RecipeStripper
Any recipe URL becomes a clean page with the amounts written into the steps.
- 3,968
- visitorslast 90 days · as of 2026-08-03
- 2,122
- recipes strippedlast 90 days · as of 2026-08-03
- 2,291
- visitorslast 30 days · as of 2026-08-03
Source: the cross-project traffic dashboard (Amplitude, bot-filtered) · read 2026-08-03
Who it's for
Home cooks who just want the recipe.
The problem
Recipe sites are bloated by economic necessity — life story, ads, pop-ups. Even once you reach the recipe, you scroll back up on every step to re-check an amount.
What it does
Paste a URL and get ingredients, instructions, servings, and a cook mode that keeps the screen awake. The core idea is inline ingredient embedding: quantities are woven into the instructions, so you never scroll back.
What I did
The failure log decides what ships. A clean audit can decide that nothing does.
Bet on AI extraction where deterministic parsers give up. Just The Recipe stops at published schema; RecipeStripper races two AI providers and takes the first valid extraction — any language, no schema.
Made failure the roadmap. Every strip writes error type, which fetch tier got the HTML, which parser won, and a ten-value AI outcome. Monthly, a coding agent is pointed at the failed-strip distribution with one instruction: retire a failure class.
Rebased the public counter DOWN — 4,886 to 1,743 — when I caught it counting page views, and deleted the endpoint that let anything with a POST inflate it.
Declined a sibling product's bot filter on evidence: this site's Linux visitors convert at 8.9%. The same cohort is bots on one property and humans on another.
Shipped nothing from the July 14 audit, on purpose — all 477 attempts decomposed into known classes. A clean audit is a valid result.
What happened
Dated results, zeros included.
- 24% → 19.3%
- attempt-level failure rate across audit windows45d ending 2026-05-18 vs 30d ending 2026-07-14 · as of 2026-07-14
- 9 batches
- shipped in July off the failure panel's distribution, each prod-verifiedas of 2026-07-27
- 1
- AI-sourced success in the entire log while the production key sat silently dead — the audit that forced per-strip AI outcomes45d audit window · as of 2026-05-19
- 402 → 697
- Bing AI citations per rolling 3-month window (+73%)3M to 2026-05-19 vs 3M to 2026-06-27 · as of 2026-06-29Overlapping windows — a level change, not a clean delta.
- rebased down4,886 → 1,743
- the public strip counter, rebased down in publicas of 2026-07-29
What's next
YouTube support builds at 15+ distinct YouTube-domain failures in 30 days — the rejected implementation is named so it can't be re-proposed. Next failure-mix read: the August 11 device-type sweep.