Methodology

Research principles

The site treats evidence discipline as part of the product. A claim can enter the content set only when the source text, publisher, publication date, access date, excerpt, archive link, and local raw snapshot are recorded in the source registry.

Statistics are not typed directly into page copy. When a page needs a statistic, the draft uses a claim tag that resolves from `content/sources.json`; if the claim is missing, unverified, or outside the source wording, the build should fail rather than soften the rule.

The research worker contract also requires the original excerpt to stay close to the source. Hedges such as company-reported, estimated, projected, up to, or according to the publisher are part of the claim, not decoration that can be removed in editing.

Dual verification

Verification has a mechanical pass and an adversarial pass. The mechanical pass fetches the source URL, falls back to archive or raw snapshots when needed, and compares the registered excerpt against the source text.

After that, a separate reviewer checks the same claim against the raw material with a hostile reading: wrong subject, changed denominator, missing caveat, mixed sources, and overstated causality are treated as failures.

Failed claims are repaired only when the source text supports a narrower statement. If the quote cannot support the proposed claim, the claim is dropped. The goal is not to preserve a storyline; it is to keep only claims that survive verification.

Source tiers

Primary sources are preferred for company actions, official product claims, investor materials, public-agency program pages, and original survey or report releases. They are useful because they anchor the exact wording, timing, and scope of the claim.

Reputable secondary sources are used when they add independent reporting, summarize inaccessible material, or provide a defensible record from established business, trade, or logistics publications.

Lower-quality sources are excluded from the source registry for quantified claims. Personal blogs, unattributed summaries, copy-farm articles, and pages that cannot be tied back to a credible publisher are not treated as statistical evidence.

Limitations

This methodology does not turn company announcements into independent audits. When a result is reported by a company, the copy should keep that attribution visible and avoid implying external validation unless a separate source provides it.

The pages also preserve timing. A claim belongs to the publication and access context recorded in the registry, so later market movement, product changes, or revised company disclosures may not be reflected until the source record is updated.

The standard is therefore modest: every statistic should have a traceable source, an archived or local copy, an excerpt that supports the wording, and a verification record. The procedure is the basis for trust, but it is not a substitute for fresh diligence before making an investment, procurement, or policy decision.