There are no magic numbers

Opening thesis

We use a phrase internally, “Innovation Physics,” and it is only useful if it means one thing: evidence beats intuition. The moment it starts to sound like we have discovered the laws of startups — exact cutoffs that decide your fate — it becomes a lie with decimal points. So let me be clear about what the numbers are and are not.

Real constraints, false precision

Some startup truths really do behave like laws, because they are true by definition. You cannot validate an offer nobody acts on. You cannot sell what you cannot deliver. If it costs more to win a customer than they are ever worth, the math does not care how much they liked the demo. Desirability, feasibility, and viability genuinely constrain each other — strength in one does not excuse a hole in another.

The problem is not that reasoning. It is pretending every numeric cutoff already has the authority of a measured constant. A threshold can be a useful interim guess and still need calibration, context, and your judgment. False precision is especially seductive early, because a clean number feels better than an honest caveat — and a number with no real basis can send you confidently in the wrong direction.

What our numbers are, and are not

So here is the honest version. We use signals to suggest whether to advance, iterate, pivot, or stop, and the direction of those suggestions rests on solid ground — accounting identities, established behavioral research, and hard-won practitioner craft. What people do outweighs what they say. That part we stand behind.

The specific cutoffs are a different thing. Today they are interim priors, not proven laws, and they are meant to be read qualitatively — a nudge, not a verdict. Benchmarks from comparable contexts can add useful context, but only your own observed evidence can validate your startup; when there is no comparable reference, a benchmark is shown as context and never blended into the call. Every gate is advisory, and a human — you — reviews and decides at each one. Nothing auto-advances or auto-kills on a signal.

I would rather admit the limits than dress them up. Sharper calibration is a path we are deliberately working toward, not something we will imply we already have. And some questions — like whether a company will ultimately succeed — can never really earn that kind of authority. They stay advisory by design.

Treat scores as support, not a sentence

Treat every score and gate as decision support, not a sentence. Ask what went into it, whether the sample was big enough, what context the benchmark came from, and how much uncertainty is left. A recommendation that admits its limits is stronger, not weaker.

Do not let a number replace judgment. Numbers can discipline intuition; they can also disguise a weak assumption. The best version of this discipline is not less quantitative — it is more honest about what the quantities mean. Used that way, Innovation Physics just means disciplined learning: respect the real constraints, chase stronger evidence, and keep the decision tied to what you have actually seen.