A model is a lens. Evidence decides whether the lens is useful.
09models audited
03interactive decision labs
26 OCT 2026research review
Before you begin
The blueprint is a starting point.
The supplied “9 Mental Models” poster is a useful memory aid. It compresses distinct traditions into short prompts and sometimes implies more certainty than the evidence supports. Here are the most important repairs.
Origins are not endorsements.
“Inspired by” does not establish invention, scientific validation or exclusive ownership of an idea. The guide distinguishes historical roots, practitioner judgment and empirical findings.
Models are not guarantees.
Game theory can reveal conflict. Leverage can amplify failure. Faster decisions can be worse decisions. Every model needs a boundary and a test.
Different futures need different stories.
Scenario planning goes beyond optimistic and pessimistic numbers. It explores different mechanisms and identifies actions that survive more than one future.
01 / The audited field guide
Nine lenses. Nine practical outputs.
Keep the original order or filter by purpose. Each model includes a correction, mechanism, application steps, failure modes and source notes. ImageFirm and Ruz examples are illustrative proposals, not claims of measured results.
Separate what must be true from what everyone assumes.
Blueprint check
Useful simplification. “What do I know for certain?” sets an unrealistic standard for most business questions. Distinguish observations, constraints and uncertain assumptions instead.
Go deeper +
What it means
Begin with the required outcome. Decompose the problem into quantities, mechanisms and constraints. Treat a supplier quote, an industry convention and a physical limit as different kinds of evidence. Rebuild candidate solutions, then test the weakest premise. The value comes from exposing assumptions, not from dismissing accumulated expertise.
Apply it
State the outcome without naming today’s solution.
Label inputs: measured fact, binding constraint, estimate or assumption.
Identify the assumption that most limits your options.
Design a cheap test that could disprove it.
Illustrative application
ImageFirm example: instead of assuming every creative request needs a full manual workflow, separate briefing, generation, review and delivery. Measure the quality floor for each step before deciding what to automate.
Failure modes
Reinventing a solved problem; mistaking confidence for evidence; excluding trust, customer behavior or compliance because they are not physical laws.
Useful output
An assumption ledger with evidence, uncertainty, owner and next test.
Attribution & evidence
Aristotle provides a historical foundation for reasoning from principles. That does not make the modern business exercise a method he tested or prescribed.
Conceptual foundation, not evidence that this branded technique reliably improves business outcomes.
Trace how people and systems respond to the first effect.
Blueprint check
Attribution needs nuance. Consequences of consequences are not identical to Howard Marks’s “second-level thinking,” which also examines expectations, consensus and what is already reflected in price.
Go deeper +
What it means
Build a short causal chain and test every arrow. Include time delays, stakeholder responses and feedback. An effect can reverse sign: faster delivery might increase demand, overfill the review queue and ultimately worsen delivery times. Stop expanding when additional branches no longer change the decision; long chains amplify uncertainty.
Apply it
Name the immediate benefit and who receives it.
Ask which behavior or incentive changes next.
Add delay, counter-response and plausible sign reversal.
Attach a measurable signal to the decisive causal link.
Illustrative application
ImageFirm example: free tools may increase acquisition, but also support load and compute usage. Measure activation, repeat use and cost per retained user before treating traffic as durable value.
Failure modes
Inventing elaborate stories without causal evidence; confusing contrarianism with insight; treating three rounds of “then what?” as a validated rule.
Useful output
A consequence map with sign, delay, evidence and an observation that would challenge each link.
Attribution & evidence
Marks is an influential practitioner of second-level investment reasoning, not an established sole originator of causal consequence analysis.
Practitioner reasoning supports the distinction; it is not a controlled test of a universal strategy formula.
Start from a failure and work back to preventable causes.
Blueprint check
“Guarantee failure” is too strong. Plausible failure paths are enough. Avoiding failure is necessary in many settings but does not, by itself, create customer value.
Go deeper +
What it means
A premortem assumes the project has failed, then elicits possible explanations. Use independent written input before discussion. Separate a vivid story from a likely failure mechanism. Prioritize severe, plausible and poorly detected risks; assign an owner and an early warning. Pair prevention with a positive theory of why the project should succeed.
Apply it
Define failure concretely at a specified date.
Collect failure explanations independently.
Group them by underlying mechanism, not wording.
Set prevention, warning signal, owner and fallback.
Illustrative application
Ruz example: assume a proposed opening is missed. Test dependencies such as long-lead equipment, contractor handoffs and commissioning readiness. Convert the largest gaps into dated checks.
Failure modes
Pessimism theater; treating every risk as equally important; allowing the most senior person’s story to dominate; canceling every uncertain opportunity.
Useful output
A short failure register with triggers and accountable mitigations.
Attribution & evidence
The poster associates inversion with Charlie Munger. Treat this as an influence credit, not proof of invention. A related named practice is Gary Klein’s project premortem.
A 1989 experiment studied how certainty and temporal framing affect explanations. It does not prove a fixed reduction in real project failure rates.
Turn new information into a bounded test and fresh feedback.
Blueprint check
Incomplete. Boyd’s model is richer than four boxes in a circle. Orientation shapes observation, decisions and action; implicit guidance and feedback matter. Speed without learning can compound mistakes.
Go deeper +
What it means
Observe credible signals. Orient by comparing them with context and competing explanations. Decide on a hypothesis and action. Act in a way that generates informative feedback. Use a short cadence for reversible decisions; add review when consequences are hard to undo. A failed test is useful if it changes the next decision.
Apply it
Observe: what changed, from which reliable signal?
Orient: which explanation fits, and what conflicts with it?
Decide: what result would justify changing course?
Act: run the bounded test and review its result.
Illustrative application
ImageFirm example: a conversion drop triggers investigation of traffic mix, page changes and measurement issues. Test one defensible hypothesis with a rollback condition instead of repeatedly redesigning the funnel.
Failure modes
Reacting to noise; skipping orientation; making multiple changes that obscure causality; rewarding decision speed rather than useful learning.
Useful output
A decision–test log with hypothesis, owner, result and revision.
Attribution & evidence
John Boyd developed OODA: observe, orient, decide and act. Business applications are adaptations of a military decision framework.
Primary doctrine describes the model. It does not establish a general measured advantage over every other business decision process.
Follow stocks, flows, incentives, feedback and delay.
Blueprint check
Correct direction, incomplete mechanism. “Everything affects everything” is not enough. Name the system boundary, accumulation, causal links and delays.
Go deeper +
What it means
A stock accumulates when inflow exceeds outflow. Reinforcing feedback amplifies a change; balancing feedback counters it. Delays can cause overcorrection. Map only enough of the system to explain the decision, then test relationships against data. Optimize the end-to-end outcome, not a department’s isolated output.
Apply it
Define the outcome, boundary and time horizon.
Identify the important stocks and their flows.
Mark reinforcing and balancing loops with delays.
Test an intervention and watch displaced costs.
Illustrative application
ImageFirm example: increasing content generation may build a review backlog. If rushed review creates rework, generation speed alone can reduce accepted output. Track accepted deliverables and queue age, not raw drafts.
Failure modes
A beautiful but untestable map; treating correlation as causation; overlooking external effects; assuming a leverage point always works in the expected direction.
Useful output
A minimal causal map plus stock/flow measures and a feedback review date.
Attribution & evidence
Donella Meadows is a major contributor and communicator within a much broader systems tradition; not the sole founder.
Meadows’s synthesis is influential conceptual guidance with explicit cautions; systems-engineering practice adds disciplined validation, not universal causal proof.
Evaluate your action together with others’ incentives and responses.
Blueprint check
Misleading implication: there is not always a win-win outcome. A Nash equilibrium can be inefficient. Strategic interaction may involve conflict, cooperation, incomplete information or several equilibria.
Go deeper +
What it means
Specify players, actions, information, timing and payoffs. Ask what each player would do if others acted as assumed. An equilibrium is a consistency condition, not a promise that real people will reach it. Repeated interaction, credible commitments and observable behavior can change incentives.
Apply it
Identify actual decision-makers and their alternatives.
Describe payoffs including effort, risk and reputation.
Compare best responses and credible commitments.
Test sensitivity to beliefs and bounded rationality.
Illustrative application
Ruz example: a contractor and owner may both prefer timely delivery but disagree about who pays for preventive work. Analyze how milestone evidence and responsibility allocation change effort incentives before negotiating terms.
Failure modes
Guessing others’ payoffs; assuming perfect rationality; treating a simplified matrix as a forecast; ignoring repeated relationships and hidden information.
Useful output
A stakeholder payoff map with alternative responses and uncertain assumptions.
Attribution & evidence
Von Neumann is foundational; Morgenstern, Nash and many others are also central. A one-name credit is not a complete history.
Mathematical results hold within specified games. They do not automatically predict behavior in a changing commercial relationship.
Improve the constraint, then scale a proven process.
Blueprint check
The poster blends three ideas: scalable resources, operational bottlenecks and systemic leverage points. Related, but not interchangeable. Leverage magnifies errors and downside too.
Go deeper +
What it means
Find what currently limits useful output: demand, production, review or delivery. Improving a non-constraint may only create inventory. Reusable software and content can reduce repeated work, but maintenance, distribution and quality costs remain. Validate unit economics and failure containment before multiplying activity.
Apply it
Measure accepted output across the full process.
Identify the binding capacity or demand constraint.
Relieve it with the smallest reversible intervention.
Re-measure: the constraint may have moved.
Illustrative application
ImageFirm example: if draft generation handles 100 items daily, review 20 and delivery 60, more generation cannot lift the deterministic capacity ceiling above 20. Review improvements can, until another step becomes limiting.
Failure modes
Automation of a flawed workflow; fragile dependence on one platform; scaling low-quality output; ignoring variability, rework and weak demand.
Useful output
A constraint test with baseline throughput, intervention cost and acceptance criteria.
Attribution & evidence
Naval Ravikant discusses labor, capital and scalable code/media. Meadows discusses system interventions. Neither is equivalent to a guarantee of large results from little effort.
Practitioner framing is useful, but leverage claims require local measurement. The lab below illustrates a simplified capacity identity, not a field benchmark.
Prepare for different worlds, not three versions of one forecast.
Blueprint check
Important correction: best/likely/worst is sensitivity analysis, not a complete scenario method. Useful scenarios differ in causal structure, not only how favorable the outcome is.
Go deeper +
What it means
Choose critical uncertainties that could alter the decision. Develop coherent, distinct worlds and test each strategic option in all of them. Identify actions that work broadly, staged commitments and observable triggers. Do not assign probabilities merely to make a worksheet look quantitative; plausible scenarios need not be exhaustive or mutually exclusive.
Apply it
Define the decision horizon and critical uncertainties.
Construct distinct worlds with coherent causal stories.
Stress-test options; identify vulnerabilities and robust actions.
Attach a signpost and response to each material shift.
Illustrative application
ImageFirm example: combine low/high customer demand with low/high inference cost. Compare full launch, staged pilot and postponement in all four worlds. No scenario is designated “most likely” without evidence.
Failure modes
Optimism dressed as a scenario; false precision; ignoring disruptive combinations; creating stories that never change a decision.
Useful output
A scenario–option matrix with signposts, response owners and commitment gates.
Attribution & evidence
Shell is a long-running institutional practitioner, not the sole inventor. Its own scenarios explicitly distinguish exploration from prediction.
Shell documents decades of use. Longevity demonstrates adoption, not controlled proof that scenario planning causes superior returns.
Start with a base rate, update with evidence, track calibration.
Blueprint check
Useful but incomplete. A numerical confidence is not automatically a good probability. Expected value also does not capture affordability, catastrophic downside or every decision-maker’s preferences.
Go deeper +
What it means
Define a resolvable event and deadline. Choose a relevant reference class, then update as evidence arrives. Distinguish evidence likely under success from evidence equally likely under failure. Record forecasts before outcomes are known. Across many predictions, calibration asks whether events assigned 70% occur roughly 70% of the time.
Apply it
Define event, deadline and resolution source.
Set an initial estimate using a relevant base rate.
Update using the diagnostic strength of new evidence.
Score resolved forecasts and review systematic errors.
Illustrative application
ImageFirm example: estimate whether a pilot will reach a defined repeat-use target by a fixed date. A favorable customer interview may be weak evidence if unsuccessful pilots often receive similar praise.
Failure modes
Ignoring the denominator; counting correlated signals twice; confusing a positive expected result with safety; scoring yourself on a single lucky outcome.
Useful output
A dated forecast register, update rationale and calibration review.
Attribution & evidence
Kahneman and Tversky studied judgment biases; they did not originate probability theory or Bayesian inference.
A two-year geopolitical forecasting tournament found benefits from training, teaming and tracking. That supports specific forecasting practices, not automatic transfer to every business decision.
These labs calculate simplified examples locally in your browser. All starting values are hypothetical. They demonstrate reasoning, not forecasts or business recommendations.
Lab 01 / Probability
How informative is the evidence?
H means “the pilot reaches its target.” E is a positive signal. A positive signal is weak evidence if it is also common when pilots fail.
Likelihoods must be justified. Reusing the same underlying evidence as independent signals overstates confidence.
Lab 02 / Leverage
Fix the limiting step.
Three sequential stages. Constant capacity, sufficient demand, no rework or variability. Long-run throughput cannot exceed the slowest stage.
Create
Review
Deliver
Real throughput can be lower. Demand, shared resources, variability and quality checks must also be modeled.
Lab 03 / Scenarios
Robustness without invented probabilities.
Hypothetical net outcome points, not money. Compare each option with the best outcome in each world. Regret is the gap to that world’s best option; minimax regret chooses the smallest worst gap.
Fixed illustrative scenario–option outcomes
Option
High demand / low cost
High demand / high cost
Low demand / low cost
Low demand / high cost
Worst regret
This rule is a choice, not an objective definition of “best.” It depends on the options, scenarios and scoring scale. It can miss unimagined futures and does not replace hard constraints.
03 / Combine deliberately
Use the model that addresses the uncertainty.
Decision problem
Start here
Then add
Leave with
“The standard approach is too costly.”
First principles
Leverage + systems thinking
Testable redesign and end-to-end metric
“Growth looks strong, but service is degrading.”
Systems thinking
Second-order thinking + OODA
Constraint hypothesis and bounded test
“This commitment is hard to reverse.”
Inversion
Scenarios + probabilistic thinking
Failure controls, stages and decision triggers
“Success depends on someone else’s response.”
Game theory
Second-order thinking + OODA
Incentive map and response-contingent action
“Our forecasts feel confident but keep missing.”
Probabilistic thinking
First principles + feedback review
Reference classes and scored forecasts
ImageFirm synthesis: this sequencing is a practical workshop design, not a separately validated methodology.
04 / Put one decision through the method
Your decision worksheet.
Capture the reasoning before you know the result. Entries stay in this page’s memory; nothing is transmitted or automatically saved. Download the text record before closing or refreshing.
05 / Evidence, without borrowed certainty
Not all support is the same.
Empirical research
Mellers and colleagues studied a two-year geopolitical forecasting tournament. Training, collaboration and performance tracking improved forecast accuracy. The outcome was forecasting performance in that setting, not company profitability.
Shell documents long-running scenario use; NASA documents systems-engineering processes. These sources show how practitioners organize uncertainty and validation. Adoption alone is not causal evidence of superior results.
Boyd’s doctrine, Meadows’s synthesis and Marks’s memos clarify mechanisms and questions. They should not be converted into universal success rates or scientific rankings of the nine models.
Source material: user-supplied “9 Mental Models for Strategic Thinking,” credited in the image to Natan Mohart. The original poster is not reproduced. Diagrams, exercises and business examples are original ImageFirm explanatory material. Source review: 26 October 2026. This is an educational decision guide; it does not claim that the nine models form a validated package.