Methodology

How a paper becomes a row you can cite

Every paper moves through the same 6 stages, and each stage records what was done.

The 6 stages

1
Discover
Automatic
Search the literature indexes or import your own references. Duplicates across sources are merged.
You write the question and the search. Uncertain duplicates wait for you.
You set the scope: the materials and properties you care about, and your inclusion criteria. Each search is saved with its exact query, source, filters, date and result count, so it can be reported and repeated. References you import from other databases join the same pool, and records that describe the same paper are merged, keeping where each one came from.
2
Screen
Suggested, you decide
Each paper is checked against your criteria, with a suggested decision and its reason. You decide.
You include, exclude or defer. A suggestion never replaces your decision.
Each title and abstract is compared with your written criteria. You see a suggested decision, the reason for it and how certain it is. You include, exclude or mark the paper as maybe; your decision is final and a later suggestion never replaces it. Criteria are versioned, so a decision can always be read against the rules in force when it was made.
3
Full text
Automatic
Open-access full text is collected automatically. For the rest, use your library or upload the file.
Papers that cannot be collected lawfully are listed for you to add.
For every included paper the platform looks for an open-access copy and records why a paper could or could not be collected. Where no open copy exists, the paper is listed for you: follow the link through your own library, or upload the file. Every file is checked against the paper it is attached to before it is used.
4
Extract
Suggested, you decide
Values are taken from text, tables and figures, each with the passage it came from and its original unit.
Every value keeps its source. Uncertain values are flagged, never silently fixed.
The document is read into sections, tables and figures. Candidate values are taken out with the sample, treatment and conditions they belong to, and with the exact passage, table cell or figure they came from. The value and unit as printed are always kept. A standardised value is added only when the conversion is unambiguous; otherwise the value waits for a reviewer.
5
Review
Decided by a person
Each value is shown beside its source. A reviewer accepts, corrects or rejects it, and that is recorded.
A named reviewer accepts, corrects or rejects every value.
The reviewer sees each value beside its source and any flags raised against it, then accepts it, corrects it, rejects it or asks for clarification. The original value, the correction, the reviewer and the time are all recorded. Rejected values are kept on record and never reach the dataset.
6
Dataset
Automatic
Approved rows export as CSV, XLSX or JSON with reported and standardised values, conditions and sources.
Only reviewer-approved rows, with their sources.
The verified dataset holds approved rows only. Exports carry the reported and standardised values, the conditions, the identifiers and the source of every row, with stable column names. A provenance file lists the searches, decisions and reviews behind the dataset.

The rules every value follows

Every value has a source: A value without an article, a page and a passage, table or figure is not stored as evidence.
Raw and standardised, side by side: The value and unit as printed are never overwritten. The standardised value records the conversion that produced it.
No conversion on a guess: A unit is converted only when the property and context make it unambiguous. “g” on a texture analyser is not treated as a force unless the paper says so; a percentage is not moved between weight and volume without a density.
A person decides: Suggestions carry a reason and a confidence. Screening and review decisions belong to people and are never overwritten.

Known limits

Accuracy depends on the material: It varies with the property, the source (text, table or figure) and the quality of the document, so it is checked per project against values a person has verified.
Values read from figures are estimates: A value taken from a plot is limited by the figure’s resolution and stays marked as read from a figure until a reviewer confirms it.
Full text is limited by access: Only open-access copies are collected automatically. Other papers wait for your library access or an upload.
Some values cannot be standardised: When a paper omits what a conversion needs, the value is kept as reported and flagged rather than converted.

More detail is in the help pages and the property glossary.

The 6 stages

Discover

Automatic

Search the literature indexes or import your own references. Duplicates across sources are merged.

You write the question and the search. Uncertain duplicates wait for you.

You set the scope: the materials and properties you care about, and your inclusion criteria. Each search is saved with its exact query, source, filters, date and result count, so it can be reported and repeated. References you import from other databases join the same pool, and records that describe the same paper are merged, keeping where each one came from.

Screen

Suggested, you decide

Each paper is checked against your criteria, with a suggested decision and its reason. You decide.

You include, exclude or defer. A suggestion never replaces your decision.

Each title and abstract is compared with your written criteria. You see a suggested decision, the reason for it and how certain it is. You include, exclude or mark the paper as maybe; your decision is final and a later suggestion never replaces it. Criteria are versioned, so a decision can always be read against the rules in force when it was made.

Full text

Automatic

Open-access full text is collected automatically. For the rest, use your library or upload the file.

Papers that cannot be collected lawfully are listed for you to add.

For every included paper the platform looks for an open-access copy and records why a paper could or could not be collected. Where no open copy exists, the paper is listed for you: follow the link through your own library, or upload the file. Every file is checked against the paper it is attached to before it is used.

Extract

Suggested, you decide

Values are taken from text, tables and figures, each with the passage it came from and its original unit.

Every value keeps its source. Uncertain values are flagged, never silently fixed.

The document is read into sections, tables and figures. Candidate values are taken out with the sample, treatment and conditions they belong to, and with the exact passage, table cell or figure they came from. The value and unit as printed are always kept. A standardised value is added only when the conversion is unambiguous; otherwise the value waits for a reviewer.

Review

Decided by a person

Each value is shown beside its source. A reviewer accepts, corrects or rejects it, and that is recorded.

A named reviewer accepts, corrects or rejects every value.

The reviewer sees each value beside its source and any flags raised against it, then accepts it, corrects it, rejects it or asks for clarification. The original value, the correction, the reviewer and the time are all recorded. Rejected values are kept on record and never reach the dataset.

Dataset

Automatic

Approved rows export as CSV, XLSX or JSON with reported and standardised values, conditions and sources.

Only reviewer-approved rows, with their sources.

The verified dataset holds approved rows only. Exports carry the reported and standardised values, the conditions, the identifiers and the source of every row, with stable column names. A provenance file lists the searches, decisions and reviews behind the dataset.

The rules every value follows

Every value has a source

A value without an article, a page and a passage, table or figure is not stored as evidence.

Raw and standardised, side by side

The value and unit as printed are never overwritten. The standardised value records the conversion that produced it.

No conversion on a guess

A unit is converted only when the property and context make it unambiguous. “g” on a texture analyser is not treated as a force unless the paper says so; a percentage is not moved between weight and volume without a density.

A person decides

Suggestions carry a reason and a confidence. Screening and review decisions belong to people and are never overwritten.

Known limits

Accuracy depends on the material

It varies with the property, the source (text, table or figure) and the quality of the document, so it is checked per project against values a person has verified.

Values read from figures are estimates

A value taken from a plot is limited by the figure’s resolution and stays marked as read from a figure until a reviewer confirms it.

Full text is limited by access

Only open-access copies are collected automatically. Other papers wait for your library access or an upload.

Some values cannot be standardised

When a paper omits what a conversion needs, the value is kept as reported and flagged rather than converted.