1. Collect permitted official data
Only enabled entries in the source registry may run. Where possible, an official metadata API identifies the current resource. Each download records its URL, timestamp, response metadata, byte size and SHA-256 checksum; raw bytes are not edited.
2. Validate before import
Parsers declare the supported schema and required columns. Unexpected schema changes fail closed. Invalid rows are rejected with a reason instead of being silently repaired, and numeric values are checked against the relevant published range and unit.
3. Preserve identity and market
Raw brand and model values are retained. Comparison keys normalise Unicode, spacing and obvious dash characters, but meaningful punctuation and market suffixes remain. Australia and France are separate jurisdictions. The 722 current regional relationship candidates remain DO_NOT_MERGE unless manually resolved with evidence.
4. Keep assertions source by source
Each assertion retains its product, market variant, field name, value, unit, source, source record, source URL, observation date, evidence tier and methodology version. Conflicting values are displayed separately rather than averaged.
5. Use careful evidence language
A French durability value is an official statutory score. Australian energy data is official registration and label data. A missing recall match means only that checked sources returned no match; it is not proof that a product is safe. Lasting Index has not physically tested the listed products.
6. Correct without erasing history
A correction does not overwrite the original assertion. Reviewed changes create a versioned re-ingestion or a superseding assertion with provenance, leaving an audit trail of what changed and why.