific tools have no market equivalent:

Where ML Transforms Tool Moat

Several "tools" are actually tool interface + trained ML model. The interface is shallow moat; the trained model is deep:

What Creates Defensible Advantage

The framework generates unique data at each pipeline stage. Aggregate that data across clients and you build models that improve with scale:

The Meta-Moat: Structured Belief → Outcome Data

The deepest moat comes from the unique data structure the framework creates:

The Network Effect

Each new firm that joins:

• Contributes to the benchmark pool (anonymized)
• Gets access to better benchmarks
• Makes the product more valuable for everyone

This is a classic data network effect moat. Competitors who copy the framework start with zero data. The historical data — years of beliefs linked to outcomes — compounds with each client.

Implications for Product Strategy

Summary

The Learning Loop

Form belief → Capture → Deploy → Monitor diffusion → Exit → SCORE → Learn → Form better beliefs
↑                                                                                           |
└────────────────────────────────────────────────────────────────────────────────────────────────┘

Tech Stack Summary