/ Frameworks

The operator frameworks I actually use.

Steal-able. Named. Versioned as I learn. Each one is a decision tree, checklist, or rubric I've used to ship product across six industries. Copy the ones that fit your team.

/ 01

Six-axis content eval rubric

Score any piece of writing — essay, post, doc — across six axes before publishing.

Use before publishing anything you want to compound over time.

Authority buildingWould another expert cite this?

Genuinely usefulCan the reader act in 24h?

Upgrades skillDid the reader gain something transferable?

SEO + AEOOne target query identified, lead-with-the-answer?

EngagingFirst sentence makes the next inevitable?

Concise + clearCould this be 30% shorter without losing substance?

gate → Total ≥ 12 / 18, no axis below 2.

/ 02

Four-question scoping frame

Cut a sprawling spec down to its decisive moves before writing a line of code.

Use the day a new product, feature, or initiative lands on your desk.

1.What is the user’s actual job? (Test: what happens to them if this disappears?)

2.What does the distribution channel reward? (Who pays vs. who uses?)

3.What is the smallest evidence we’re useful? (Evidence-generating, not shippable.)

4.Where is the team strongest? (Don’t fight the shape of the room.)

gate → If you can’t answer #1 in one sentence, scope is wrong.

/ 03

LLM eval pattern stack

Build a real eval suite when there is no ground truth and no labeled dataset.

Use the week your AI demo starts crossing the chasm into product.

Pairwise preferenceIs A better than B? (Easier than scoring absolute.)

Rubric gradingScore 4–6 named dimensions 0–3 with an LLM judge.

Behavioral assertionsUnit tests for every failure mode you’ve actually seen.

Drift detectionSample 1% of prod outputs, alert on distribution shift.

User feedbackThumbs + conversation completion as eventual ground truth.

gate → No single signal is enough. Compose at least three.

/ 04

First-10-customers playbook (B2B AI)

Recruit ten customers who teach you a generalizable product, not ten logos.

Use from day one of a B2B AI startup, until you can predict yes/no before a demo.

Move 1Pick a workflow you can run live in the meeting.

Move 2Trade hand-built integrations for design-partner rights.

Move 3Charge a small but real price — corporate card-and-forget.

Move 4One founder, ten customers, no salespeople.

gate → Ready for #11 when you can predict the close before the demo with 70% accuracy.