The canvas that runs itself.

01 · what the canvas was for

The canvas family, Osterwalder’s business model canvas and Ash Maurya’s leaner adaptation, was built to solve a coordination problem: business plans were forty pages of fiction, and teams needed one page of testable claims instead.^1,2 Behind it sat the lean startup argument from Eric Ries and Steve Blank, which is really one sentence: a startup is not a small company executing a plan, it is a search for a business model that works, run as a loop of build, measure, and learn.^3,4

The approach has real evidence behind it, not just conference talks. In a randomized trial with 116 Italian startups, founders taught to treat their assumptions as hypotheses and test them formally earned more revenue, pivoted more decisively, and killed bad ideas earlier than the control group.⁵ Thinking like a scientist paid. The problem was never the method. The problem was that each turn of the loop cost weeks, and founders have quarters, so the loop turned three or four times when it needed thirty.

02 · every box becomes a job

Here is the inversion this note exists for: with Fable 5, the canvas stops being a form you fill and becomes a set of jobs you dispatch. Tap each box below to see what the model actually runs. The pattern to notice is that no box gets an opinion; every box gets work with an output that can be wrong, which is what makes it evidence rather than decoration.

03 · the loop, run weekly

The canvas earns version numbers the way software does, and the discipline comes from Maurya’s ordering rule: attack the riskiest assumption first, not the easiest one.² Each week runs the same cycle. The model ranks the canvas’s guesses by how much the business dies if they are wrong, and for the riskiest one it designs a falsifiable test with a pass threshold written down before the data arrives, which is the same pre-registration discipline I use in measurement everywhere. Then the work splits: the model runs the desk half, the evidence scans, the pricing analogs, the channel math, the interview guide, and a human runs the field half, which is the actual conversations with actual buyers. The model then synthesizes transcripts against the guesses, the canvas increments, and the changelog records what died and why.

The changelog matters more than the canvas, and this surprises people. A canvas is a claim about the present; the changelog is the accumulated evidence trail, and it is what makes version 1.4 trustworthy where version 1.0 was a guess wearing a grid. It is also, not incidentally, the exploration transcript from Field Note 04 in another costume: the dead guesses carry priced reasoning, and they make the next venture’s canvas start half-drawn.

canvas: route-density data product · spinout candidate
v1.0  problem guessed as "dispatchers lack visibility."
v1.1  KILLED by six interviews: dispatchers have visibility
      and ignore it. real problem is trust in the reroute
      suggestion. uvp rewritten.                     [field]
v1.2  segment narrowed: 30-80 truck fleets. below 30 the
      dispatcher IS the algorithm; above 80 they run Samsara.
      channel box redone around two trade associations. [desk]
v1.3  pricing test: anchored at $18/truck/mo against the
      $30-40 enterprise tools; 4 of 5 design partners accepted
      without negotiation, which means the anchor is too low.
      revenue box revised upward.            [field + note_01]
v1.4  unfair advantage passes the added-value test only with
      pooled cross-fleet data. exclusivity clause added to
      design-partner agreements.                   [note_03]

04 · what it returns

The return is loop speed, and loop speed compounds. When the desk half of an iteration costs a day instead of three weeks, the loop turns every week, the kills happen at version 1.1 instead of at launch, and the Camuffo result, that hypothesis-driven founders terminate bad ideas earlier, stops being a discipline you aspire to and becomes the default cadence of the work.⁵ In the Foundry, no company gets formed until its canvas has survived several versions of this loop with design partners supplying the field half, which is cheap insurance against the base rates in Field Note 01 of this series’ parent thesis: most spinouts should die as canvases, where dying costs a few thousand dollars, not as funded companies, where it costs years.

There is a second return that owners feel immediately: the canvas becomes a shared instrument between me and the client. An owner who can read a changelog can see exactly why the venture believes what it believes, which converts the usual consulting faith into something closer to an audit trail. Trust built on evidence survives bad quarters; trust built on charisma does not.

05 · the honest limits

Three, as always. The model’s desk evidence is desk evidence: Fable 5 can summarize a market’s consensus with beautiful fluency, and a market’s consensus about itself is frequently wrong, so desk findings set priors and only field contact settles claims. Validation theater is now cheaper too: the same tool that runs an honest canvas can generate a persuasive fake one, which is why every test gets its pass threshold written before the data arrives, not after. And the canvas is a search tool, not a running company: once the model is found, the discipline inverts from exploration to execution, per March in Field Note 04, and a team still iterating its canvas in year two is not being lean, it is being lost.