Structure handbook

Handbook · How DeOS is structured

1. The structure handbook#

From solutions to single criteria: the anatomy of a decision#

How DeOS organizes decisions into solutions, decision policies and analyses, how every analysis runs through four components, and how every recommendation is built from conviction, options, actions, goals, a causal decision graph, criteria and insights.

Summary

2. In brief#

DeOS turns investment decisions into engineered, reusable and auditable objects. This handbook explains the structure that makes that possible, from the framework at the top down to a single decision criterion, with examples from long/short equity, credit, growth, private equity, activism and corporate capital allocation.

Framework and structure

3. The framework and the three levels#

DeOS is built on one framework and organized as one hierarchy. The framework says what decision optimization needs. The hierarchy says where each part of it lives.

Decision optimization = Strategy + Intelligence + Platform#

Each part has to run end to end. A partial solution moves the weak link rather than removing it: a sharp analysis is wasted if the decision that uses it has no mandate, and a good decision is wasted if nothing records and learns from it.

The five platform capabilities#

processes with guardrails and constraints; keeps the audit trail of which insight was created on what data, by which analysis and by which person; versions policies, analyses and recommendations; holds the decision assets and DeOS Studio.

Context

Builds the context through data connectors to the client's proprietary sources, internal systems and third-party data providers: data rooms, ontology, entities, drivers and comparables.

Analysis

Runs causal diagnosis and counterfactual prediction as one end-to-end analysis: regime-aware, causal and probabilistic foundation models propagate predictions from one model to the next, fused with expert judgment, with the human brought in at defined points.

outcomes on the optimization goal, conviction, ranked options, aggregation from asset to portfolio, and a fully explainable why chain from several perspectives.

Learning

Compares outcomes with predictions and updates beliefs, models and policies, on the client's instance alone.

How the three levels map onto the framework#

The three parts of the framework show up at every level of DeOS, with a different weight at each. Governance spans all three levels: solutions, decision policies and analyses.

The four components of every analysis#

Intelligence runs one fixed flow. Each step is a component of every analysis, and each runs on one platform capability. Diagnose runs on Analysis, not on Context: it estimates what drives outcomes, which is a modelling task.

The four decision policy types#

Every decision policy has one of four types: Selection, Policy optimization, Go / No-Go or Monitoring. The type sets the question its Recommend step answers and the shape of its output. The type sets what the policy's Recommend step delivers. Inside an analysis, contextual optimization uses the first three as its modes (Selection, policy optimization, Go / No-Go), and monitoring analyses set triggers and thresholds. Every output comes with an explanation of why.

The hierarchy#

The rule of the hierarchy is simple. A DeOS instance holds one or more solutions. Each solution holds use-case specific stages, each with a set of decision policies. Each decision policy holds one or more analyses, and every analysis is built from the four components, each assembled from analysis archetypes of the catalogue. Reusable components, such as data models, expert questionnaires, criteria and causal decision graphs, are shared across the hierarchy so that a new policy rarely starts from zero.

Optimization at any level#

Because every level is explicit, DeOS can improve decisions where the value is, not only where the data is. The three levels below are not alternatives; most clients start at one and grow into the next.

LEVEL 1 · ANALYSIS

A better valuation

The fair-value distribution of a mid-cap industrial is sharpened by stacking the analyst's model with AI forecasts of margins and multiples. The decision itself is unchanged; its main input is better.

Why sequences matter. Many investment outcomes are decided less by any single choice than by how choices interact: an attractive entry is wasted by a poor size; a correct thesis is wasted by an early exit. Optimizing the decision lifecycle makes those interactions explicit.

LEVEL 2 · DECISION

Better position sizing

Valuation, catalyst likelihood, crowding and liquidity are fused into one recommended size, with conviction, a predicted contribution to fund risk and the reasons behind it.

LEVEL 3 · DECISION LIFECYCLE

A better path

Entry, sizing, adds, trims and exit are optimized as one path: a starter position before results, an add on confirmation, a trim at the price where risk/reward falls below 2:1.

02

Solutions

4. Solutions: stages, decisions, stakeholders, mandate#

A solution is how DeOS shows up for a team: one decision lifecycle for one domain, made of use-case specific stages, each holding a set of decisions, run by named stakeholders under one mandate.

What a solution holds#

Stages. The stages of this decision lifecycle, in order. They are use-case specific: a buy-and-build fund has a thesis, a platform, an add-on pipeline, value creation and exit; a long / short book has idea selection, entry and sizing, monitoring and exit. There is no generic stage list.

A set of decisions per stage. One decision policy per decision, or several where a decision needs more than one type (for example a Go / No-Go and a financing policy at signing).

Stakeholders. The sponsor, the solution owner, a decision owner per stage, the contributors (analysts, experts, data and technology) and the control functions and recipients (risk, compliance, auditors, investors, lenders).

One mandate. The goals and constraints every decision must respect: return targets, risk budget, leverage, liquidity, concentration, ESG and regulation.

Hand-offs and decision lifecycles. What each decision passes to the next, and which decisions are optimized together as one path.

Decision assets and aggregation. The predefined, standardized components every policy reuses, so that the same decision runs the same way every time: policy templates, data ontologies, metrics, criteria libraries, causal decision graphs, expert prediction questions, data connectors and reusable analyses; and views across all decisions by sector, book, fund or vintage.

Stakeholders and mandate#

Example solutions and their stages#

The same structure carries very different domains. The examples below list typical stages and decision points of each solution. They are illustrations for the use case, not a fixed list; every decision point is served by one or more decision policies. Chapter 7 gives the full directory of solution archetypes and the end-to-end template.

Development

A solution can be as broad as a whole fund or as narrow as one stage of the process. Many clients start with a single solution around their highest-value decision, and extend it stage by stage. Because policies share assets, every new decision is cheaper to add than the last.

03

Decision policies

5. Decision policies: the main artefact of the Engineering phase#

A decision policy is one decision, formalized. It combines its analyses into one recommendation through its own Recommend step, and holds together everything needed to produce it: type, objective, owner, constraints, recommendation, analyses and process.

Why the policy is the central artefact#

In most investment organizations the way a decision is made lives in people's heads, templates and e-mail threads. It changes with who is in the room. A decision policy makes it explicit and reusable. That has four consequences:

The fields of every policy#

Every policy is specified with the same fields. The decision policy template in chapter 8 lists them with guidance, a filled example and a blank version.

How a policy combines its analyses#

The policy's own Recommend step takes the outputs of its analyses and turns them into one recommendation against the policy's objective and constraints. What it delivers follows from the policy's type: an Selection policy ranks options, a policy optimization policy sets actions, a Go / No-Go policy answers yes or no with the terms under which the yes holds, and a monitoring policy sets triggers and explains variance. A policy has one type but may draw on more: a Go / No-Go on an add-on can set price and financing by policy optimization in the same step. The recommendation is explained, and the named decision owner decides; every override is possible and recorded with its reason.

Context and process#

From policy to decision#

A policy is designed once and used many times. The same object moves through a short decision lifecycle, and every step is recorded.

Design

The policy is assembled in DeOS Studio from the archetype directory and the shared building blocks: graph, criteria, metrics, perspectives, actions, goals, analyses.

Publish

The published policy becomes the standard for its decision and can be reused as a template; a policy in use is changed only through a new version.

• Start a decision · One click instantiates the policy for a concrete case: data contexts, analyses, tasks and reviewers are set up.
• Run the process · Context, analysis and recommendation steps run with their owners; guardrails block submission while required inputs are missing.
• Decide and record · The decision owner decides; the recommendation, the decision, any override and its reason are stored point in time.

Learn

Outcomes are compared with predictions; findings change the analyses, the criteria or the graph in the next version.

04

Analyses

6. Analyses: four components#

An analysis is the smallest optimizable unit in DeOS and the place where everything connects. It is one run of the intelligence flow, built from four components: Perceive context, Causal diagnosis, Counterfactual prediction and Contextual optimization.

For building, the four components split into six analysis steps: context engineering; causal diagnosis; probabilistic prediction and fusion (will); counterfactual simulation (could, would); contextual optimization; and explanation. Not every analysis runs every analysis archetype, but every analysis has all four components. The DeOS Analysis Manual is the detailed companion to this chapter: twelve rules for a good analysis, the step-by-step method and the full analysis archetype reference.

Analyses are reusable and they chain. The same valuation analysis can serve a Go / No-Go policy, a Price policy and a Monitoring policy, each time with the same method and data definitions. Analyses can also depend on each other along a path: a regime analysis conditions the margin forecast, which conditions the earnings-revision path, which conditions fair value. DeOS keeps these path dependencies explicit, so uncertainty travels through the chain instead of being dropped at each step.

4.1

Analyses · Component 1 · Perceive context

7. Perceive context: as much as necessary, as little as possible#

Big data creates a paradox: endless information, but more noise than signal. Perceive context selects carefully what matters for the decision and what it is comparable to, and leaves out what only adds noise. It runs on the Context capability, through context engineering.

The goal is not more data. It is the smallest set of data that fully describes the decision.#

data from internal and external sources. The ontology maps it to real-world objects.

• Context scan · Automated or augmented selection of comparable entities and situations, building short and long lists that give analysts and models their context.
• Context construction · Builds the sets an analysis reasons in: investable universe, sector, theme, peers.
• Context generation · Generates context where data does not exist or is scarce: imputation, synthetic data, contextual base rates, simulated comparables in private markets.
• Feature engineering and representation · The processing steps that turn data into inputs, and the final decision-relevant representation of the asset and its situation that the next components consume.

The decision-centric data room#

A decision-centric data room starts from the decision, not from the data. It asks: from all the data available on this asset, what is required for the downstream analysis? That has a practical consequence. DeOS does not need all data integrated first before any value appears. It integrates what each decision needs, and the foundation grows with every decision. In DeOS the data room is organized as data contexts: business objects such as revenue model, debt, operations and taxation.

Data and sources#

One ecosystem, any provider. The vencortex ecosystem draws on third-party sources, and it does not lock a firm into any of them. The data is synchronised, harmonised and aligned across providers: the same company, the same field and the same point in time mean the same thing whichever provider delivered them, on one identifier and one definition. So it does not matter which provider a firm prefers. It can keep the one it already pays for, switch, or combine several, and every analysis, criterion and decision policy runs unchanged.

The ontology: from rows and columns to real-world objects#

The ontology works as a digital twin. It turns rows and columns into the objects a decision is about: companies, business units, product lines, assets, deals, people. It represents a company or asset in data at any level of granularity, independent of accounting standards: at the level of business units or product lines instead of the consolidated corporate; by geography, by industry sector or by thematic sector; and at mixed frequencies, combining high-frequency and low-frequency data.

Reusable foundation. Every decision adds to a structured data foundation that is reused by the next one.

Proprietary to the firm. Internal and external, structured and unstructured data are combined in one model that belongs to the firm.

One click into the analysis. Once mapped, data feeds the analyses seamlessly; a new decision fills its data room automatically.

Data loader and feature engineering#

The data loader runs the ETL pipeline that collects, cleans and loads data point in time, so every analysis sees exactly what was known at the decision date. Feature engineering then applies the processing steps that turn raw data into inputs: choosing the perspective (absolute, relative to history, relative to peers), building panels across entities and time, computing cross-sectional ranks, momentum, volatility, seasonality and anomaly features. The steps follow the same grammar as insights (chapter 17), so the features a model uses and the insights a human reads stay consistent.

Panel

Quarterly panel of 400 industrials over 15 years for margin and revision models

• Cross-sectional · Percentile rank of ROIC, valuation and short interest within the universe
• Time series · Margin momentum, revision velocity, volatility and seasonality-adjusted working capital
• Representation · A compact vector of the asset's characteristics, causal drivers and trajectory, used by the context scan and the AI skills

Context scan: grounding every analysis in comparable context#

The context scan finds the entities and situations that are truly comparable to the decision at hand. It grounds peer benchmarking, base rates and AI predictions in context, and makes them regime-aware.

Matching on inputs (characteristics and causal drivers) and reading the outputs (trajectories) is what turns peer groups into evidence. If eight of the ten closest set-ups re-rated within twelve months, that is a base rate the analysis can use and the expert can check their view against.

Why it matters#

Comparable benchmarking. Peer comparisons use peers that are actually comparable, not only those in the same sector code.

Contextual situations. Base rates come from situations like this one, not from all history.

Regime awareness. Predictions are conditioned on the regime the decision faces, estimated at four scopes at once: global, local, macro (the economic cycle) and micro (the investable universe and the corporate lifecycle). A model trained mostly on falling rates is not blindly applied in a rising-rate world, and a growth-stage playbook is not applied to a company in stress.

Context generation#

Sometimes the context does not exist in data. In private markets comparables are often scarce or undisclosed; for a new theme there may be no history at all. Context generation fills the gap, for example by imputing missing financials, simulating synthetic comparables from the causal drivers of known assets, deriving contextual base rates, or generating counterfactual situations from the causal decision graph. Generated context is always labelled as such and carries wider uncertainty.

4.2

Analyses · Component 2 · Causal diagnosis

8. Causal diagnosis: what really drives outcomes#

Causal diagnosis separates the true drivers from correlations and noise, and says where the asset stands against its context. It runs on the Analysis capability: estimating drivers is a modelling task, not a data task.

Analysis archetypes of the catalogue#

AI skills for diagnosis#

Diagnosis is carried by purpose-built AI skills (chapter 4.3), guided by the causal decision graph and checked by experts where the stakes justify it.

• AI SKILL · FINANCE EXAMPLE
• Anomaly detection · Flag that DSO jumped 18 days against a stable history
• Clustering · Group issuers by similar leverage and cash-flow behavior
• Regime models · Identify that the market is in a late-cycle, falling-rate regime
• Factor analysis · Decompose a position's return into market, sector, style and specific
• Causal driver estimation · Estimate how much input costs drive gross margin for this company
• Mispricing · Gap between market-implied and predicted fundamentals

Why diagnosis comes before prediction#

The default human / AI mode of diagnosis is Augmented for valuation, mispricing, drivers, regime and causal rating, where the expert's view of the business adds information, and Automated for relative performance and anomaly, where the task repeats and errors are visible (chapter 4.5).

4.3

Analyses · Component 3 · Counterfactual prediction

9. Counterfactual prediction: human and AI fused#

Counterfactual prediction estimates what will, could and would happen. It is where human judgment and AI meet: expert prediction and purpose-built AI skills are fused into one calibrated signal.

Expert prediction: turning judgment into signal#

Human expertise and intuition are a crucial source of information, especially where data is thin, the situation is new or the answer depends on tacit knowledge. DeOS collects that judgment in a structured way and fuses it into one calibrated signal.

Many expert inputs, combined with AI. Analysts, PMs, operating partners and external experts each answer the same structured questions.

Every input carries a conviction or confidence level. Experts give a most likely value and a range, or a probability, not just a view, on fixed answer scales that are colour-coded red, amber and green.

Knowledge fusion. Inputs are fused taking into account each expert's calibration, historical performance and decisiveness. Signal is performance plus metacognition, building on the psychology of judgment and forecasting research associated with Daniel Kahneman, Philip Tetlock's superforecasters and Nate Silver.

The goal. Maximum knowledge aggregation, best signal extraction, and reduction of error and bias.

Typical expert questions#

• EXPERT INPUT · FINANCE EXAMPLE
• Risk judgment · How likely is a material downside surprise in the next four quarters?
• Investment time horizon · Over what horizon will the thesis play out?
• Target assumption · What EBITDA margin will the platform reach by exit?
• Hurdle rate · What minimum return does this risk profile require?
• Time to target · How many quarters until the margin target is reached?
• Effect size · How much will the price increase lift gross margin?
• Effect duration · How long will the pricing benefit last before competitors respond?
• Effect likelihood · How likely is the procurement synergy to be realized at all?
• Implementation cost · What will the ERP consolidation across add-ons cost?
• Synergy time to market · When will cross-selling synergies show up in revenue?
• Commercialization time · When will the new product line generate revenue?
• Customer and employee feedback · Structured feedback from customer calls and employee interviews, scored
• Target preference · Which of the shortlisted add-on targets does the deal team prefer, and how strongly?

Answer scales for expert questions#

Every expert question is answered on one of five scales. Each point on a scale has a value and a label, so an answer can be read without knowing the method, compared across experts and fused. The colour says how the answer bears on the decision: green supports it, amber is neither for nor against, red speaks against it. On a negative judgment such as risk or cost a high value is red; on a positive judgment such as quality a high value is green. Confidence and uncertainty are the same idea from two ends: confidence is most certain at 100, uncertainty at 1.

Probability

How likely is it? Events, catalysts, targets and effects.

■ green: supports the decision · ■ amber: neither for nor against · ■ red: speaks against it. For negative judgments (risk, cost) a high value is red.

What expert prediction delivers#

Standalone or fused. Expert predictions can be used on their own or fused with AI for downstream predictions within the analysis.

Tacit knowledge that stays. Each expert carries a latent model for each context. Structured collection captures performance and tacit knowledge that cannot easily be quantified, and it stays with the firm when people leave.

Better experts. It helps experts solve problems better, learn continuously from other experts and from AI, and learn about themselves ("in this context I am overconfident"), with nudges at the moment of judgment.

Skill separated from luck. Measured judgment makes it possible to separate true skill from luck, and to link incentives and bonus payments to it.

AI skills: purpose-built intelligence#

Skills chain along the path of the analysis. A regime model conditions the margin forecast; causal driver estimation tells the forecast which inputs matter; the catalyst skills sit on top. Because each skill reports a distribution, uncertainty is carried forward through the chain to the recommendation. Where experts and skills answer the same question, knowledge fusion weights each by its track record in this context.

4.4

Analyses · Component 4 · Contextual optimization

10. Contextual optimization and explanation#

The last component turns predictions into the best options or actions against the goals and constraints, in context, and explains every single prediction and recommendation.

Modes inside an analysis#

Inside an analysis, contextual optimization works in the modes of the first three decision policy types (chapter 1): Selection ranks options, policy optimization sets the best actions, Go / No-Go tests conviction against a threshold. Monitoring analyses set triggers and thresholds and attribute variance. An analysis that feeds a policy may end at a light recommendation (a rank, a flag or a threshold) and leave the full optimization to the policy's own Recommend step. The step ingests the goals and constraints of the policy and the mandate. The models used include:

• MODEL · ROLE
• Probabilistic Selection · Ranks options across many criteria, under uncertainty, into the probability that each option is best
• Conviction fusion · Fuses criteria into perspectives and perspectives into the decision conviction score, corrected for dependence between criteria
• Goal-based policy optimization · Finds the action values (price, size, leverage, timing, weights, triggers) that maximize the probability of meeting the goals within the constraints
• Go / No-Go threshold · Tests the conviction against the threshold and states the terms under which the yes holds
• Decision lifecycle optimization · Optimizes linked decisions such as entry, sizing, add, trim and exit as one path

An explanation for every prediction#

DeOS explains every single prediction, not only the final recommendation. Explanations come in three families, each answering a different question a decision maker asks.

The explanation types use the same causal and systemic vocabulary as insights (chapter 17), and they follow the causal decision graph (Part V), so a "why" at any level leads back to the same structure.

Optimizing a single analysis#

Each analysis is evaluated against realized outcomes. If the fair-value analysis is systematically too optimistic for cyclical names, DeOS shows it, and the weights between the analyst model, the expert inputs and the AI skills are recalibrated. Because policies share analyses, one better analysis improves many decisions.

4.5

Analyses · Human and AI

11. Human / AI modes per component#

Each component of every analysis states who does the work: DeOS, people, or both, in detail per analysis archetype. The mode is fixed in advance, so the split does not depend on who has time that week, and it is calibrated from outcomes.

Human role per mode#

Modes in practice#

Some analyses are best served by AI (forecasting revisions across 3,000 names), some by experts (judging management credibility), many by both. The mode is set per component, not per analysis: the scope of a management-credibility analysis can be automated while its prediction stays judgement only.

The analysis template (chapter 9) records the mode for each component, and the analysis family directory gives the default mode per analysis family. A mode moves only on evidence: a component moves from augmented to automated when its calibration shows that the human adjustment no longer adds accuracy, and back when it does.

Part i i

12. Decision policy types and decision archetypes#

Every decision policy, in every asset class, has one of four types. Within them, eleven decision archetypes recur so consistently that they can be standardized once and reused everywhere.

05

The four decision policy types

06

Examples by investor type

05

Decision policy types

13. The four decision policy types#

Every decision policy has one of four types: Selection, policy optimization, Go / No-Go or monitoring. They follow the decision lifecycle of an investment, and each answers a different question and produces a different output.

Decision archetypes#

Each type breaks down into decision archetypes: decisions with one clear objective that recur across asset classes. A type plus a decision archetype is a policy archetype, the starting point from which concrete decision policies are built. Chapter 8 turns them into the decision policy archetype directory, with the output and the typical chain of analyses for each.

Decision

Decision

optimization

Treatment

What operational levers must management pull?

Timing

What is the optimal window for execution?

How the pieces fit together#

Each type has a typical recommendation shape#

06

Examples

14. Examples by investor type#

The same decision archetypes appear in every investor type, with very different concrete decisions. The tables below show how. Rows marked as added extend the original framework with the Hedging decision archetype.

Long/Short Equity

Portfolio construction dominates: sizing, gross and net, factor balance.

Selection

Rank 60 names in a sector by risk-adjusted upside and downside and by catalyst density, to decide where analysts spend their time

Price

Define the entry zone and the price at which risk/reward falls below 2:1 (trim or exit)

Timing

Enter before or after earnings; time a short against an upcoming catalyst or lock-up expiry

Financing

Express the view through stock, options or swaps, and weigh borrow cost against financing cost

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Size positions by conviction, liquidity and correlation; manage gross/net and factor exposures

Decide how much of the sector beta and the momentum factor to hedge out, and with which instrument

Thesis-break triggers: estimate revisions, short-interest spikes, KPI misses

Decompose a losing position into market beta, factor, thesis error and timing error

Multi-Asset Credit

Allocation and portfolio construction both matter: the risk budget first, then issuer sizing.

Selection

Compare spread per unit of risk across IG, HY, loans, CLO tranches and EM to find the best relative value

Price

Set the minimum spread or OID, or the yield-to-worst floor, to bid

Timing

Primary vs. secondary entry; position ahead of refinancing walls or call dates; spread-cycle entry

Financing

Choose the tranche (senior, second lien, mezzanine); decide on fund-level leverage, repo or TRS; hedge FX and rates

Allocation

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Distribute the risk budget across rating buckets, sectors, duration and liquidity tiers

Size issuers by downside loss given default, correlation and liquidity-to-redemption match

Set the rates-duration and FX hedge ratios, and decide whether to buy CDS index protection into a spread-widening regime

Covenant headroom, spread widening, rating watch and liquidity triggers

Separate loss into spread move, default, recovery shortfall and carry

Fundamental Growth

Allocation between new names and follow-on reserves, then concentration in proven winners.

Selection

Rank high-growth companies by durability of growth, unit economics and reinvestment runway

Price

Determine the maximum entry multiple (e.g. EV/ARR) that still delivers the target IRR after growth decay

Timing

Enter after a drawdown; time pre-IPO, follow-on or post-lock-up entry

Financing

Primary vs. secondary, preferred terms, liquidation preferences, ratchets and convertible structures

Allocation

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Split capital between new investments and follow-on reserves

Concentrate in proven winners; size by growth durability and exit visibility

Hedge the long-duration growth factor against a rates shock while keeping the stock-specific exposure

Net revenue retention, cohort curves, burn multiple and growth decay vs. plan

Explain a growth miss: pipeline vs. churn vs. pricing vs. macro

Buy-and-Build Private Equity

Allocation dominates: add-ons vs. organic investment vs. deleveraging.

Selection

Choose the sub-sector to build in and the platform to anchor on; rank add-on targets by strategic fit and price

Price

Platform bid price, and the maximum add-on multiple that preserves arbitrage against the expected exit multiple

Timing

Cadence and sequencing of add-ons; when to stop acquiring and start integrating; exit window

Financing

Acquisition facility or DDTL sizing, equity vs. debt per add-on, earn-outs, rollover equity

Allocation

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Split capital between add-ons, organic investment and deleveraging, and across platforms in the fund

Concentration limits per platform and per vintage

Decide the interest-rate cap or swap level on the acquisition facility as leverage rises with each add-on

Pipeline conversion, integration milestones, leverage covenants, synergy realization

Decompose an EBITDA bridge miss into synergy leakage, organic decline and overpayment

Activist Investors

Portfolio construction is a concentration question: few campaigns, high conviction.

Selection

Screen for undervalued companies with fixable issues; rank by value gap against feasibility (ownership structure, board vulnerability)

Price

Accumulation price ceiling and the value that justifies the campaign cost

Timing

Accumulation window before disclosure thresholds; nomination deadlines; proxy season; post-earnings entry

Financing

Hold via common, options or swaps; fund the proxy-fight budget

Allocation

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Share of fund capital in campaigns; number of concurrent campaigns

Concentration in a few high-conviction campaigns vs. diversification

Hedge market and sector beta during a campaign that may run for 12 to 24 months, so the outcome reflects the ask

Management response, proxy-advisor stance, shareholder support; escalation triggers (private letter, public letter, proxy fight, settlement)

Why the stock did not re-rate despite concessions: wrong ask, wrong audience or execution failure

Corporate Capital Allocation

Allocation dominates: capex, M&A, R&D, dividends, buybacks, deleveraging.

Selection

Rank competing uses of capital (capex, M&A, R&D, buybacks) by ROIC, NPV and strategic fit

Price

Maximum acquisition price; buyback price ceiling relative to intrinsic value

Timing

Sequence capex, divestitures and buybacks against the cycle and funding conditions

Financing

Debt vs. equity vs. internal cash; leases; rating target; dividend policy

Allocation

Portfolio construction

Hedging

ADDED

Monitoring

Root cause

Split between growth capex, M&A, dividends, buybacks and deleveraging, and across business units

Balance the business portfolio by risk, cyclicality and return profile

Set the commodity, FX and interest-rate hedge ratios that protect the business case of a large project

Post-investment ROIC vs. return hurdle; kill triggers on underperforming projects

Explain the gap between planned and realized project returns: cost overrun, demand shortfall or execution delay

Part i i i

15. Archetypes, examples and templates#

Nothing in DeOS starts from a blank page. Each level has an archetype directory to start from, a filled example and a blank template. The solution template specifies a whole decision lifecycle end to end and points to the policy and analysis templates for every decision in it.

07

Solutions: directory and end-to-end template

08

Decision policies: directory and template

09

Analyses: directory and template

07

Solutions · Directory, example, template

16. Solutions: directory and end-to-end template#

A solution starts from an archetype: a typical decision lifecycle with its stages, key decisions and stakeholders. The solution template then specifies one solution end to end, stage by stage and decision by decision.

Solution archetype directory#

Ten solution archetypes cover most investment organizations. The stages are typical for the archetype and are adapted to the client; they are never a generic list.

Solution template for end to end#

The solution template is the entry point for specifying a solution. It fixes identity, mandate and stakeholders, lists every stage with its decisions, and points to one decision policy template (chapter 8) per decision and one analysis template (chapter 9) per analysis. Filled in, it specifies one solution end to end: nothing is left between two decisions.

Six steps from decision lifecycle to solution

Defining a solution means finding where in the decision lifecycle each decision is made, by whom and on what, before any policy is engineered. Steps 1 to 4 are one workshop per stage with its decision owner; steps 5 and 6 use the decision heat map and the audit.

Decision points hide in five moments. Walk each stage with its owner and look for them; each is a row in the decision map.

Terms

A price, size, structure, timing or route is set: offer price, leverage, exit window.

Watch

A signal is checked against a trigger and the owner must decide whether to act: early warning, thesis break, rebalancing.

Review

An outcome is judged and the next cycle changes: post-investment review, mandate review.

The decision map: which decision at which point#

The decision map is the first artefact of a solution and the source of its stage register: one row per decision point, with the stage, the moment and its trigger, the decision made there with its type and owner, what it needs from earlier decisions and what it hands on, the human / AI mode today and the target, and its place on the roadmap. The example is the PE buy-and-build platform used throughout this handbook.

One row per decision point. The decision column becomes a decision use case (one policy template each); the hands-on column fills the solution template's hand-offs; the roadmap column orders the decision portfolio. Example: PE buy and build, illustrative.

Example: the solution template filled for PE buy and build#

Proof

Calibration of synergy forecasts; add-on outcomes against case

Field

Example · PE buy and build

Roadmap

Add-on go / no-go first (L3), then the full decision lifecycle (L4)

Example: PE buy and build. Names, numbers and thresholds are illustrative, not client data or product output.

End to end: every stage, its decisions and their templates#

The core of the solution template is the map of stages and decisions. Each row is one decision policy, specified with its own policy template; each analysis named is specified with its own analysis template. The add-on Go / No-Go row is filled in full in chapters 8 and 9.

Example: PE buy and build. Names, numbers and thresholds are illustrative, not client data or product output. Decision lifecycles: the platform bid, its financing and the first add-on are optimized as one path.

Blank solution template#

Blank decision map#

Blank stage register#

Policy

08

Decision policies · Directory, example, template

17. Decision policies: directory and template#

The policy archetype directory builds on the four decision policy types and eleven decision archetypes of chapter 5. A policy archetype is a type plus a decision archetype. For each it adds the output of the Recommend step, the typical chain of analyses and example policies, so a new policy starts from a proven shape.

Decision policy archetype directory#

Read it with chapter 5, which gives each decision archetype its objective, and chapter 6, which gives concrete decisions per investor type. The analysis chains use the catalogue terms of chapter 9 and run in the order of the four components.

Example: the decision policy template filled for an add-on Go / No-Go#

The example is one row of the end-to-end map in chapter 7: the add-on Go / No-Go and price at stage 3 of the buy-and-build solution.

Field

Example · PE buy and build

identity

Type

Go / No-Go, with price and financing set by policy optimization · decision archetypes: conviction test, price, financing

Objective

Raise platform equity IRR by at least 2 pp per add-on within the leverage limit

Owner

Deal partner recommends; investment committee decides; operating partner signs the integration plan

Field

Example · PE buy and build

constraints

• Regulatory constraints · Merger control thresholds; clean-team rules before signing; LPA concentration limit; AI regulation classification
• Mandate constraints · Leverage at most 5.5x; add-on at most 25% of platform EBITDA

recommendation

Options

Acquire at ask · acquire at walk-away price · wait two quarters · pass

Actions

Bid range, financing mix, first three integration levers, review triggers

Insights

Target's cash conversion cycle 12 days longer than its contextual peers

analyses

process and record

Process

Perceive the context, run the analyses, recommend, IC decision

Example: PE buy and build. Names, numbers and thresholds are illustrative, not client data or product output.

Blank decision policy template#

When a policy template is complete#

09

Analyses · Directory, example, template

18. Analyses: directory and template#

Analyses are built from the four components, and each component from analysis archetypes of the catalogue. The directory groups them into analysis families, each with a default human / AI mode; the template specifies one analysis with its constraints, data and components.

Analysis family directory#

The default mode is where a component usually starts; where one analysis archetype differs, it is named. The mode is set per analysis in the template and moves on evidence (chapter 4.5). The DeOS Analysis Manual gives the full analysis archetype reference and the extended analysis template with dependencies, proof and value.

Example: the analysis template filled for synergy realization#

The example is one analysis of the add-on Go / No-Go policy in chapter 8. It is reused by the value creation plan and the exit case of the same solution.

Output

Synergy distribution: expected $4.1m run-rate EBITDA, 5% to 95%: $1.6m to $6.0m, by month 18

Example: PE buy and build. Names, numbers and thresholds are illustrative, not client data or product output.

Blank analysis template#

Part IV

19. The anatomy of a decision recommendation#

Every decision policy ends in a recommendation. Whatever the asset class and whatever the type, the recommendation is built from the same components, so it can be read, compared and audited the same way.

10

The recommendation at a glance

11

Level, conviction and perspectives

12

Options, ranking and actions

13

Optimization goals

10

Anatomy

20. The recommendation at a glance#

A DeOS recommendation answers four questions at once: what should we do, how sure are we, why, and what would change the answer.