How GüdDish works

The world's recipes hold an incredible amount of knowledge. We do the work of finding it, measuring it, understanding it, and turning it into something you can rely on for every meal.

GüdScore tells us where the strongest evidence is. GüdDish turns that evidence into the recipe you should cook.

The work

01. We search

We find the strongest recipes across the web. Thousands of recipes. Dozens of sources. Rated pages with verifiable reviews, and unrated methods (including video) when they show a real technique worth studying.

02. We measure

GüdScore evaluates the evidence and finds the recipes that deserve our attention. It does not claim to hand you "the best recipe" by itself. It surfaces the strongest rated methods to learn from.

03. We learn

We study what the strongest recipes have in common: ingredients, techniques, proportions, and details. A winning move can come from the #1 recipe, the #7 recipe, or an unrated method that never entered the score.

04. We build

We then distill the best of what we find into a single recipe — the GüdDish. Not a reprint of whoever sits at #1 on the Rank tab.

That is the primary methodology. Everything below is how we prove the measurement step is honest.

The evidence

How GüdScore works

GüdScore is the evidence engine behind GüdDish. It accounts for how different publishers rate recipes, then orders versions of a dish by how strongly the public evidence supports each one. We use that order to decide what to study deeply. GüdDish does the rest.

How a recipe earns attention

  • Site baseline. Every publisher has its own typical rating level. A 4.9 on a source that usually rates near 4.9 is different from a 4.9 on a stricter source.
  • Lift. We measure how far a recipe sits above (or below) its own site's baseline.
  • Shrinkage. Low-vote recipes are pulled toward the mean so a handful of five-stars cannot outrank a well-tested classic.
  • Conservative ordering. Recipes are ordered by a lower-confidence bound on that lift, not by raw star averages alone.
  • Contender floor. The public ranking only includes recipes with enough ratings for the size of that dish's pool (from about 10 votes on very small dishes up to 30 on large ones). That keeps thin dishes honest without a one-size-fits-all cutoff.

What enters the score

We only put a recipe in the scored ranking when we can check the evidence behind its rating at the review level.

  • Sites that only publish closed aggregate stars, without recoverable review data, stay outside the score (not quietly down-weighted). Their recipes may still appear as method references or labeled publisher claims.
  • Where advertised averages fail integrity checks, we rebuild ratings from raw public reviews when those reviews are available, before a recipe competes.
  • Sites with no public star ratings (for example many magazine-style recipes) are out of scope for the score by design.
  • Cookbooks, restaurants, non-English sources, and social video are not in the scored ranking. Unrated and video methods can still inform The GüdDish when they teach a strong technique.

What the Rank tab is allowed to claim

On the ranking layer, the public claim is order: which recipe ranks #1, #2, #3 among recipes we could fairly measure. We do not put a universal 0 to 100 grade on every recipe. Places may use short labels (#1, #2 through #5, and so on) so a thin corpus can still show a clear leader without reading like a school score. Close places can reverse as more ratings arrive.

  • A clear #1 pick with corpus size and a link to the publisher's recipe
  • A ranked shortlist of contenders, with a note that close places can reverse as more ratings accumulate
  • Corpus stats: recipes rated, sites, total ratings counted, overall confidence
  • Freeze or reopen status when the completion policy has been applied

What we host and what we don't

We don't host the ranked publisher recipes. We publish the ranking and link to the publisher for the method. Those cooks did the original work. We studied it.

The GüdDish is different. It is our synthesis of the field: one recipe you can cook on the Cook tab, built from what the strongest methods taught us. We are not pretending we invented every technique. We are saying we did the research and wrote the best version.

For tools and citations

Dish pages include structured rank lists, links to this methodology and coverage, and machine-readable snapshots under /api/ for tools that need a stable citation. When you cite the ranking layer, use the winner and rank (for example "#1 pot roast: …"), not an invented letter grade. When you cite The GüdDish, say it is GüdDish's synthesized method for that dish.

See also what we scanned and llms.txt.