Entimema
Contact us

Entimema Labs / 01 — Research remit

The questions behind financial decisions.

Entimema Labs investigates what happens when financial methodology, risk reasoning and AI-assisted systems must operate inside real decisions.

We formulate questions, develop methods and examine how selected elements can work in financial systems. The agenda remains open to scrutiny, exceptions and revision.

Explore the research agenda

02 / The investigative problem

What connects an analytical result to an accountable decision?

Data, models, reports and automation each provide part of an answer. The research question concerns their interaction: which evidence supports a claim, which rule can test it, and who should decide when uncertainty remains?

A reconciliation can establish arithmetic consistency without establishing meaning. A model can interpret meaning without establishing whether action is justified. Labs investigates those boundaries and the hand-offs between them.

03 / Research agenda

Three domains. Questions that cross their boundaries.

Financial controls, AI governance, traceability and human judgement run through the agenda. They are shared concerns, not separate programmes.

Financial Intelligence

How can heterogeneous financial information become an analytical model without losing accounting meaning?

Scope boundary. The research scope is wider than the current Income Statement workflow. Reporting, planning and forecasting topics are not all implemented product capabilities.

What we investigate

  • Canonical schemas, data normalisation and the meaning of periods, currencies and units.
  • Reporting, planning and forecasting logic connected to the decisions they support.
  • Reconciliation, evidence lineage and the point at which interpretation needs human review.

Selected existing work

Credit Risk

How should predictive risk models connect to decision strategy, monitoring and governance?

Scope boundary. This is methodological research. The publications do not imply that Entimema operates a production lending platform or a live credit-decision engine.

What we investigate

  • Scorecards, probability of default and calibration: what a risk estimate does and does not establish.
  • Portfolio monitoring, decision strategies and early-warning signals across the credit lifecycle.
  • Validation and explainability in relation to the decision a model is intended to support.

Selected existing work

Decision Systems

How can model interpretation, deterministic controls and human judgement form one auditable decision system?

Scope boundary. An architecture defines responsibilities; it does not, by itself, demonstrate accuracy, safe autonomy or empirical validation.

What we investigate

  • The relationships between source evidence, claims, assumptions and unresolved unknowns.
  • Confidence, materiality, exceptions and the conditions for escalation or abstention.
  • How rules and human review constrain AI-assisted workflows while preserving decision lineage.

Selected existing work

04 / How we investigate

A method for making the reasoning inspectable.

Research is valuable when it can withstand scrutiny—and still work inside a real decision.

  1. Observe

    Identify a recurring practitioner problem and the operating conditions in which it appears.

  2. Formalise

    Separate evidence, definitions, assumptions, calculations and judgement so each can be examined.

  3. Test

    Challenge methodology, calculations and controls through analytical examples and explicit failure modes.

  4. Operationalise

    Translate elements that meet their stated checks into workflows, rules or analytical systems, preserving the limits of those checks.

  5. Improve

    Use observed exceptions and validated corrections to identify where a method or implementation needs refinement.

Methodological position

Models interpret; rules control. Evidence, assumptions and unknowns must remain distinguishable, with consequential ambiguity available for human review.

Exceptions and corrections can expose a gap. They do not automatically constitute empirical research learning, nor does every correction train or improve an AI model.

Decision architecture

A decision is more than a model output.

Evidence, interpretation, deterministic control and professional judgement have different responsibilities. Follow the relationships; they are not a single automatic chain.

Select or focus a concept to inspect its connections. Every responsibility is visible without interaction.

supports / challenges Claim

may generate Hypothesis

may be investigated through Model

may require clarification or review Human Judgment

may surface a control exception Human Judgment

may surface material ambiguity Human Judgment

informs Decision

Evidence → supports / challenges → Claim
Claim → may generate → Hypothesis
Hypothesis → may be investigated through → Model
Unknown → may require clarification or review → Human Judgment
Rule → may surface a control exception → Human Judgment
Model → may surface material ambiguity → Human Judgment
Human Judgment → informs → Decision

Separate responsibilities.
A model does not validate its own financial controls. An unresolved unknown need not become a hypothesis.

Decision → retains the relevant path

  • evidenceEvidence
  • interpretationModel
  • controlsRule
  • judgementHuman Judgment

These are retained references, not instructions to repeat the process. A model output is not the final decision.

Read the architecture

The upper branch separates a source, an assertion and a testable proposition. Unknowns remain separate. Interpretation and controls can each expose a review need; the decision keeps references to its relevant path.

This is a methodological architecture, not a map of implemented Income Statement v1 features. The applied-system example below defines that workflow’s actual scope.

05 / Methodology in an applied system

Implemented example

Traceable Income Statement

Financial Intelligence / Income Statement v1

One bounded example of how selected methodology becomes an operational workflow, from a financial value to its evidence, checks and review.

  1. Research question

    What must be established before an extracted financial value can support analysis?

    Interpretation and control

  2. Methodology

    Preserve accounting meaning through canonical concepts, periods, currency and scale. Keep interpretation separate from proof of arithmetic.

    Financial data normalisation

  3. Implementation

    Income Statement v1 connects extracted values and canonical mappings to source and evidence references that can be inspected.

    Evidence lineage

  4. Control

    Deterministic reconciliation checks financial relationships. Readiness gates identify failed controls, missing evidence and unresolved material issues.

    Validation controls

  5. Human review

    Review tasks expose ambiguity and proposed mappings. Supported review decisions are recorded and the affected checks are recalculated.

    Confidence and review

Implementation boundary. The current workflow is for eligible English Income Statements in XLSX, XLSM, CSV or text-based PDF files. It does not implement the entire Labs agenda, and its existence is not evidence of empirical validation across other financial or credit decisions.

See Financial Intelligence in practice

Secure workspace · sign-in required

06 / Selected methodological work

The work behind the questions.

Published research makes the methods available for inspection. These selections connect to the agenda; they are not a catalogue of implemented capabilities.

Explore all research

07 / Open questions

What remains unresolved.

These questions guide further investigation. They are research directions, not promised features or claims of completed validation.

  • How much ambiguity can be resolved automatically before a financial interpretation requires human review?
  • How should confidence interact with materiality when a small error can change a consequential decision?
  • How should decision lineage remain inspectable as source data, definitions and models change?
  • How can financial and risk workflows preserve professional judgement while making routine execution repeatable?

08 / Explore

Inspect the method. Explore its application.

Entimema Labs is practitioner-led by Alexander Dimitrov.

Explore Entimema Research
See Financial Intelligence in practice

Secure workspace · sign-in required

About the Founder Start a conversation