The order of operations: what 14,362 recipe steps say about the rules you were taught

Mentions of parsley, cilantro, basil, chives, dill or mint (n=966 step-mentions) sit at a mean relative position of 0.79, with a median of 1.00 — the single most common place to meet a fresh herb is the final step. 72.7% of all fresh-herb mentions fall in the last third of the recipe. Dried herbs and bay leaves (n=195) sit at 0.51 — the middle of the dish, right where the simmer starts.

The rule’s logic is sound — garlic burns faster than onion sweats. But of the 785 recipes whose steps mention both:

The most common practice is the one the rule forbids. Either a lot of recipe writers are wrong, or — more likely — the rule matters at the margin (high heat, long sauté) and most home recipes never reach the margin.

862 of 2,345 recipes — 36.8% — never mention salt in any step. Not “add salt,” not “salted water,” not “season with salt” (our matcher counts all of these). Salt appears in ingredient lists and then silently vanishes from the instructions in over a third of recipes. If you learned to cook from written recipes alone, a third of your teachers never told you when to salt anything.

Mentions of cumin, paprika, curry powder, chili powder and turmeric (n=531) sit at a mean position of 0.26; 60.5% fall in the first third of the recipe and only 5.3% in the final third. Written practice agrees with the rule: ground spices are an opening move. (The exception the data can’t see: finishing spices like garam masala — deliberately excluded from the list because it’s taught both ways.)

An earlier internal probe of the same data reported fresh herbs at mean position 0.93 (we measure 0.79 — herb lists differ, and ours is published) and 29.8% of recipes never mentioning salt (we measure 36.8%, despite our matcher also counting “salted” and “salting”). Its garlic-vs-onion split we reproduce exactly — once plurals count as mentions; our first run with singular-only matching found 694 recipes and a 40/41/19 split, a reminder of how much the matcher is the method. Directions never flipped; magnitudes move with the term lists. That is exactly why every term list is published — change them and see.

• Mentions ≠ additions. “Remove the bay leaf” counts as a bay-leaf mention; “season to taste” without the word “salt” does not count as salt. Step text is a proxy for practice, not a log of it.
• The step data is partly machine-structured, and ~40% of recipes are project-authored — their step text was written against a schema, then verified. Authored recipes may state sequencing more explicitly than the wild average.
• Relative position flattens real time. Step 3 of 4 might be “simmer for three hours.” Only 47.5% of steps carry an explicit duration, so we used the step count, not the clock.
• Term lists are choices. We published ours; reasonable alternatives move the numbers by a few points. None we tried flip a conclusion.

Every number on this page is a deterministic measurement over the same recipe corpus — nothing here was estimated or embellished. Want to check the work, or run your own analysis? A 250-dish sample of the corpus — ingredients, 18-note flavor profiles, timing, macros — is free to download.