Where teams and agents find playbooks that work

The empirical design reference for onboarding, paywalls, and habit loops. 175 scored A/B tests and 18 full-funnel blueprints.

Synthesized from 175 controlled experiments shipped at

Duolingo Brilliant Blinkist Headspace MasterClass Codecademy Babbel Quizlet Pinterest Slopes Coursera Speak

Growth research took weeks. Now, it's one search.

Every team has that folder. Dozens of paywall screenshots saved in Figma, yet nobody knows if Variant B actually beat Control.

Interactive 3D Deck · Drag or Click to Flick

Flick through the highest-lift winners.

Swipe left or right—or use your arrow keys—to cycle through 7 shipped A/B test screens with spring physics.

01 / 07

One library for every growth lever.

Give your AI agents access to the library.

Stop letting coding agents invent generic SaaS templates. Connect the Growth Library skill so Claude, Cursor, and Figma ground every flow in shipped experiments and verified lift data.

Claude Code Cursor ChatGPT Figma
Explore `.claude/skills` Included in repository

Building in EdTech or Subscription SaaS?

Explore 32 deep-dive experiments from Duolingo, Brilliant, MasterClass, Quizlet, Babbel, Codecademy, and Coursera—including negative tests that failed in production.

Open EdTech Vault
FAQ

How the playbooks are verified.

Every readout is graded across sample size, p-value disclosure, and source tier (Tier A official engineering blogs like Duolingo & Netflix; Tier B verified teardowns via abtest.design and SubClub). We also include negative and flat experiments—such as Duolingo's failed pre-signup streak commitment and Babbel's raw traffic test—so teams see where patterns break.

A single A/B screen shows one isolated variable change. A Playbook chains 3 to 5 sequential screens into a complete behavioral flow—for example, pairing an unauthenticated interactive first lesson (Step 1) with a soft registration wall (Step 2) and a 3-milestone trial reminder timeline (Step 3).

Yes. The repository ships with structured JSON schemas (`public/data/seed.json`) and `.claude/skills` definitions so AI coding agents can look up empirical lift distributions, behavioral mechanisms, and UI copy before generating onboarding or paywall components.

Find what works. Ship what converts.

Started as a curated sheet of 68 A/B readouts. Today it holds 175 scored tests, 61 meta-analyzed learnings, and 48 side-by-side screen comparisons.