I have always liked games with odds: markets, venture capital, product bets, and any situation where smart people can see the same facts and still disagree. Certainty is unavailable, but a decision is still required.

Trading during college made this painfully concrete. A good outcome could follow a weak decision, and a strong decision could still lose money. Looking only at the result made it easy to learn the wrong lesson.

Separate the decision from the outcome

Outcomes contain skill, luck and conditions you did not control. A useful review asks what information was available, what assumptions were made, what would invalidate the thesis and whether the exposure matched the uncertainty.

Product teams need the same separation. A successful launch does not prove every choice was correct, and a failed experiment does not make the reasoning useless. The quality of the learning depends on whether the bet was explicit before the result arrived.

Size the bet to what you know

Conviction should change exposure, not eliminate humility. When evidence is limited, the decision should be reversible or small. As evidence improves, commitment can grow.

Feature flags, staged rollouts, limited cohorts and operational guardrails are product versions of position sizing. They preserve the ability to learn without making one uncertain assumption existential.

Protect against ruin

Expected upside is irrelevant if one failure can destroy the system. Markets call this risk of ruin. Products encounter it through compliance failures, irreversible customer harm, broken money movement, security incidents and operational overload.

Good systems make ordinary failure survivable. They create limits, recovery paths, audit trails and humans in the loop where uncertainty carries asymmetric cost.

Update without rewriting history

Changing your mind after new evidence is rational. Pretending you always believed the new version is not. A decision log protects against that temptation and reveals which assumptions repeatedly fail.

The goal is not to become a perfect predictor. It is to become less surprised by being wrong—and better prepared when you are.

TakeawayMake the thesis explicit, size the bet, define what would change your mind and protect the system from a single bad outcome.