PKU Market Atlas Everything PKU, at your fingertips. Open the atlas
For Nutritionists · Researchers · Clinicians

A standard you can inspect, test, and trust.

Until now there's been no way to judge whether a phe-estimator app is accurate. PKU Commons gives you the method, the data provenance, and a public benchmark, open to your review.

How a phe estimate is produced

The Skill encodes the method a trained PKU parent uses, made explicit and reviewable.

  1. Countable? Screen the label to the diet-relevant range (≤ 2 g protein per serving), where precise phe counting matters most.
  2. Find phe-bearing ingredients. Parse the ingredient list for items that contribute phe.
  3. Classify by protein source. Assign a phe-per-gram-protein value per ingredient by type.
  4. Apply a recipe factor. Estimate each phe-bearing ingredient's relative weight in the recipe from basic food science.
  5. Scale to portion & log. Multiply by the weighed portion; record for tracking and clinic sharing.
We deliberately constrain the AI's non-deterministic reasoning and add explicit guidance for foreseeable conditions (e.g. weak grocery Wi-Fi, partial labels), reducing variance so estimates are consistent and defensible.

How accuracy is validated

This is the part built for you: a reproducible standard, verified independently of vendor claims.

Reproducible ground truth. Each benchmark case pairs a food label with a phe value computed from documented public sources, USDA FoodData Central and Open Food Facts, with the source id recorded. You (or anyone) can re-derive the "true" value independently.

Transparent metrics. Every estimator is scored the same way: mean absolute error, per-case error, and the share of estimates falling within a clinically-relevant tolerance band.

Versioned & audited. Methodology rules and food-list rows carry a citation to authority, a version, and a reviewer of record. Your challenges become tracked issues with documented resolutions, recorded in an audit trail anyone can inspect.

Your sign-off has standing. Clinician / RD review is a first-class authority layer: a named professional endorsement is a citable source in the record.

Read the benchmark method   The two-layer review model

Download & test Phebe

Phebe is a phe-estimator and logger. Try it on labels you know and compare against your own judgment.

Phebe is a reference implementation of the PKU Commons standard a working example of the open method and benchmark, not the standard itself. Any app can implement the standard.

Tell us what you think

What must a phe-estimator get right to be useful in your clinic or study? Where would an estimate be dangerous if wrong? Your input directly shapes the tolerance bands and the method.

Shape the standard:

Tell us the accuracy your clinic or study needs, the failure modes that concern you, and the data you'd want to see. Your input directly informs the tolerance bands and the method, and becomes part of the review record once filed here.

No login required. Prefer GitHub? Open a feedback issue instead