Why product-market fit alone never got anyone to $100M — and the three other fits nobody checks. Balfour's framework, with four cases.

The default shape for any "design X" or "improve Y" question.

For "critique this app / flow" questions: audit, do not redesign.

Decide when an AI should act on its own, ask, or stay out of the way.

Who pays, for what value, how much, in what model, packaged how.

Pick a North Star that measures value delivered, not activity generated.

For "how would you grow / 10x / monetise X" — one defended bet, not a list.

Retrieval-Augmented Generation — let the model look things up instead of making them up.

Turn meaning into coordinates so "similar" becomes a distance you can compute.

Teach the model new behaviour vs just asking better.

The model states false things with total confidence — by design, not by bug.

Unit tests for model behaviour — the difference between shipping and guessing.

Models that take actions in a loop — plan, act, observe, repeat.

During a partition you can’t be both linearizable and available. Most systems are neither CP nor AP.

Rows by key in milliseconds, or billions of rows by column in seconds.

Split the data, then decide where the pieces live, and what happens when one gets hot.

Elect a leader with an up-to-date log, replicate to a majority, and mind the current-term rule.

Promise, then accept: two majority rounds, and a chosen value can never change.

Queue or log, at-least-once, idempotent consumers, and the outbox.

Read the code, call the Big-O before the reveal. Six snippets, from O(1) to O(2ⁿ).

An exam-prep app with cinematic videos, PYQs and audio notes wants a new home feed. What do you look at before designing anything?

You onboard 100 users and 8 purchase. Where do the other 92 go — and what should the push journey do about each of them?

One repeatable method for any "go look at the data" problem — decision, trend, segment, funnel, hypothesis, rule.

AARRR — Acquisition, Activation, Retention, Referral, Revenue. Find which stage leaks before fixing anything.

Group users by when they joined and watch each group over time — the only honest way to read growth.

Recency, Frequency, Monetary — split users into behavioural groups before believing any average.