Evidence and measurements
The verified dataset
What a verified value is, how the dataset is organised, and how to filter it.
What a verified value is
A verified value is a candidate that passed the quality checks and was approved by a named person. Verified values are stored one row per measurement, and that table is the project’s scientific record. The legacy FoodProt workbook is one view of it.
One row per measurement keeps the dataset easy to analyse: a protein and property pair exists without empty cells for all the properties a paper did not measure.
Current and replaced rows
Rows are versioned, and only the current version is shown by default. Correcting a verified value creates a new current row and marks the old one as replaced, so the dataset can show what changed between two exports and who changed it.
Filtering
You can filter by property, protein, source type, quality-check status, unit and reviewer. The count on screen is the real number of matching rows.
- Filter by source type to separate values read from tables from values read from figures.
- Filter by quality-check status to find rows that passed with warnings.
- Filter by reviewer to check consistency across a team.
What it does not contain
The verified dataset contains only what a person approved. Papers excluded at screening, rejected candidates and values with open flags are not in it, and each can still be found in the stage that produced it.
Related
Quality checks
Fixed rules, three levels of flag, and why open warnings also hold a value back.
Exports and the audit trail
What the FoodProt AI export contains, who can export, and what the provenance columns let a reader check.
Reviewing extracted values
How to read a candidate value, the four review decisions, and who can give final approval.
Data dictionary
Every field in the FoodProt AI export, listed in the order it is written.