Skip to content
Forrest Miller

Reliability & operations

RecipeStripper

Strip recipes down to just ingredients and cooking instructions — no ads, no life stories, no scrolling.

Open recipestripper.com (opens in a new tab)

Who it's for
Home cooks who just want the recipe.
The problem
Recipe sites are bloated because bloat pays. And you still scroll back up on every step to re-check an amount.
What it does
Paste a URL; get ingredients, steps, and a cook mode. Every amount is written into the step that uses it — no scrolling back.
RecipeStripper's mobile homepage: the strip-recipes headline, URL input with Go button, and the before/after of an ad-choked recipe blog beside the same recipe stripped clean.
recipestripper.com — paste a URL, get the recipe
A stripped recipe on mobile: Chicken tikka masala with Email, Share and Save actions, a serves stepper, Cook Mode toggled on, and the numbered ingredient list.
The result — serves stepper, Cook Mode, clean ingredients
Recipe Steps on mobile: numbered steps with each ingredient amount highlighted inline in the step that uses it.
Recipe Steps — every amount inline, no scrolling back

Opportunity

Recipe traffic is enormous, and the bloat is the business model.

Recipe traffic is enormous — Allrecipes alone takes about 104 million visits a month — and the pages are bloated on purpose: more scroll means more ads. The escape hatch has measurable demand: 3,100 monthly searches for 'just the recipe.'

RecipeStripper races a deterministic parser against two AI providers — any page, any language, no markup required. Each attempt records its source and parser path.

What I learned

The failure log knew more than the roadmap.

The failures clustered into root-causeable classes. A 45-day audit read a 24% failure rate and explained every miss — and exposed a production AI key that had been silently dead.

The failure log knew more than any feature backlog. Each logged failure names its cause, so one fix retires a whole class.

Pivot & results

The log became the roadmap, and the success rate followed.

What changed

  1. Made the failure log the roadmap. Monthly, the evidence goes to a coding agent with one goal: retire a failure class.

  2. Shipped nine fix batches in one day, each straight off the log, each verified in production.

  3. Accepted the clean audit. One audit found nothing worth fixing and shipped nothing — a valid result, on the record.

  4. Corrected the public counter down and closed the endpoint that let outside scripts inflate it.

Measured results

45.2% → 92.3%

monthly strip success rate, first month vs this month

2,480

recipes stripped across 663 different sites since launch

39 → 675

Google clicks, May baseline vs current

Corrections & failed tests

corrected down

4,886 → 1,743

public recipe counter corrected down — it was counting page views

What's next

YouTube support builds when 15 distinct visitors fail on YouTube links in a month — the trigger is pre-committed.