Getting started
The six stages, end to end
What each stage does, and the usual reasons a stage stops and waits.
One sequence of six stages
FoodProt AI works as one sequence. A paper enters at Discover and leaves as approved values in the dataset, and every page in a project belongs to one of the six stages below. Each stage uses what the previous one produced, so skipping a stage leaves the next one with nothing to work on.
The six stages, in order, with the reasons each one stops
01
- Stage
- Discover
- What it does
- Builds the corpus: runs your search against the databases the platform may query, imports reference exports from the ones it may not, and merges records that are the same paper.
- Why it might stop
- A database is unreachable or rate-limited, no search has been run, or duplicate groups are waiting for a decision.
02
- Stage
- Screen
- What it does
- Decides which papers are in scope from title and abstract, against the criteria in the study configuration.
- Why it might stop
- Automatic suggestions are not switched on, papers arrived without abstracts, or papers are still waiting for your decision.
03
- Stage
- Full text
- What it does
- Obtains the full text of included papers where that is lawful and automatic, queues the rest for you, and parses each attached PDF into sections, tables and figures.
- Why it might stop
- The licence does not allow automatic download, the paper needs institutional access or an upload, background processing is not running, or a PDF could not be read.
04
- Stage
- Extract
- What it does
- Produces candidate measurements from tables, text and figures, each linked to the evidence it was read from, with names standardised and quality checks applied.
- Why it might stop
- No parsed full text exists for the paper, the evidence could not be verified, or a candidate carries an unresolved flag.
05
- Stage
- Review
- What it does
- A named person approves, edits or rejects every candidate beside its evidence. Nothing reaches the verified dataset without this.
- Why it might stop
- Nothing is queued, or a candidate still carries a blocking quality flag.
06
- Stage
- Dataset
- What it does
- Holds the approved measurements only: browsable, filterable, exportable, and checkable against a truth set.
- Why it might stop
- Unresolved quality errors block an export, or a truth set is uploaded but not mapped.
Two stages are decided by people
Screening asks a person to decide on every paper, and review asks a reviewer to approve every value individually. Neither can be switched off, because the dataset is called verified only when a person has checked it.
The table above uses the same descriptions as the Help panel at the top of each stage page, so the two always agree.
A stage that has stopped is usually waiting for you
Most stops are the platform declining to guess: a licence that does not allow automatic download, a conversion that needs a density the paper did not report, or a value whose evidence could not be confirmed. These wait for a person and are shown that way.
If a stage fails for every paper, or a page will not load at all, the problem is with the platform and not with your data. The troubleshooting guide says what to check.
Related
Getting started in FoodProt AI
What a project is, what the study configuration sets, and the order the stages run in.
Writing a clear study configuration
Criteria that a colleague and the platform apply the same way, and target units chosen before extraction.
Reviewing extracted values
How to read a candidate value, the four review decisions, and who can give final approval.
Troubleshooting
Common problems, what each one means, and the first thing to check.