DeOS AlphaBook

Chapter 05 · Part II · The framework: What decision optimization is

The Decision Optimization Framework

pages 105–11711 figures7 tables

Part II · The framework: What decision optimization is

Part II

1. The framework What decision optimization is#

Part II turns the case into a structure. One formula, Strategy + Intelligence + Platform; one hierarchy, solution, decision policy, analysis, analysis archetype; four decision policy types and eleven decision archetypes; and the handbook that shows how each level is built, with examples and screens.

Chapter 05The Decision Optimization Framework

What decision optimization is: strategy, intelligence and platform; solution, decision policy and analysis; the analysis catalogue.

Chapter 06Structure handbook

How solutions, decision policies and analyses are structured in DeOS, with archetypes, examples and screens.

Chapter 05 · The framework

Framework · Decision optimization

2. The decision optimization framework#

Strategy, intelligence and platform, and how solutions, decision policies and analyses fit together#

One equation, one four-step intelligence flow, five platform capabilities, four levels of granularity and one analysis catalogue: the common structure behind every DeOS use case.

01 · strategy 02 · intelligence 03 · platform
What we decide, and for what How we reason about it Where it runs and learns
End-to-end decision policy and mandate. Perceive, Diagnose, Predict, Recommend. Governance at every level; Context, Analysis, Recommendation, Learning

Summary

3. In brief#

Decision optimization has three parts and four levels. The parts say what it is made of: a strategy, an intelligence and a platform. The levels say how the work is cut: solution, decision policy, analysis and analysis archetype. Every DeOS use case is built from the same parts, at the same levels.

Decision optimization = Strategy + Intelligence + Platform.

Each part has to run end to end. A partial solution moves the weak link rather than removing it.

Strategy says what we decide, and for what.

The end-to-end decision policy plus the mandate, its goals and constraints. Every decision across the decision lifecycle is connected and governed by one mandate.

Intelligence says how we reason about it.

Hybrid intelligence in four steps: Perceive, Diagnose, Predict, Recommend. Every analysis runs from context to the recommended action and its explanation.

Predict answers will, could and would.

What will happen is the expected path. What could happen is the range of possible paths, and what would happen is the path each action or catalyst creates. Only the third connects a prediction to a decision.

Recommend turns a prediction into a decision.

It brings in the mandate, so the same prediction leads to different actions for different goals and constraints. One of four decision policy types answers: Selection, Policy optimization, Go / No-Go or Monitoring, always with an explanation of why.

Human and AI: three modes, set on purpose.

Every analysis states, per component, who does the work: Automated, Augmented or Judgement only. Repeated data-rich tasks are automated; where data and judgment both carry information, experts are fused with the model by their track record.

Four levels cut the work.

A solution is a decision lifecycle of use-case specific stages and decisions. A decision policy is one decision whose Recommend step combines several analyses. An analysis is one run of the intelligence flow. An analysis archetype is one reusable method.

One analysis catalogue, shared across use cases.

Analysis archetypes are reusable methods, each sitting in one of six analysis steps and grouped into families, with a default human / AI mode. The second use case starts from the first.

3 + 4parts and levels: Strategy, Intelligence and Platform, cut into solution, decision policy, analysis and analysis archetype.
5platform capabilities: Governance across every level, then Context, Analysis, Recommendation and Learning.
6analysis steps in the catalogue, from context engineering and causal diagnosis to contextual optimization and explanation.

01

Strategy

4. What we decide, and for what#

Strategy = End-to-end decision policy + Mandate. The strategy fixes which decisions the firm makes, how it makes each one, and what every decision has to achieve and respect.

End-to-end decision policy

The codified way the firm makes each decision across the whole decision lifecycle, and how the decisions connect: which decisions exist, in which order, what each one needs as input and what it hands on to the next. The decision lifecycle stages are specific to each use case; the framework prescribes no generic list.

Mandate

The set of goals and constraints every decision must respect. Goals say what we want to achieve: return, risk-adjusted value, impact, growth. Constraints say what we may not do or must stay within: risk budget, liquidity, capacity, regulation, ESG, concentration.

Strategy = end-to-end decision policy + mandate

Mandateapplies to every decision
GoalsReturnRisk-adjusted valueImpactGrowth
ConstraintsRisk budgetLiquidityCapacityRegulationConcentration

the mandate sets the goals and constraints of each decision

  1. Policy 1Decision 1

    use-case specific

  2. Policy 2Decision 2

    use-case specific

  3. Policy 3Decision 3

    use-case specific

  4. Policy 4Decision 4

    use-case specific

  5. Policy 5Decision 5

    use-case specific

  6. Policy 6Decision 6

    use-case specific

  7. Policy 7Decision n

    use-case specific

End-to-end decision policy: every decision in the decision lifecycle, in sequence, each handing its result to the next

Each use case defines its own decisions and their order.

Figure 2 Strategy end to end. The mandate applies to every decision; the decision policies run in sequence across the decision lifecycle. Which decisions, and

in which order, is defined by each use case.

Why both. A policy without a mandate optimizes toward nothing; a mandate without a policy is a wish that cannot be applied consistently. Two firms with the same forecast should often act differently, because their mandates differ. The mandate enters every analysis at contextual optimization.

02

Intelligence

5. How we reason about it#

Intelligence = Hybrid intelligence. The firm's data and its experts' knowledge, combined with purpose-built AI, running one fixed flow of four steps: from what we should look at to what we should do about it, and why. The four steps are the four components of every analysis; for building, they split into six analysis steps.

INTELLIGENCE = HYBRID INTELLIGENCE: ONE ANALYSIS, FOUR STEPS 01 02 03 04 Perceive context Causal diagnosis Counterfactual prediction Contextual optimization What drives outcomes? What should we look at? What will, could, would happen? What to do about it, and why? 1 Context engineering: 2 Causal dia INTELLIGENCE = HYBRID INTELLIGENCE: ONE ANALYSIS, FOUR STEPS 01 Perceive context What should we look at? 1 Context engineering: entities, comparables, universe, base rates 02 Causal diagnosis What drives outcomes? 2 Causal diagnosis: true drivers, regime, value and where we stand 03 Counterfactual prediction What will, could, would happen? 3 Probabilistic prediction and fusion (will); 4 Counter- factual simulation (could, would) 04 Contextual optimization What to do about it, and why? 5 Contextual optimization: what · 6 Explanation: why, for our goals and constraints Every outcome feeds the next analysis: the scope, the diagnosis and the predictions are updated HYBRID: the firm's data + its experts' knowledge + purpose-built AI, at every step
Figure 3 The intelligence flow. One analysis runs the four components in order; every outcome feeds the next analysis.
Step Component Question Analysis steps Platform
1 Perceive Perceive context What should we look at, and what is it comparable to? 1 Context engineering Context
2 Diagnose Causal diagnosis What really drives outcomes here, and where do we stand? 2 Causal diagnosis Analysis
3 Predict Counterfactual prediction What will, could and would happen? 3 Probabilistic prediction and fusion (will); 4 Counterfactual simulation (could, would) Analysis
4 Recommend Contextual optimization What should we do, given our goals and constraints, and why? 5 Contextual optimization (what); 6 Explanation (why) Recommendation

Not every analysis runs every analysis archetype, but every analysis has all four components. An analysis whose output feeds another decision policy may end at a light Recommend, a flag, a rank or a threshold, with the full optimization done in the policy's own Recommend step.

Intelligence · Step 3

6. Will, could, would#

A counterfactual prediction answers three questions, not one. What will happen is the expected path, from probabilistic prediction and fusion. What could happen is the range of paths that are possible, and what would happen is the path each action or catalyst creates; both come from counterfactual simulation. Only the third connects a prediction to a decision.

STEP 3 · COUNTERFACTUAL PREDICTION: THREE QUESTIONS would WILL What will happen? The expected path, if we do nothing new. will What could happen? COULD The alternative paths that are possible: scenarios, regimes, tails. N O W What would happen if we act? WOULD The counterfactual of each action we co STEP 3 · COUNTERFACTUAL PREDICTION: THREE QUESTIONS N O W would will WILL What will happen? The expected path, if we do nothing new. COULD What could happen? The alternative paths that are possible: scenarios, regimes, tails. WOULD What would happen if we act? The counterfactual of each action we could take. Paths simulated for illustration.
Figure 4 Three questions in one prediction. Will is the expected path, could the paths that are possible, would the path an action creates.

Intelligence · Step 4

7. Recommend: four decision policy types#

Step 4 is where a prediction becomes a decision. It brings in the mandate from the strategy, so the same prediction leads to different actions for firms with different goals and constraints. It answers one of four question shapes, always with an explanation of why.

STEP 4 · RECOMMEND: FOUR DECISION POLICY TYPES

  1. TYPE 1

    Selection Which options? A ranked, scored shortlist, trading off several goals and constraints at once

  2. TYPE 2

    Policy optimization Which actions? The best action or setting: price, lever, timing, structure, weights, triggers

  3. TYPE 3

    Go / No-Go Yes or no? A binary decision with its conviction threshold and the terms under which it holds

  4. TYPE 4

    Monitoring What should trigger action? Signals, thresholds and triggers that reopen the decision

Inside one decision policy the types often chain: pick the options, set the terms, decide, then watch what reopens it.

Figure 5 The four decision policy types. Every decision policy has one type; inside an analysis the same shapes are the modes of contextual optimization.

Intelligence · Hybrid

8. Human and AI: three modes#

Hybrid intelligence is set on purpose. Every analysis states, per component and in detail per analysis archetype, who does the work: automated, augmented or judgement only.

THE HUMAN / AI CONTINUUM · ONE SCALE, SET PER COMPONENT OF EVERY ANALYSIS Human share of the work AI share of the work Judgement only Augmented Automated People decide; DeOS records the judgment DeOS proposes; experts adjust and are fused DeOS runs it; people watch the exceptions WHO DOES THE WORK W THE HUMAN / AI CONTINUUM · ONE SCALE, SET PER COMPONENT OF EVERY ANALYSIS Human share of the work AI share of the work Judgement only People decide; DeOS records the judgment Augmented DeOS proposes; experts adjust and are fused Automated DeOS runs it; people watch the exceptions WHO DOES THE WORK People decide the component. DeOS captures the judgment in a structured form (ranges, confidence, rationale) and records it. WHEN IT FITS No data or precedent, a new situation, or accountability that must stay with a person (mandate trade-offs, final approval). HUMAN ROLE Decides and records the rationale EXAMPLES The final go / no-go of the investment committee; a relationship or integrity assessment WHO DOES THE WORK DeOS proposes and quantifies; experts add, adjust and are fused with the model by their track record. WHEN IT FITS Data and judgment both carry information; the stakes justify a human look. HUMAN ROLE Gives structured predictions, adjusts and signs off EXAMPLES Synergy capture likelihood fused from model and operating partners; contextual valuation WHO DOES THE WORK DeOS runs the component end to end. People set it up, watch its quality and handle exceptions. WHEN IT FITS Data is rich, the task repeats, errors are visible and cheap to correct. HUMAN ROLE Reviews exceptions and quality reports EXAMPLES Connecting a data room; a comparable- firms scan; imputation of missing financials Set per component: Perceive context · Causal diagnosis · Counterfactual prediction · Contextual optimization. Fixed before the analysis runs, calibrated from outcomes. Automate what is high in volume and low in ambiguity; augment where expert knowledge is scarce in the data; keep judgement only where accountability requires a person. The decision itself, at policy level, always stays with the named decision owner.
Figure 6 The human / AI continuum. Judgement only, augmented and automated are one scale, set per component of every analysis and calibrated from outcomes; the decision itself stays with the decision owner.
Mode What IT means When IT fits Human role Example
Automated DeOS runs the component end to end. People set it up, watch its quality and handle exceptions. Data is rich, the task repeats, errors are visible and cheap to correct. Reviews exceptions and quality reports Connecting a data room; a comparable-firms scan; imputation of missing financials
Augmented DeOS proposes and quantifies; experts add, adjust and are fused with the model by their track record. Data and judgment both carry information; the stakes justify a human look. Gives structured predictions, adjusts and signs off Synergy capture likelihood fused from model and operating partners; contextual valuation
Judgement only People decide the component. DeOS captures the judgment in a structured form (ranges, confidence, rationale) and records it. No data or precedent, a new situation, or accountability that must stay with a person (mandate trade-offs, final approval). Decides and records the rationale The final go / no-go of the investment committee; a relationship or integrity assessment

Automate what is high in volume and low in ambiguity; augment where expert knowledge is scarce in the data; keep judgement only where accountability, values or regulation require a person. The mode is set per component of every analysis (in detail per analysis archetype), before the analysis runs, and calibrated from outcomes. The decision itself, at policy level, always stays with the named decision owner.

03

Platform

9. Where it runs and learns#

Platform = Governance + Context + Analysis + Recommendation + Learning. Five capabilities, delivered as one system: DeOS. Governance spans every level and every step; the other four run the intelligence flow and close the loop.

PLATFORM = GOVERNANCE + CONTEXT + ANALYSIS + RECOMMENDATION + LEARNING Governance Codifies the solution, every decision policy and the mandate; versions and approves 01 analyses and their models; logs every decision, override and reason; auditable and replayable spans every level and every step 02 0 PLATFORM = GOVERNANCE + CONTEXT + ANALYSIS + RECOMMENDATION + LEARNING 01 Governance spans every level and every step Codifies the solution, every decision policy and the mandate; versions and approves analyses and their models; logs every decision, override and reason; auditable and replayable 02 Context Ontology, entities, drivers and causal relationships 03 Analysis Causal diagnosis and counterfactual prediction; full outcome distributions 04 Recommendation Actions optimized against goals and constraints, explained 05 Learning Outcome vs prediction; decision quality scored; models and policies updated Learning closes the loop: every outcome updates the policies, the context and the models DeOS: one system, modular, the firm's own infrastructure, learning for that firm only
Figure 7 The five platform capabilities. Governance spans solutions, decision policies and analyses; Learning closes the loop.
Capability What IT does Where IT runs
Governance Codifies the solution (its decisions and their order), every decision policy and the mandate; versions and approves analyses and their models; logs every decision with its context, recommendation, explanation, human decision, override and reason; makes everything auditable and replayable. Across all levels (solution, decision policy, analysis) and all steps; stores the output of 6 Explanation
Context Builds and maintains the model of the environment: data ontology, entities, drivers and the causal relationships between them. 1 Context engineering
Analysis Runs the causal diagnosis and the counterfactual predictions, and produces outcome distributions, including the tails. 2 Causal diagnosis, 3 Probabilistic prediction and fusion, 4 Counterfactual simulation
Recommendation Optimizes actions against goals and constraints and explains the ranking, with alternatives and conviction. 5 Contextual optimization, 6 Explanation
Learning Compares realized outcomes with predictions, scores decision quality separately from the outcome, and updates beliefs, models and policies. After the outcome; feeds back into 1

04

Structure

10. Solution, decision policy, analysis, analysis archetype#

Four levels of granularity, each made of the one below it. A solution is the decision lifecycle of several decisions: use-case specific stages, a set of decisions and the stakeholders who make them. A decision policy is a single decision, and it combines its analyses into one recommendation through its own Recommend step. An analysis is one run of the intelligence flow, with four components. An analysis archetype is a reusable method within one analysis step.

Four levels: solution > decision policy > analysis > analysis archetype

Solutionstages, decisions and stakeholders
Stage 1
  1. Decision policy

    decision 1

  2. Decision policy

    decision 2

Stage 2
  1. Decision policy

    decision 3

  2. Decision policy

    decision 4

Stage n
  1. Decision policy n

    last decision

Stakeholders: deal and operating partners, IC · one mandate

Decision policyone decision
Analysisfour components

Decision policy 3 · e.g. acquire add-on target X?

mandate: goals and constraints of this decision

Analysis Ae.g. synergy
1 Scope2 Diagnose3 Predict4 Recommend
Analysis Be.g. price
1 Scope2 Diagnose3 Predict4 Recommend
Analysis Ce.g. financing
1 Scope2 Diagnose3 Predict4 Recommend
Policy recommend

combines them into one, as Selection, policy optimization, Go / No-Go or Monitoring

Each analysis runs the four components; the policy's own Recommend step combines them into one recommendation.

Here: one Go / No-Go on the add-on, at a price and structure, with its explanation.

Analysis archetypeone reusable method from the catalogue

Analysis archetypes used in analysis A (synergy), one or more per component

1 Perceive

Similar entity: codependency

2 Diagnose

Causal driver decomposition

3 Predict

Catalyst prediction; catalyst effect and impact

4 Recommend

Treatment optimization; contextual synergy

Figure 8 The four levels. A solution is a decision lifecycle of decision policies across its stages; each decision policy combines several analyses through its own Recommend step; each analysis runs the four components, built from analysis archetypes. Example from a private equity buy and build, illustrative.

Structure · Levels

11. Four levels, one example#

Level Definition Contains Owner (typical)
Solution The decision lifecycle of several decisions for one domain: use-case specific stages, each holding a set of decisions, run by named stakeholders under one mandate. Stages, decision policies, hand-offs, decision assets Solution owner (CIO, head of strategy, managing partner)
Decision policy One decision, formalized. It combines its analyses into one recommendation through its own Recommend step. Type, objective, owner, constraints, recommendation, analyses, process Decision owner (PM, deal partner, CFO, committee)
Analysis One run of the intelligence flow, built from four components: Perceive context, Causal diagnosis, Counterfactual prediction, Contextual optimization. Four components, each built from analysis archetypes, with a human / AI mode each Analysis owner (analyst, associate, operating partner)
Analysis archetype A reusable building block within one analysis step, from the catalogue (section 06). One method, such as contextual valuation, catalyst prediction or price optimization Decision architect

Example: one decision policy, three analyses#

Illustrative, PE buy and build: the solution is the platform's life from strategy to exit. One decision policy is "acquire add-on target X or not". Its analyses are synergy, price and financing, each running the four components with analysis archetypes from the catalogue:

Analysis Perceive context Causal diagnosis Counterfactual prediction Contextual optimization
Synergy Similar entity: codependency Causal driver decomposition Catalyst prediction (synergy capture); catalyst effect and impact Treatment optimization; contextual synergy
Price Peers; contextual base rates Contextual valuation; causal mispricing Contextual returns; stress testing Price optimization; contextual impact
Financing Similar situation: point in time Contextual regime Macro forecasting; ripple-through across levels Financing optimization; systemic response

The policy's own Recommend step combines them into one Go / No-Go on the add-on, at a price and structure, with its explanation.

05

How IT fits

12. One framework, three parts, four levels#

The parts of the equation and the levels of the structure describe the same system from two sides. The strategy is defined at the solution level and applied to each decision policy; the intelligence runs in every analysis; the platform carries all four levels, with governance across every one of them.

The equation across the four levels

StrategyIntelligencePlatform
SolutionEnd-to-end decision policy across the decision lifecycle; one mandate for all decisionsResults of each decision feed the next decisionGovernance of the decision lifecycle and the mandate; Learning across decisions
Decision policyOne decision, with the goals and constraints the mandate sets for itIts own Recommend step combines the analyses into one recommendationGovernance: policy, log, audit, override with reason; Recommendation
AnalysisTakes the goals and constraints of its policyPerceive context · Causal diagnosis · Counterfactual prediction · Contextual optim.Governance: versions, approval, model risk; Context, Analysis, Recommendation
Analysis archetypeOne method within one of the six analysis stepsThe capability that runs that step, under Governance
Figure 9 The equation across the four levels.
WHERE EACH STEP RUNS ON THE PLATFORM 1 Scope 2 Diagnose 3 Predict 4 Recommend Outcome 1 Context 2 Causal 3 Probabilistic 4 Counterfac- 5 Contextual realized vs 6 Explanation engineering diagnosis prediction, fusion tual simulation optimization predicted PLATFORM CAPABILITY Context Analysis Recommend WHERE EACH STEP RUNS ON THE PLATFORM 1 Scope 2 Diagnose 3 Predict 4 Recommend Outcome 1 Context engineering 2 Causal diagnosis 3 Probabilistic prediction, fusion 4 Counterfac- tual simulation 5 Contextual optimization 6 Explanation realized vs predicted PLATFORM CAPABILITY Context Analysis Recommendation Learning Governance: across every level (solution, decision policy, analysis) and every step; stores every explanation
Figure 10 Where each step runs on the platform. Perceive runs on the context capability; Diagnose and Predict on the analysis capability; Recommend on the recommendation capability; governance spans every level and every step.

Engineer the decision, end to end. The strategy says what to decide and for what, the intelligence reasons from context to the

optimal action, and the platform governs, runs and learns from every decision, so the firm gets better with every cycle.

06

Analysis catalogue

13. The building blocks of every analysis#

Every analysis is assembled from analysis archetypes: reusable methods, each sitting in one of the six analysis steps and grouped into families. The catalogue is shared across use cases, so the second use case starts from the first. The default mode is the usual human / AI split; each analysis states its own.

Family Analysis archetypes Typical output Default mode
1 Context engineering · Perceive context
Context representation Ontology, embedding, modelling Objects and relations; similarity space; causal graph of drivers Augmented
Context scan Similar entity: thematic, causal drivers, historic and predicted behaviour, codependency. Similar situation: trajectory, point in time Ranked comparables and past situations Automated
Context construction Investable universe, sector, theme, peers Eligible set; driver-based groupings Augmented
Context generation Imputation, synthetic data generation, contextual base rates Estimates with intervals; labelled generated histories; frequencies with sample size Automated
2 Causal diagnosis · Causal diagnosis
Valuation and mispricing Contextual valuation, causal mispricing Value distribution, decomposed; the price gap and its causes Augmented
Drivers and regime Causal driver decomposition, contextual regime Systemic beta, structural alpha, tactical alpha; regime with probability at four scopes (global, local, macro cycle, micro: universe and corporate lifecycle) Augmented
Standing Relative performance, causal rating, anomaly Gap to peers and base rate; rating on causal drivers; flag with severity Automated; causal rating augmented
3 Probabilistic prediction and fusion · Counterfactual prediction (will)
Prediction and fusion Knowledge fusion, contextual returns, default prediction, fundamental forecasting, macro forecasting One fused belief; return distributions; default probability by horizon; probability-weighted paths Augmented
Catalyst prediction Management, macro, market and fundamental catalysts Catalyst, likelihood, time horizon, effect size, counterfactual outcome Augmented
4 Counterfactual simulation · Counterfactual prediction (could, would)
Simulation Stress testing, scenario simulation, ripple-through across levels Loss in mandate units per scenario; outcome distribution per option; impact by level Automated; stress design augmented, scenario selection judgement only
Catalyst effect and impact Management, macro, market and fundamental catalysts Effect on the driver; impact on value, risk and portfolio, with against without Augmented
5 Contextual optimization · Contextual optimization (what)
Selection Ranking and selection Ranked, scored shortlist Augmented; criteria weights judgement only
Policy Price, treatment, timing, financing, Walk-away price; ranked levers; window; Augmented optimization allocation, portfolio construction, capital structure; capital split; position sizes; hedging hedges Go / No-Go Go / No-Go Yes or no, with conviction and terms Judgement only at the final call Policy optimization Price, treatment, timing, financing, allocation, portfolio construction, hedging Walk-away price; ranked levers; window; capital structure; capital split; position sizes; hedges Augmented Go / No-Go Go / No-Go Yes or no, with conviction and terms Judgement only at the final call Monitoring Monitoring Triggers and thresholds that reopen the decision Automated; thresholds judgement only

6 Explanation · Contextual optimization (why)

Attribution Contextual impact, contextual synergy, Driver weights and interaction effects Automated systemic impact, systemic synergy Response Contextual response, systemic Sensitivity curves, elasticities Automated response Archetype Turnaround target, archetype profiles Nearest improving comparabl Attribution Contextual impact, contextual synergy, systemic impact, systemic synergy Driver weights and interaction effects Automated Response Contextual response, systemic response Sensitivity curves, elasticities Automated Archetype Turnaround target, archetype profiles Nearest improving comparable and its path; matching prototype and its base rates Automated; turnaround target augmented

Every catalyst is predicted with the same five outputs: catalyst, likelihood, time horizon, effect size and counterfactual outcome, in four families: management, macro, market and fundamental. Catalyst prediction gives whether and when it happens; catalyst effect and impact simulates the world with it against the world without it. In Explanation, contextual means the entity's own drivers in its context, systemic broad forces shared across the system.

© vencortex®. This paper describes how decision optimization and DeOS structure decisions. It is not investment advice.

← Quant, quantamental and hybrid intelligenceStructure handbook →

Decision optimization with DeOS · vencortex® – DeOS | Decision Optimization System. Converted from DeOS_Compendium_Light.pdf by scripts/pdf_to_markdown.py; built by scripts/build_book_site.py.