Methodology

Every figure, its source

Every fact is an observation, a link or an event with at least one citation. The citation points to an immutable document, identified by its sha256 fingerprint, and to a typed locator: page, sheet and cell, CSV row, service record or JSON path. The only figures without a direct citation are derived ones, which state the cited data they come from.

Explicit basis, never mixed

Every figure states its basis. We only add up figures at 100% of distinct assets. The attributable figure is computed as 100% of the asset times the ownership percentage.

100% of asset

The asset's total, not adjusted for ownership

Attributable

The share attributable to an owner: 100% of the asset × its stake

Company own

The company's output at its own operations

Company incl. equity stakes

Includes stakes in other operations; never summed

No basis

The metric doesn't depend on ownership

No silent pro-rating

When a source reports several units together, for example two divisions in one figure, the data belongs to an asset group. We do not split it among its members.

Versions and restatements

Nothing is overwritten. A new version of a source inserts a new record that supersedes the previous one, and a change larger than 1% is recorded as a restatement.

Not disclosed is not the same as no data

“Not disclosed” links the document we checked that does not report the figure. “No data yet” means we have not added a source for that item.

Review badges

Every data point shows how it reached the platform:

• StructuredStructured (official source)
• AI-verifiedAI-verified (dual extraction, quote checked)
• Manual reviewManually reviewed (founder)

Independent dual extraction

Values we read from PDF reports are extracted twice, independently and blind. They are published only when both extractions agree and a deterministic check finds the verbatim quote on the stated page. Disagreements go to manual review.

Licences and visibility

Each source has a legal status (cleared, provisional or blocked) and a display policy. The database enforces both: the public site only shows cleared sources and never quotes beyond what the licence allows.

Provenance audit

Every week we check a sample of published values against the stored excerpt of the document and its locator. The result is published on Sources.

Personal data

For people we store only name, role, employer, dates and source. Never their national ID (RUT), email, phone, address or photo.