Educational

Every so often a recipe comes through with an ingredient that can't be pinned down. A regional flour. A brand-specific sauce. A quantity written as "a splash."

At that moment there are two options. Substitute a plausible average, produce a clean total, and say nothing. Or say what happened.

We do the second one, and it's worth explaining why, because a flagged calculation looks less impressive than a confident one.

How a recipe becomes a number

The pipeline has a few stages, and each is a place where certainty can be lost.

First, the page gets read. If it publishes structured recipe data — the machine-readable format most large recipe sites use — that's the fast path, and the ingredients arrive clean. When that data is missing or incomplete, the text gets parsed instead, which is a harder problem: ingredient lists are written for humans and humans tolerate a lot of ambiguity.

Second, each ingredient gets a quantity in grams. "Two cups of flour" needs a density. "One large onion" needs an assumed size. "Chicken thighs, bone-in" needs a bone deduction.

Third, each ingredient is matched against reference data for carbohydrate, fibre and glycemic index. Then the glycemic load arithmetic runs on what survived all three stages.

Where the uncertainty actually comes from

Most of it isn't exotic:

• Vague quantities. "A splash of soy sauce", "season to taste", "a handful of coriander". These are fine in cooking and unhelpful in arithmetic.
• Ingredients with no tested GI value. The published glycemic index literature covers a few thousand foods. It doesn't cover everything in every cuisine, and coverage is thinner outside the foods that Western research groups happened to test.
• Brand and regional variation. Two products with the same name can differ substantially in sugar content.
• Divided ingredients. "1 cup oil, divided" where half goes into the pan and gets drained away.
• Serving counts that don't match reality. A recipe claiming eight servings from a tray that realistically feeds four.

That last one is the most common and the least visible. Every per-serving number on any site depends entirely on the recipe author's opinion about how many people the dish feeds.

Why we flag rather than fill

A confident-looking total that quietly contains a guess is worse than a slightly awkward one that shows where the guess is, for three reasons.

You can't check what you can't see. If an unusual ingredient got assigned an average value, the only person who could catch that is the one who knows what the ingredient actually is — the cook. Hiding it removes the one correction mechanism available.

Errors compound quietly. One substituted value in a dish with fifteen ingredients is usually minor. The same habit applied across a whole catalogue produces a database that looks precise and drifts.

And it's the same promise the tool is built on. The whole reason for showing the calculation is that a number without its working is something you have to take on faith. A gap that gets papered over is a place where you're being asked to take it on faith without being told.

What to do when you see a flag

Usually the fix takes ten seconds, because you know something the parser didn't.

If a quantity was ambiguous, put a real number on it — you know whether that was a tablespoon or three. If the serving count is wrong for how you actually eat the dish, change it; that single edit moves the per-serving figures more than anything else. If an unusual ingredient couldn't be matched, check whether it's carbohydrate-bearing at all. A missing spice changes nothing. A missing sweetener changes everything.

The breakdown is there to be argued with. It's laid out per ingredient specifically so a wrong line is visible as a wrong line rather than buried in a total.

What a flagged number is still good for

A calculation with one uncertain minor ingredient is not useless. It's a good estimate with a known soft spot, which is a normal state for any measurement.

The practical question is whether the uncertain part is big enough to matter. An unmatched pinch of something doesn't change a glycemic load. An unmatched cup of a starch does. The flag tells you which situation you're in, and that's the entire point of it.

The one gap we can't flag

There's a category of uncertainty the tool can't see at all: what you actually do in the kitchen.

Recipes get adapted. People use two-thirds of the sugar, swap the rice for cauliflower, leave out the honey, add an extra tin of beans. Every one of those changes the number, and none of them is visible to anything reading the original page.

That's not a defect to fix so much as a reason to treat an extracted recipe as a starting document rather than a finished record. If you cook something regularly with your own modifications, editing the ingredients to match what you actually do makes every number downstream real rather than theoretical.

The broader version of this

Published GI values themselves come from testing groups of ten to fifteen people, and individual responses to the same food vary considerably. So even a fully matched, perfectly parsed recipe produces an estimate rather than a prediction for your body.

That's not a reason to distrust the number. It's a reason to treat it as what it is: a well-grounded starting point that ranks foods reliably and estimates your personal response approximately.

Showing the gaps is how a tool tells you which parts of its answer to lean on. We'd rather hand you something honest and slightly untidy than something clean that quietly made things up.