Methodology

A trace runs through one pipeline: research → normalize → score → cache. When you enter a product name or URL, an AI model with live web search researches it — instructed to prefer official company pages, regulatory filings, sustainability reports, and major business press, and to cite the real pages it used.

We then normalize what it returns: countries are canonicalized to a single spelling, percentage shares are clamped to a sane range, and every source link is checked for well-formedness before it reaches your screen. Only after that does the deterministic scorer run. Results are cached so a repeat lookup is fast and consistent.

This is research assistance, not a certified audit. Treat a report as a well-sourced first pass and click through to the sources.

Every ownership, manufacturing, and materials fact carries one of four labels:

• Confirmed — multiple authoritative sources agree.
• Reported — a credible source states it, not independently corroborated.
• Inferred — a reasonable deduction from indirect evidence.
• Unknown — not disclosed, or sources contradict. We never invent a value to fill the gap.

The Local Score is a 0-100 blend of how much of a product sits in your home country, weighted across three facets. The weights are fixed and identical for every product:

• Manufacturing — up to 55 points. Earned in proportion to the share of manufacturing located in your country.
• Materials — up to 25 points. Earned in proportion to the share of key materials sourced from your country.
• Brand ownership — 20 points. All-or-nothing: awarded only when the ultimate parent company is headquartered in your country.

An unknown facet earns nothing — honesty over inflation. When both manufacturing and ownership are unknown, we show "—" instead of a number, because a score we cannot support is worse than no score.

The final number maps to a plain-language verdict:

• 85+ — Homegrown
• 60+ — Mostly local
• 30+ — Mixed origin
• 5+ — Mostly imported
• 0+ — Imported

This matters, so we state it plainly. A report can be incomplete or out of date: supply chains shift, companies disclose selectively, and the model can make mistakes. Nothing here is a guarantee — verify anything important against the linked sources.

Percentages are estimates, with their uncertainty carried in the confidence labels rather than in decimal places. Niche or newly released products may resolve poorly. A URL trace depends on how much the URL itself reveals — a product name usually works better. And the local-alternatives section only lists products the model could actually verify; an empty list is an honest answer, not a bug.