RUNNING LEAN · THE METHOD STRUCTURE OF THE 3RD EDITION
From canvas to evidence — think the next step through.
A canvas expresses assumptions. A set of tools puts those assumptions up against reality. This atlas explains what each tool is worth, how they depend on one another, and why “finishing the process” does not imply “the business works.”
00 / REUSE
How to reread this, and how to apply it to a real problem
Method structure atlas · 2026-10-02. Text, styles, interactions and figure downloads are all embedded in the page; figures can be enlarged, and each can be downloaded as an original SVG.
First pass: see the whole
Read “How many tools” and the “Overview map” first. Separate business assumptions, customer facts, system metrics and stage goals, then read the four core representations.
Second pass: check the reasoning
Read “Necessary vs. sufficient” and interrogate every arrow: is this an input, a prerequisite, supporting evidence, or feedback? Use the scenario example to check whether any evidence was skipped.
Third pass: reuse it on a case
Write down the current goal, the known facts and the largest uncertainty, then pick only the tools that help with the judgment at hand. Do not treat twelve teaching units as twelve mandatory forms.
Six steps of one real application
Pin down the situation: a bare idea, paying customers already, or scaling.
Write existing evidence and guesses down separately, and locate the current stage.
Pick the main assumption that limits progress, and read the inputs and limits of the matching tool.
Design a validation action you can afford; before running it, write down the expectation, the measurement and the decision rule.
After running it, record the result, alternative explanations and shortcomings, then feed them back into the business model and risks.
State the next step, and what new evidence would change the judgment.
When moving across industries, redefine customer behavior, time windows, delivery constraints and the cost of trial and error. Repeat orders in food service and repeat purchases of equipment, for example, cannot share the same retention window. Moving to another industry changes how you validate; it does not let you skip evidence or lower necessary quality and safety requirements.
How work is saved and resumed in Lean Forge
The four core files are still input.md, canvas.md, risk-ledger.md and verdict.md. Interviews and experiments use the templates in cases/_template/_optional/ as needed; detailed evidence goes into the case’s evidence/, and older tool worksheets are located via engine/artifacts.md. Report summaries are written after manual review; the next session resumes from the latest evidence and verdict, never by guessing progress from the highest gate number.
This handbook explains the method and guides its use. Scoring for any specific gate follows the current project contract; a handbook version does not automatically override new evidence, nor does it imply that the book’s whole theory has been fully covered.
01 / CLASSIFY
First, tell tools, models and gates apart
Counting only canvases the text uses directly for this method, two kinds can be identified: the Lean Canvas and the Customer Forces Canvas; there is also a three-box Lean Canvas Lite, a variant of the former. The Business Model Canvas, mentioned as a theoretical source, is not counted as one of this book’s main working tools. Blueprints, roadmaps, estimation methods, interviews and cycles all play a role, but they should not all be called “canvases.”
Checked against the book: names, purposes, chaptersInterpretive synthesis: this page’s grouping and relationship mapsEngine convention: eight models, eleven gates
Green · tools / canvases
Concrete vehicles for expressing, interviewing, estimating or planning. The 01–12 on this page number teaching units — not the order of gates, and not the number of model files.
Blue · models
Analytical frameworks that interpret facts and generate hypotheses. Lean Forge currently has 8 separate model files; a tool may map to a model, or may have no model of the same name.
Orange · gates
Process checkpoints that decide, on evidence, whether to pass, loop back, continue or pause. Lean Forge currently has 11 gates; gates call on models and tools, but finishing a diagram or form does not mean passing a gate.
Order of use: first see what the case needs to decide and locate the current gate; read the models that gate requires; then choose tools that can obtain or express the evidence. Exact entry conditions and exit paths follow the gate files.
2 main canvases + 1 variant
The Lean Canvas describes the business model; the Forces Canvas records the drivers and resistance behind a customer’s choice and use; the Lite version focuses on customer, problem and value proposition. See chapters 1, 2 and 8 of the book.
4 core representations
This page groups the two canvases with the Customer Factory blueprint and the Traction Roadmap, expressing “business story, customer behavior, system flow, stage goals” respectively. This is a teaching grouping.
12 teaching units
This page adds 8 supporting tools / methods to form a complete path to understanding. It is not the author’s published “total number of tools in the book,” nor a count of illustrations.
A correction to the earlier “seven models” view
Seven models was how Lean Forge filed things at first review. Customer forces analysis was later added, bringing it to eight models, together with routing for interviews, experiment records and older tool outputs. The business pitch and customer progress are still carried as needed; the number of model files is not the book’s total number of tools.
02 / CONNECT
Together they form a learning loop
Modeling helps you ask better questions; experiments help you revise the model. Below is an interpretive synthesis across chapters showing the main flows of information. A real business enters according to its current stage and evidence — no project is required to walk mechanically from left to right again.
Interpretive relationship map · synthesizing the modeling, evidence and review structure across chapters
The overview map is Robert’s synthesis of relationships — not a ready-made flowchart from the book.Download overview map SVG ↓
A triple check runs through the design: desirability — do customers want it; viability — is the economic scale worth it; feasibility — can it be built at the current stage. Passing all three raises confidence in the judgment, but still cannot logically guarantee that the venture succeeds. See chapters 2–4 of the book.
03 / EXPLORE
Each tool does one specific job
Every card gives the inputs, outputs, next link and the conclusions it cannot support. Each figure can be downloaded on its own; expand a card to see the same fictional meal-ordering scenario and how Lean Forge takes up the tool. On narrow screens, figures scroll horizontally.
Puts customer, problem, value, channels and economics in one field of view so that contradictory assumptions come to light.
Functional structure redrawn · see book ch. 1, 11 · Enlarge ↗
Input
The venture idea, known facts, and assumptions not yet validated.
Output
A nine-box version of a single business story; split into several candidate variants when needed.
Feeds into
Supplies assumptions for stress testing; receives revisions from interviews, commitments and operating results.
Cannot be used to claim
Nine filled boxes only show the story is fairly fully expressed; they prove neither that the customer exists nor that the business works.
See a teaching example and the engine mapping
Fictional meal-ordering case: assuming “office workers need meal ordering,” you must go on to distinguish who, in which situation, and how they solve it today.
Lean Forge mapping: has its own model; maps mainly to gates 1.1 and 1.2, and runs through later back-filling.
Records, for one real switch, the push, pull, inertia and friction, and what happened afterwards; the blanks on the canvas are information still to be gathered in interviews and verified — not a nine-box grid to be filled out of thin air.
Functional structure redrawn · book ch. 8, fig. 8-4; behavior model in ch. 2, 13 · Download draw.io ↓ · Enlarge ↗
Input
From specific interviews: the switching event, the old solution, the choice process, anxieties, costs and outcomes of use.
Output
One forces-and-next-step record per customer, then a comparison of shared trigger situations, desired outcomes and alternatives.
Feeds into
Revises customer segments and the problem; gives leads for improving the offer, pitch, activation and retention.
Cannot be used to claim
The four forces are a framework for explaining behavior; a completed canvas cannot prove causation, nor can drivers and resistance be added up as precise numbers.
See a teaching example and the engine mapping
Fictional meal-ordering case: a manager starts looking for a new supplier because of late deliveries, but worries that switching will disrupt the invoicing process; the event, the choice and the worry are recorded separately.
Lean Forge mapping: the figure redraws the main record areas of the book’s fig. 8-4; the current interview template is used for record-keeping and is not the book’s canvas itself.
How are customers acquired, how do they get value, and how do they keep contributing value?
Looks at the business along the whole customer chain, separating “lots of traffic” from “a steady output of satisfied customers.”
Functional structure redrawn · see book ch. 3, 6, 12, 13 · Enlarge ↗
Input
Customer behavior, event definitions per stage, headcounts, conversion rates, time windows and customer cohorts.
Output
Metrics and loops for the five stages, exposing the stage that may limit overall output.
Feeds into
Feeds Fermi estimates and constraint diagnosis; takes in measured results from the MVP and growth loops.
Cannot be used to claim
The five stages are an analytical structure; not every industry has to charge in the same time order. Cumulative headcounts are not enough to explain causes either.
See a teaching example and the engine mapping
Fictional meal-ordering case: meals are paid for before they are experienced, so revenue comes before repeat orders; first value and later retention still have to be measured separately.
Lean Forge mapping: has its own model; chapter 3 already uses it for modeling, so it should not wait until gate 3.2 to be used.
To be worth doing, what progress must be reached by when?
Unfolds the minimum success criteria into stage milestones, linking present actions to future goals.
Functional structure redrawn · see book ch. 4, 11 · Enlarge ↗
Input
Founder goals, Fermi estimates, growth assumptions and evidence for the current stage.
Output
A now–next–later stage plan with traction goals to compare against.
Feeds into
Guides current risk choices and 90-day goals; at review, overlay measured results and revise.
Cannot be used to claim
A drawn goal curve is not a guaranteed forecast; an assumed growth rate must not be treated as fact. Nor is it a feature schedule.
See a teaching example and the engine mapping
Fictional meal-ordering case: first prove there are paying commitments, then that the service can be delivered repeatedly, and finally validate sustainable expansion.
Lean Forge mapping: maps to the “Traction Roadmap” model, but it is not the same tool as the growth loops of chapter 14.
Works back from a few key assumptions to the customer and channel requirements, catching models that are mathematically unlikely to hold as early as possible.
Functional structure redrawn · see book ch. 3, 4 · Enlarge ↗
Input
Minimum success criteria, revenue per customer, lifetime or churn rate, conversion rates and the reachable customer pool.
Output
The customer base, replacement acquisition and lead throughput needed to reach the goal.
Feeds into
Shapes the Traction Roadmap, and sends the most sensitive, least grounded numbers off for validation.
Cannot be used to claim
Numbers that add up do not mean it can be done. The formula holds only if units, time windows, and the definitions of revenue and churn are consistent.
See a teaching example and the engine mapping
Fictional meal-ordering case: assuming an annual revenue goal of ¥1.2 million and ¥1,200 of annual revenue per customer, about 1,000 active customers are needed; this is only a scenario estimate.
Lean Forge mapping: housed mainly in the “Viability math” model; chapter 3 of the book uses it for the viability test.
Pulls vague opinions back to specific times, events and behavior, supplying material for forces analysis.
Functional structure redrawn · see book ch. 8 · Enlarge ↗
Input
People who recently went through the target job, adopted an alternative or made a switch, plus event records that can be checked.
Output
Stories rebuilt in time order and interview notes; patterns that emerge across interviews.
Feeds into
Feeds the Customer Forces Canvas, which in turn revises business assumptions and the offer.
Cannot be used to claim
Convenience samples, leading questions and selective excerpts manufacture false convergence; an interview is not an order.
See a teaching example and the engine mapping
Fictional meal-ordering case: asking “which day last week did you switch lunch suppliers, and why” yields more facts than asking “do you want on-time delivery?”
Lean Forge mapping: the timeline and forces records are now merged into an on-demand interview template, so one interview is not written up twice.
What should improve right now to change the overall result?
Directs resources to the assumption that truly limits progress at the current stage, instead of everyone doing whatever is easy.
Functional structure redrawn · see book ch. 4, 6 · Enlarge ↗
Input
Stage goals, business model, Customer Factory readings and known evidence.
Output
The current constraint, its possible causes and the assumptions to test first.
Feeds into
Orders cycle activities and minimum experiments; reorders when new evidence appears.
Cannot be used to claim
The lowest conversion rate is not necessarily the most important constraint; align goals and measurement first, then test causes.
See a teaching example and the engine mapping
Fictional meal-ordering case: there are plenty of leads, but fulfillment errors stop customers from coming back; buying more traffic now may only magnify the waste.
Lean Forge mapping: one of the original seven models, now part of eight; the 2×2 rough screen, the constraint column and the exact format of gate 1.4 belong to the engine’s implementation.
How do you turn judgments into actions that produce evidence on schedule?
Organizes models, activity proposals and five short sprints around a shared cycle goal. Fig. 11-1 shows exploration, design and offers in the problem–solution stage; fig. 13-3 shows the product–market stage optimizing the aha moment, the switching moment and the customer happiness loop.
Structure traced from the book · ch. 11, fig. 11-1 · Enlarge ↗
The second figure, fig. 13-3, covers the customer happiness loop once a product exists; its sprint goals differ from the figure above.
Traction goals, the updated canvas and roadmap, activity proposals and resource limits.
Output
Expected vs. measured results for each sprint, stage-appropriate check evidence, a progress report and a persevere / pivot / pause decision.
Feeds into
Updates the canvas, roadmap and next cycle’s goals; measured results in turn correct the models.
Cannot be used to claim
The two figures are sample plans for different stages, not five fixed gates for every case; reaching the deadline or finishing a sprint does not mean validation passed.
See a teaching example and the engine mapping
Fictional meal-ordering case: a meal-ordering business with a product first checks whether the first delivery gives customers value, then checks continued use and switching; specific experiments and decision rules must be set from the case’s own evidence.
Lean Forge mapping: the two figures redraw the functional structure of the book’s fig. 11-1 and fig. 13-3; the record fields and the gate 2.4 verdict are Lean Forge implementation conventions.
Tool · teaching unit 09Model · no dedicated model; used as neededGates involved · no dedicated gate; used as needed
How do you get stakeholders to understand — and question — the business?
Connects the business story, orders of magnitude and stage plan, exposing gaps and making clear what support is needed.
Functional structure redrawn · see book ch. 5, 11 · Enlarge ↗
Input
The Lean Canvas, Customer Factory / Fermi estimate, Traction Roadmap and current progress.
Output
A business narrative open to discussion, feedback, and a clear resource ask.
Feeds into
Valuable feedback is fed back into assumptions; audience approval is not logged as market validation.
Cannot be used to claim
Advisor approval, successful fundraising and customers willing to buy are different kinds of evidence.
See a teaching example and the engine mapping
Fictional meal-ordering case: when an investor asks “how many customers do you need to hit the goal,” you should be able to go back to the estimate rather than just saying the market is huge.
Lean Forge mapping: the book has a dedicated chapter; the current eleven gates have no separate pitch gate.
Will customers pay a price to commit to a specific value?
Before delivering at scale, tests whether the problem understanding, solution, price and call to action can produce a real choice.
Functional structure redrawn · see book ch. 7, 9, 10 · Enlarge ↗
Input
Customer stories, forces leads, the minimum solution, pricing and delivery limits.
Output
A demo and explicit offer, commitments with a real cost, reasons for rejection and conversion records.
Feeds into
Feeds back into the solution or customer judgment; once commitments meet the bar, deliver in batches and keep testing whether value is realized.
Cannot be used to claim
Commitment can still end in failed delivery, refunds or no repeat purchase; high conversion also requires checking the sample and acquisition cost.
See a teaching example and the engine mapping
Fictional meal-ordering case: showing a one-week trial plan and taking trial orders gives stronger behavioral evidence than asking “would you buy later?”
Lean Forge mapping: maps to the demo–sell–build model and gates 2.2 and 2.3; a PASS at 2.3 should not be read as all business assumptions being confirmed.
How do existing customers power the acquisition of new ones?
Identifies revenue, content / data or referrals that can be reinvested, freeing growth from a constant drain on the founder’s time.
Functional structure redrawn · see book ch. 14 · Enlarge ↗
Input
A running Customer Factory, retention and economics data, and reinvestable assets.
Output
Candidate loops and their test results, narrowing step by step to the growth methods backed by evidence.
Feeds into
Brings new customers into the factory; changes in scale will also expose new constraints.
Cannot be used to claim
Repeat use does not automatically bring newcomers; one burst of subsidized growth does not prove long-term sustainability.
See a teaching example and the engine mapping
Fictional meal-ordering case: renewals from existing customers produce gross margin that is reinvested in effective acquisition; the payback period must be checked, not just sales.
Lean Forge mapping: housed mainly in the Traction Roadmap model and gate 3.3; understanding it on its own keeps a stage diagram from being mistaken for a growth mechanism.
04 / REASON
Not every arrow can be read as “therefore it holds”
The logical analysis here is Robert’s interpretive layer. A strict necessary / sufficient relationship requires explicit propositions and conditions of application; between tools, the more common relations are input, methodological prerequisite, evidential support and feedback.
Robert’s logical analysis · separating procedural eligibility, evidential support and real-world outcomes · Download SVG ↓
Canvas completed → business succeeds? It does not follow.
Counterexample: all nine boxes can be filled in and still nobody pays, so it is not sufficient. A business can also explore by other means and succeed, so “using this particular canvas” is not a necessary condition for success either.
Prerequisite → universal law? That does not follow either.
“Validate commitment before investing in building” is a strategy for controlling risk. A Lean Forge PASS is procedural eligibility to enter the next gate; it cannot be upgraded into the proposition that “every successful business must pass through this gate.”
Between tools
Kind of relation
What is passed on
Must not be swapped for
Lean Canvas → Fermi estimate
Input dependency
Customer, pricing and channel assumptions enter an order-of-magnitude check.
Nine complete boxes ≠ the economics or scale work.
Customer Factory → constraint diagnosis
Diagnostic input
Uses behavior and flow to see what limits the goal.
One weak metric ≠ the root cause has been found.
Fermi estimate → Traction Roadmap
Modeling handoff
Breaks scale requirements down into time and stage goals.
A goal curve ≠ it can already be done.
Interviews → Forces Canvas → offer
Evidence & interpretation
Distills situations, drivers and obstacles from real experiences.
Hearing a pain point ≠ willingness to pay for this solution.
Offer → purchase commitment
Experimental test
Observes whether customers pay a real cost.
Sign-ups, intent and deposits differ in strength.
Commitment → MVP → retention
Process prerequisite, then further validation
Following the book’s strategy: validate commitment first, then deliver and test.
Prepayment ≠ successful delivery ≠ product–market fit.
Retention → growth loop
Enabling condition
Existing customers may supply the assets that new customers require.
Good retention is not sufficient; the assets must also bring newcomers.
Measured results → canvas / roadmap
Feedback correction
Changes assumptions, forecasts, stage judgments and the order of experiments.
Revising a model does not automatically prove the revision correct.
If the numbers add up, why is it still only an assumption?
A fictional subscription business with an annualized revenue goal of ¥1.2 million. Drag the annualized revenue per active customer and the monthly churn rate to see the steady-state customer count and the churn you must replace. All numbers are teaching assumptions.
Active customers needed at the target steady state
1,000 customers
About 50 new customers a month are needed just to replace churn.
Calculation: steady-state active customers ≈ annualized revenue goal ÷ annualized revenue per customer; monthly churn replacement ≈ steady-state active customers × monthly churn rate. Annualized revenue is current monthly revenue × 12; customer heterogeneity, seasonality, growth and costs are not included. These equations hold within their assumptions; real price acceptance, the customer pool and channel capacity still need validating.
05 / APPLY
Same business, different evidence — a different next step
Fictional scenario: a weekday lunch subscription for small teams. The suggestions below are a teaching exercise, not a verdict on a real case; without new evidence, pressing a button on this page does not “pass a gate.”
Each round leaves behind: what the evidence is → which method or rule was used → which part is inference → what the next experiment is → what result would change the judgment.
06 / TRACE
What this atlas rests on — and what it does not prove
Basis in the book: a personally owned copy of Ash Maurya’s Running Lean (3rd edition). Definitions and connections in chapters 1–4, 6, 8, 10, 13 and 14 were checked in a targeted way around the relevant tools; item 02 is checked against the canvas fields of fig. 8-4 in chapter 8, and item 08 against the problem–solution cycle of fig. 11-1 in chapter 11 and the customer happiness loop of fig. 13-3 in chapter 13. Figures are independently redrawn by structure, and none of the book’s images are embedded on this page; this is neither a certified word-for-word reading nor a full list of illustrations.
Edition scope:O’Reilly · Running Lean, 3rd Edition provides the edition and chapter structure. The author’s later articles may evolve the ideas; this page is based mainly on the 3rd edition and does not automatically write later changes back into the book.
Project implementation: the engine uses eight models, eleven gates and a set of parsing code to express how it encodes the theory. The eight models and eleven gates belong to the project’s structure; automatically generated reports do not decide whether a market is real.
Interpretation and derivation: the four-core / twelve-unit grouping, the overview map, the necessary-vs-sufficient analysis, the interactive calculation and the meal-ordering scenarios were all organized by Robert for this teaching purpose. All SVGs are original drawings; the two figures in item 08 distill the book’s examples for different stages, and are neither pixel copies of the originals nor a single fixed sprint schedule.
