Home · Working notes · Exploratory notes · About
Model type: Explanatory and Application Model
Scope: Construction, verification, compression, use, diagnosis, correction, and revision of explicit models
Status: Exploratory Note
Version: v0.1
Last updated: 2026-09-15
A model is a simplified representation of something.
It may help us orient ourselves, explain observations, organize knowledge, anticipate developments, compare possibilities, make decisions, or guide action.
Not every model is written down. Whenever we interpret a situation, expect an outcome, or choose an action, we already rely on some representation of how things work.
Making a model explicit does not make it correct.
It makes it possible to examine.
This note proposes a general process for constructing and maintaining useful explicit models:
Explore & Build → Verify → Compress → Use, Revise & Correct
The objective is neither maximal complexity nor maximal simplicity.
It is to preserve enough of reality to serve a defined purpose, expose important assumptions and uncertainty, identify the model’s limits, and retain enough structure for diagnosis and revision.
Models can describe physical systems, organizations, social situations, personal plans, or abstract ideas.
The underlying representation may be explicit and examined, or it may remain implicit and largely invisible.
An implicit model can work well when a situation is familiar, its consequences are limited, or errors are easily corrected.
Choosing between a red and a yellow pepper usually requires no explicit model. A preference, habit, or immediate choice is sufficient. The cost of deeper analysis would exceed its likely value.
The balance changes when the subject has greater consequences.
Explicit modeling becomes more useful when:
A model that takes several days to build may be excessive for a minor purchase and negligible for a commitment that shapes several years.
The required modeling effort should therefore be proportional to the significance of its intended use.
Make a model as explicit as its consequences require.
More modeling is not automatically better.
Additional structure is valuable only if it improves understanding, application, or the ability to detect error.
A model should not begin by perfecting one isolated component.
It should first reach a simple but structurally complete state.
A first stable version should address every necessary part of the model sufficiently, but not completely. Its purpose is to make the whole structure visible enough to inspect.
This prevents local optimization from hiding structural absence.
A detailed causal mechanism is of limited value if the model has no defined purpose. Precise evidence cannot rescue a distorted boundary. Elegant compression cannot repair an important missing relationship.
Once the whole structure exists, improvement can be directed toward the weakness that most limits the model’s usefulness.
The process is therefore not:
Perfect each part before continuing.
It is:
Construct a complete first structure, then improve its limiting weakness.
The first stable version is not the final model.
It is the first version sufficiently complete to be examined as a whole.
The process is organized into four phases:
These phases describe a logical structure, not a rigid chronological sequence.
Model building is iterative. Verification may reveal missing structure. Compression may expose unclear assumptions. Use may require a new boundary.
The phases identify what kind of work is being performed and what that work should produce.
The first phase constructs a sufficiently complete representation of the relevant part of reality.
Its purpose is not final precision.
Its purpose is to create a structure that can be examined.
Begin with what the model should help you understand or do.
A model may be intended to:
These purposes may overlap, but they are not identical.
A model designed to explain why something happened may not be sufficient to predict what happens next.
An orientation model may show where to look without explaining every underlying mechanism.
An application model must preserve enough structure to guide action.
The purpose determines which parts of reality are relevant.
A model of a company built to understand profitability will differ from one built to understand employee motivation. Both describe the same organization, but they preserve different structures.
A purpose can be clear and still be framed badly.
A question such as:
How can I maximize my income?
establishes a measurable purpose, but it may exclude health, dependence, risk, or long-term capability even when these factors materially affect the value of the resulting income.
The purpose does not need to contain everything.
But it should not exclude something that could reverse its own conclusions.
A poorly framed purpose can also contain a preferred explanation before the model has been built. The model may then become a justification for an existing position rather than an examination of reality.
The purpose should therefore be defined and tested before it determines the model’s structure and boundary.
A model without a defined purpose has no reliable stopping point.
Reality contains more detail than any model can preserve. Without a stopping condition, almost any additional detail can appear potentially relevant.
The appropriate stopping point is not completeness.
It is sufficient usefulness for the stated purpose.
Before expanding the model, establish what it must be able to do and how much confidence its intended use requires.
Build only as much model as its purpose requires—but enough to understand its assumptions and limits.
Identify the parts of the situation that matter for the stated purpose.
Depending on the subject, these may include:
A list of relevant elements is not yet a model.
The model begins to become useful when the relationships between them are made visible.
For example:
Higher demand does not automatically produce higher profit.
The result may depend on available capacity:
Higher demand → higher use of existing capacity → higher profit
Once the capacity limit is reached:
Higher demand → need for additional structure → higher fixed costs → temporarily lower profit
The same input produces different effects depending on the state of the system.
Preserving this relationship is more important than collecting many isolated facts about revenue and cost.
Do not begin by collecting every available fact.
Begin by asking which differences can alter what the model helps you understand or do.
The goal is not to reproduce reality.
It is to preserve the structure necessary for the model’s purpose.
A useful model removes detail without removing relevant differences.
Every model describes only part of reality.
The working boundary determines:
A model of short-term liquidity may treat business strategy as fixed.
A model of long-term viability cannot.
A model of one employee’s performance may treat organizational structures as context, while another model may need to make those structures central.
The boundary does not need to be permanent.
It needs to be appropriate for the current purpose and sufficient to construct the first model.
A boundary is inadequate when an excluded factor can regularly dominate the relationships inside it.
If this occurs, the factor must either enter the model or be stated as a condition under which the model no longer applies.
The boundary established during construction remains provisional.
Verification may later show that an excluded factor matters, that the time horizon is too short, or that the selected scale hides an important mechanism.
Models depend on conditions that may not be directly represented within them.
Some assumptions concern:
If these assumptions remain implicit, the model may appear more certain and more general than it is.
Not every assumption requires immediate proof.
Important assumptions should, however, be visible enough to examine, test, or revise.
The most consequential assumptions deserve the most attention.
An uncertain assumption that barely changes the result may be accepted provisionally.
An uncertain assumption that reverses the result may require evidence, experimentation, or an alternative model.
Making assumptions explicit also separates two sources of failure:
A model may therefore remain structurally useful even when a particular application fails because one of its assumptions was violated.
Not every uncertainty can be reduced in the same way.
Some uncertainties result from missing information.
Others require practical experience.
Some cannot be reduced at an acceptable cost or within the relevant time.
The answer exists and can reasonably be found through research, measurement, or consultation.
Examples include:
The appropriate response is to obtain better information.
The answer depends on interaction with reality and cannot be established through description alone.
Examples include:
The appropriate response is a test, prototype, trial period, or limited application.
The uncertainty cannot currently be reduced at a reasonable cost or with sufficient reliability.
Examples include:
The appropriate response is not endless analysis.
It is to:
This distinction prevents two opposite errors:
Construction creates a coherent candidate model.
Verification tests whether that model has a defensible relationship with reality.
A coherent model can still describe the wrong structure.
It must therefore be tested against:
Where direct testing is impossible, it should at least be clear what evidence would increase or reduce confidence in the model.
A model should be able to distinguish cases that produce different outcomes.
If meaningfully different situations receive the same explanation or prediction, an important variable, condition, threshold, or relationship may be missing.
The objective is not to defend the model.
It is to discover:
A model cannot become more reliable than the observations on which its critical assumptions depend.
Evidence should therefore be examined for:
An isolated observation may reveal a possibility without establishing a pattern.
Historical evidence may describe a system that has since changed.
A source may provide accurate information while still presenting only the part that supports its interests.
The strength of a conclusion should remain proportional to the evidence supporting it.
Verification is not a single final test.
It may send the model back into construction:
Model → Observation → Comparison → Structural adjustment
The model should not proceed to compression merely because it is internally coherent.
It should proceed when the available evidence is sufficient for its intended use.
Once the model contains the relevant structure and has been compared with reality, it can be reduced for use.
Compression begins by establishing what the model can and cannot support.
The working boundary describes what the model initially attempts to include.
The limits describe what the model can actually support after verification.
This distinction matters.
A boundary is a construction decision.
A limit is a conclusion about valid use.
The model’s limits may arise from:
Knowing the limits of a model is not additional knowledge about the model.
It is part of knowing the model.
A model is not sufficiently usable until its user can recognize when not to use it.
This does not require complete theoretical understanding.
It requires enough understanding of the model’s purpose, assumptions, structure, and limits to prevent blind application.
A model does not become weak by stating what it cannot explain.
It becomes more reliable in use because its valid range is visible.
Defining the limits begins compression because it determines which claims, distinctions, and conditions must survive simplification.
Once the valid range of the model is visible, unnecessary complexity can be reduced.
Elements that behave similarly can be combined.
Details that do not affect the model’s use can be removed.
Long explanations can be replaced with clearer:
But compression has a limit.
If two meaningfully different situations become indistinguishable, the model has been compressed too far.
Compression should preserve not only the expected output of the model, but also enough of its underlying structure to understand:
Compress for use. Preserve enough structure for diagnosis and revision.
A model should be simple enough to work with and structured enough to be corrected.
A model should be used for the function it was designed to perform.
An orientation model may identify where to look next without explaining every mechanism.
An explanatory model may clarify why something happens without prescribing an action.
An application model may translate understanding into a repeatable process.
A model does not need to eliminate uncertainty.
It needs to make the relevant structure and remaining uncertainty more manageable.
Using a model outside its purpose may still produce an answer, but not necessarily one the model is capable of supporting.
An explanatory model does not automatically produce a good intervention.
A model that works at one scale may fail at another.
A model built for stable conditions may become misleading during rapid change.
Legitimate use should therefore state:
Application remains conditional on purpose, assumptions, evidence, and limits.
Application exposes the model to reality.
The resulting observations provide new evidence about both the model and its use.
Use the model according to its defined purpose and valid range.
The model may help:
Its value should be assessed relative to the implicit model it replaces.
The relevant question is not whether the explicit model is perfect.
It is whether it improves understanding or action enough to justify its cost.
Application also creates new observations.
These observations should not be treated only as outcomes.
They are feedback about:
A useful model should indicate when it needs to change.
Revision should not depend only on a vague feeling that the model no longer works.
Wherever possible, observations or changes that require reconsideration should be defined in advance.
A difference between expectation and observation does not show by itself what failed.
Several failures are possible:
Before changing the model, ask:
Did the model fail, did one of its assumptions fail, or was it applied outside its limits?
A correction addresses a local error without necessarily changing the model’s underlying structure.
Examples include:
A revision changes the model itself.
Examples include:
Revision should be directed at the source of failure.
Otherwise, a useful model may be abandoned because of poor application, or repeatedly patched when its underlying structure is wrong.
After structural revision, the changed model should be verified again before it is recompressed and returned to use.
The full cycle is therefore:
Explore & Build → Verify → Compress → Use → Observe → Diagnose → Correct or Revise
And after revision:
Revision → Verification → Compression → Use
A model is not a final statement about reality.
It is a maintained structure for working with reality.
The preceding sections explain the structure of model building.
The following questions translate that structure into a practical working sequence.
Not every question must be answered in equal depth.
Their relevance depends on the model’s purpose, consequences, uncertainty, and current state of development.
For a low-consequence use, a few short answers may be sufficient.
A model with long-term, difficult-to-reverse, or system-wide consequences may require considerably more examination.
The following use case applies the model to a common question.
It is not intended to provide a universal answer.
It demonstrates how an initially unstructured question can be transformed into a bounded and revisable model.
The question:
What should I do after I finish school?
is understandable, but not yet structured enough to answer well.
It combines several different questions:
A person may answer intuitively.
Because the consequences can extend across several years, however, making the underlying model explicit is likely to be useful.
The objective is not to calculate the correct life.
It is to construct a model that improves orientation and supports a sufficiently informed next step.
The original question is too broad because “what should I do?” contains no explicit standard for what counts as a good answer.
A first attempt might be:
Which path will lead to the best career?
But “best” may refer to:
These aims can conflict.
The question also assumes that a complete career path must already be selected. That may demand more certainty than the current situation allows.
A more useful purpose is:
Identify a next step that fits my current interests and constraints, develops valuable capabilities, reduces important uncertainty, and preserves sufficiently attractive future options.
This framing does not assume that maximum income is the only objective.
It does not assume that every option must be preserved.
It recognizes both the value and cost of exploration.
It limits itself to the next step without treating long-term effects as irrelevant.
The model should:
It does not need to identify the one ideal occupation for the rest of the person’s life.
Relevant elements include:
The elements do not merely coexist.
They affect one another.
Practical exposure can improve knowledge of personal fit:
Practical exposure → better information about fit → lower risk of long-term mismatch
Exploration also has costs:
Longer exploration → delayed qualification → delayed specialized development and income
Specialization can accelerate progress:
Earlier specialization → faster field-specific development → earlier access to specialized opportunities
At the same time, specialization can increase dependence on the initial choice:
Earlier specialization → greater commitment → higher switching costs if the assumed fit is wrong
A broad or reversible step preserves alternatives:
Higher reversibility → lower cost of error → more preserved future options
Preserving options is not free:
More preserved options → less immediate commitment → potentially slower specialized development
Information changes the value of commitment:
Better information about fit → higher expected value of specialization
This creates a sequence rather than a static comparison:
Exploration → information gain → improved selection → informed specialization
The usefulness of exploration depends on the initial level of uncertainty.
If fit is already well tested, additional exploration may have little value.
If fit is largely imagined, the same exploration may prevent a costly mismatch.
The central relationship is:
The value of information rises with the cost of being wrong and falls as relevant uncertainty is reduced.
The best next step therefore depends not only on its direct benefits.
It also depends on how it changes the quality of later choices.
For this use case, the model includes:
The model does not attempt to include:
These excluded factors may matter later.
They remain outside the current boundary because they cannot be represented with enough reliability or are not necessary for choosing the next step.
The boundary transforms the question.
Instead of:
What should I do with my life?
the model addresses:
What is a useful next step from my current position?
The model depends on several assumptions:
These assumptions will not hold equally for every person.
For someone with a strong and repeatedly tested interest in medicine, immediate commitment may be reasonable.
For someone whose preferences are based mainly on school subjects and imagined professions, information gain may deserve much greater weight.
| Uncertainty | Type | Appropriate response |
|---|---|---|
| Entry requirements | Informational | Research official requirements |
| Costs and financial support | Informational | Calculate and verify |
| Actual content of a program | Informational and experiential | Review curricula and speak with participants |
| Ordinary working conditions | Experiential | Observe, interview, or complete an internship |
| Personal fit with the work | Experiential | Test through practical exposure |
| Employment prospects in the next few years | Partly informational | Use current evidence and scenarios |
| Labor market in twenty years | Residual | Preserve adaptability rather than predict precisely |
| Future personal preferences | Residual and experiential | Maintain revision possibilities |
Missing entry requirements justify research.
Uncertainty about personal fit justifies experience.
Uncertainty about the distant future justifies flexibility, not indefinite analysis.
Useful evidence includes:
The distinction between imagined and observed fit is important.
A person may like the idea of psychology but dislike research methods, documentation, and emotionally demanding client contact.
Another may enjoy computer games but not software development.
Someone interested in medicine may discover that working conditions matter more than the subject itself.
The reliability of the evidence also matters.
Promotional material from a university, employer, or professional association may be informative but selective.
A frustrated individual may provide genuine experience without representing the field as a whole.
Important claims should therefore rely on multiple types of evidence where possible.
This model cannot determine a uniquely correct path.
It cannot reliably predict:
It also cannot derive personal priorities from facts alone.
The model can make conflicts visible, but it cannot objectively determine how much income, security, freedom, meaning, or status a person should value.
The result is therefore conditional:
Given the current information, priorities, constraints, and assumptions, this appears to be the most useful next step.
It is not:
This is the correct life path.
The model should not be used as proof that later changes are failures.
Changing direction may instead show that the original step successfully produced new information.
The full structure can be compressed into five criteria:
| Criterion | Question |
|---|---|
| Fit | How well does the option match current interests, abilities, and preferred ways of working? |
| Feasibility | Can the option realistically be entered and sustained? |
| Development | Which capabilities, qualifications, and opportunities does it create? |
| Reversibility | What happens if the option turns out to be wrong? |
| Information gain | How much will it reveal about personal fit and the remaining options? |
Possible paths can then be compared:
| Option | Fit | Feasibility | Development | Reversibility | Information gain |
|---|---|---|---|---|---|
| University degree | Depends on tested subject interest | Depends on admission and funding | Often high, but delayed | Medium | Medium |
| Vocational training | Often easier to test concretely | Usually financially accessible | High within a defined field | Medium | High |
| Immediate employment | Depends strongly on the available role | Often high | Variable | Often high | High |
| Structured exploration | Useful when fit is uncertain | Depends on funding | Low formal development | High | Very high |
| Travel or voluntary service | Indirect occupational fit | Depends on cost and access | Broad personal development | High | Medium |
The table does not calculate the answer.
It preserves the distinctions needed for comparison.
Scores may help reveal preferences, but numerical precision should not be confused with objective certainty.
Assume a student is considering three options:
The student has strong academic ability but little practical experience.
Several subjects appear interesting, but none has been tested outside school.
Financially, all three options are possible.
A superficial comparison may favor the degree because it appears to create the highest long-term qualification.
The model identifies a different structure:
The relevant comparison is:
Value of additional information versus cost of delayed specialization
If the expected information gain is high and the delay cost is limited, exploration may be the more useful next step.
The resulting action could be:
Complete a structured exploration period with defined internships, reflection points, and application deadlines that preserve access to later education or training.
This is not an unstructured gap year.
It is a test designed to reduce a specific uncertainty.
If the student had already completed several internships, worked in the field, and developed a stable interest over several years, the information gain from further exploration would be lower.
The same model could then support immediate specialization.
The model does not provide a universal answer.
It shows which differences should change the answer.
The conclusion should be reconsidered if:
If an internship feels wrong, the model should not immediately conclude that the entire field is unsuitable.
First diagnose:
The model is then corrected or revised at the appropriate level.
The model did not answer the original question directly.
It first transformed:
What should I do after I finish school?
into:
Which next step fits my current position, develops valuable capabilities, reduces the most important uncertainty, and preserves sufficiently attractive future options?
It then identified a causal structure:
Exploration can reduce uncertainty about fit, but delays specialization.
The result depends on which side currently dominates:
This transformation is part of modeling.
Many questions are difficult not because no answer exists, but because the original question combines:
A useful model separates these elements, identifies their relationships, and reduces the question to a form that can be examined.
The answer remains uncertain.
But the uncertainty is no longer unstructured.
Everyone already forms models.
The practical difference is whether those models remain implicit or become available for examination.
Explicit modeling is not necessary for every situation.
It becomes rational when the value of improved understanding exceeds the cost of constructing and maintaining the model.
The objective is neither maximal complexity nor maximal simplicity.
It is an adequately structured representation of reality that serves a defined purpose, exposes its assumptions, reveals its limits, distinguishes reducible from irreducible uncertainty, and remains open to correction.
Explore and build what matters. Verify it against reality. Define its limits and compress it for use. Then use, revise, and correct.
Initial stable formulation.
Includes:
This document is an exploratory note.
It proposes a general framework for constructing and maintaining useful explicit models.
It is not a complete theory of scientific modeling, rationality, forecasting, or decision-making.
Its purpose is structural clarification:
to make models sufficiently explicit to inspect, verify, use, diagnose, and revise.
Wende, A. (2026).
How to Model a Model: Building Useful Representations of Reality.
Exploratory Notes, systemic-effect.org. Version 0.1.
https://systemic-effect.org/exploratory-notes/how-to-model-a-model/v0.1
Version overview · Exploratory notes
systemic-effect.org
Exploratory Notes