How We Verify Our Furikake Data

Where the data comes from

We source each product's name, ingredients, and allergen facts from the manufacturer's official page or package label—not a fan wiki or third-party roundup. A maker-operated shop may supply a product identifier such as a JAN only when the source note isolates that use; it does not replace the maker's label as ingredient evidence. No official primary source for the product facts means no publication.

Display images follow a separate provenance rule. We first-party host a documented product image only when it is cleared under our image policy. If reuse permission cannot be confirmed, we omit package art; compact product cards use a flavor symbol instead. A site-original neutral illustration may be used when its publication basis is recorded, and its visible alt text identifies it as illustration—not package art. Official image URLs remain source references only.

Cross-checking the label itself

Automated extraction isn't always reliable. Some manufacturer pages show allergens as a color-coded matrix that an automated summary can misread as "every allergen applies." Some package ingredient text exists only inside a product photo, not as selectable text. And some labels put the legally required "contains" notice at the very end of a long ingredient string, where a summary can drop it. Wherever extraction looked uncertain, we checked the raw page source or the actual package image by eye instead of trusting the first pass.

Every file is machine-validated

Each product's data is stored as a structured file and validated against a schema before publishing: required fields must be present, allergen and category values must match an approved list, and no product can list the same allergen twice. This catches typos and structural mistakes before they reach the page you're reading.

How we store label evidence

Across 28 current product records, we store 75 positive ingredient-category flags. These categories include broader dietary checks such as fish, meat, and alcohol; the total is not a count of statutory allergen claims.

Every positive flag cites at least one literal Japanese excerpt from the recorded ingredient list or maker allergen list. Those positive flags contain 137 source-bound quote records. One flag can cite several excerpts, and the same excerpt can support more than one category, so quote and flag totals are intentionally different.

We keep 7 editorial notes in a separate field. Notes explain scope or uncertainty; they are never counted or displayed as quotations. The database also has 1 evidence-bearing "unable to determine" flag. That evidence stays separate from positive claims rather than turning uncertainty into a yes.

We don't invent English names

Not every manufacturer publishes an official English product or brand name. When one doesn't exist, we leave that field blank rather than write our own translation and present it as official. An empty English name field means exactly that — no official name exists yet, not that we forgot to fill it in. For shelf matching, we separately show an editorial romanization tied to the exact Japanese source name. It is a navigation alias, not an official English product name or a translated ingredient claim.

Unreviewed or stale translations stay out of public product copy

Our working translation files use automation and a shared glossary for recurring terms such as おかか, 明太子, and 昆布. Each file carries a review status. While it remains an "ai_draft," its expanded English name, ingredient translation, taste summary, and usage copy are not published in page text, search descriptions, or Product structured data. Public product pages use the original Japanese label plus normalized ingredient and allergen facts. English prose can appear only after a separate human review records the reviewer and date, the exact primary-source URL and crawl timestamp, and a SHA-256 fingerprint of the fixed Japanese source payload. The site recomputes those bindings from the current product record before publishing. Any mismatch returns the translation to pending until a person reviews it again. The physical packet remains the final source because formulas can change.

Every product shows its date and source

Every product page displays the date its label was last checked and links directly to the official page it was checked against — so you can check the primary source yourself.

What we know we still get wrong

Japanese ingredient labels don't sort seafood the way English labels do. Squid, octopus, krill, and shellfish-derived ingredients like calcium don't always fall cleanly under "fish" or "shellfish." We hit this directly: a product that looked plant-based at a glance — no listed fish, no listed shellfish — turned out to contain squid and krill once we checked its full ingredient list and allergen codes. We now cross-check the specific allergen list against the fish/shellfish fields before calling anything plant-based — a real gap in mapping Japanese labels to English dietary categories, and a big reason we mark things "unable to determine" rather than guess.

Found a mistake?

We're one person checking every product by hand against official sources — we will get things wrong sometimes, or fall behind a label change. A public contact channel for error reports is coming soon; until then, treat any product page's "Label verified" date as the point to double-check against the manufacturer's current official page, linked from every product.