# Accuracy and limits

> Where Kallo gets it wrong, by how much, and when not to trust the number.

Every nutrition app is estimating. The difference is whether it says so.

This page says so.

## Why a range, not a number

A bowl of noodle soup has no single correct calorie value. It depends on how much noodle
you got, the cut of meat, how fatty the broth is, and whether you finished the
liquid.

Kallo computes a plausible range, then collapses it to one figure for your log.
The number you see is a point inside that range, not a measurement.

> Useful uncertainty, not fake certainty.

## The three biggest sources of error

### 1. Portion size

By far the largest. If Kallo assumes your bowl held 500ml and it actually held
350ml, every figure is 40% off — even when the ingredient recognition was
perfect.

**How to reduce it:** answer the vessel question when Kallo asks, and give
weights when you know them.

### 2. Oil and cooking method

Oil in stir-fried or deep-fried dishes varies enormously between kitchens. The
same plate of stir-fried greens can differ twofold in fat.

**How to reduce it:** set your oil habit accurately in your
[profile](https://kallo.fit/en/docs/profile-and-targets), and state the cooking method when you
describe a meal.

### 3. Reference data coverage

Protein, carbohydrate and fat data is reasonably complete. Micronutrients are
not — many rows leave vitamin and mineral columns empty.

**Consequence:** a micronutrient showing low may mean you ate too little, or that
the data has no figure. The [Nutrition](https://kallo.fit/en/docs/tracking/nutrition) page carries a
confidence reading precisely so you can tell those apart.

## Where Kallo is strong

- **Vietnamese and Asian food** — the reference data and the recognition were
  built around it.
- **Multi-component dishes** — exactly where ingredient-by-ingredient search in
  other apps becomes impractical.
- **Repeated meals** — [re-logging](https://kallo.fit/en/docs/logging/relog) gives consistent
  figures across days.
- **Trends over time** — random error cancels out across many days.

## Where Kallo is weak

- **Unusual, rare or invented dishes** — little reference data to lean on.
- **Restaurant food with proprietary recipes** — nobody knows how much butter
  went in.
- **Micronutrients on a single day** — look at longer windows.
- **Any day you only partly logged** — missing data is worse than rough data.

## Using the numbers well

> **The trend is more trustworthy than any single day**
>
> A single day's error can be 20–30%. The average error across three weeks is far
> smaller, because random errors cancel each other out.
>
> If you are cutting, the thing to watch is your weight trend line against your
> target adherence — not whether yesterday came in 80 kcal over.

## When not to use Kallo

Kallo is not a medical device. Some situations need accuracy an estimating app
cannot provide:

- Carbohydrate counting for insulin dosing.
- A therapeutic diet prescribed by a clinician.
- Managing an allergy or intolerance — Kallo estimates nutrition, it does
  **not** verify allergens.
- Recovering from an eating disorder, where counting can do more harm than good.

In these cases, work with a health professional.

## Helping Kallo improve

If an ingredient is missing or matched wrongly, send feedback from **Settings**.
It arrives with the technical context needed to reproduce it, and goes straight
to the team.

---

_Source: [https://kallo.fit/en/docs/estimates/accuracy-and-limits](https://kallo.fit/en/docs/estimates/accuracy-and-limits) · Last updated: 2026-08-03 · Kallo docs index: [https://kallo.fit/llms.txt](https://kallo.fit/llms.txt)_