Founder OS · Model integrity
Every model is part record and part hope. Most do not say which is which.
A spreadsheet presents a number taken from a signed contract and a number someone guessed in identical type. This layer makes the difference visible, and then shows what the model looks like once the evidence is weighted.
Assumption Integrity Layer
- Asks
- How much of this model is known, and how much is hoped?
- Decides
- Every input carries its source, weighted by strength, from direct evidence down to plain assumption. The model is then shown twice: as the founder built it, and adjusted for the evidence. Gaps above 20% are flagged. The runway is shown both ways, and the difference is the assumption risk.
- Result
- Ten metrics rated green, amber or red, and a recommendation on whether the model can be shown as it stands, only with stated conditions, or not yet.
The failure it is built to catch.
Financial models fail quietly. Not because the arithmetic is wrong — the arithmetic is almost always right — but because an assumption entered eighteen months ago has been compounding ever since, and nobody can now remember it was an assumption.
By the time a model is in front of a board or an investor, its weakest input looks exactly like its strongest. The founder cannot say which numbers would survive a question, because the model does not record where any of them came from.
This is also the firm’s second rule applied to a spreadsheet: weak evidence counts for less, so a plan built on hope shows its gap rather than hiding it.
Every input carries its source.
Strength of evidence is recorded alongside the number itself, and the model is then recalculated with each input weighted accordingly.
- Direct evidence
- The strong end of the scale. A signed contract, an invoice paid, a conversion rate measured in your own data. It is what it says it is.
- Plain assumption
- The weak end. A figure someone chose. It may well be right, and it carries the least weight, and the model says so rather than letting it pass as fact.
- Everything in between
- Most inputs sit between those two ends, and are weighted by how much supports them rather than by how confidently they were entered.
- Gaps above 20%
- Where the evidence-adjusted figure diverges from the figure as built by more than a fifth, the input is flagged rather than quietly absorbed.
- Runway, shown twice
- As the founder built it, and adjusted for evidence. The difference between the two is the assumption risk, stated as a number of months rather than a feeling.
Ten metrics, and one recommendation.
- Ten metrics rated
- Each is green, amber or red on the strength of what sits behind it — not on whether the number is impressive.
- Show as it stands
- The model can go in front of a board or a data room as built. Its inputs would survive being asked about.
- Show with stated conditions
- Usable, provided the weak inputs are named when it is presented. Naming them yourself is considerably better than being asked.
- Not yet
- Too much of the model rests on inputs that cannot be evidenced. The work is to evidence them, not to redraw the chart.
Common questions.
Does this tell us whether our numbers are good?
No, and it is deliberate. It measures whether a number can be evidenced, not whether it is impressive. A modest figure from a signed contract is rated above an excellent one that someone chose.
What if most of our model is assumption?
At pre-seed and seed that is normal and expected. The problem is never that assumptions exist; it is that they are indistinguishable from facts by the time anyone reads the model. Labelled assumptions are defensible.
Will this make our model look worse?
The evidence-adjusted view is usually more conservative, yes. It is also the version that survives diligence. The alternative is that someone else finds the gap and you learn about it from them.
Is this check available on its own?
It runs inside Founder OS, which programmes use during a Pilot Cohort with us. The weighting scheme above can be applied by hand to any model — a source column and an honest rating gets most of the value.
Founder OS runs this during a Pilot Cohort.
Programmes use it with their cohort, and Dineshwara reviews every cohort's results. If you want a read on your own company first, the founder readiness check is twelve questions and open to anyone.