1
A checker that clears everythingis indistinguishable from no checker at all. If every item comes back fine, the answer carries no information and you would be better off not asking.
REFUSAL RATE
How often we decline.
Most of what we are asked about does not clear. This page says how much, why, and what would have to change. Counted for both kinds of check: what must not be in the food, and what must stay under a number.
A high refusal rate is the product working.
That is a convenient thing for us to say, so here is the reasoning rather than the assertion. Check it against the numbers underneath.
2
So the clears are worth exactlywhat the refusals cost. Every refusal costs us coverage, disappoints a user, and will eventually disappoint someone paying. That cost is what makes an approval mean something.
3
Which is why we publish the numberand why we do not work to bring it down. The easy way to improve it is to clear on thinner evidence, and that would make the figure look better while making every verdict worth less.
The test you should apply to us.
If this figure falls sharply and coverage has not risen, we have started clearing on thinner evidence. That is the failure to watch for, it is visible on this page, and you should tell us about it.
LAYER 1 · LIVE
What has been ruled since launch
Straight from GET /v0/stats. Zero is a true number and this page is built to show it.
rolling 30 daysupdated dailysince 30 Aug 2026
STATE · LOW N
89
couldn't verify
52
ask one question
41
conflict found
32
no conflict found
214 checks ruled · no percentages shown below 500
On 30–31 Aug we found 63,601 rows in our own corpus recording “may contain” as “contains”.
We fixed the reading and re-derived every affected row. This page exists so that kind of error has nowhere to hide.
LAYER 2 · CORPUS
Why the gap exists
Measured over our evidence corpus, every menu item and product we hold, ruled against each FDA-9 restriction. Dated, reproducible, and independent of traffic.
Measured 31 Aug 2026against production evidence· 67,791 products · 711,900 store shelves · 30 metros
363,832 (51%) carry ≥1 allergen finding. The other 49% is the shop arm's refusal rate. 446,371 (63%) have a typed nutrition panel · 87% of panels state a serving basis · trans fat quantified on only 1.8% of panels · added sugar on 43%.
Eat-out: 11 chains are machine-readable · 6,858 dishes ruled · sit-down lane: 24 venues, 681 items, 227 presence-only findings.
| Cause | Share of refusals | Who could close it |
|---|---|---|
| No preparation evidence published | [CORPUS] | The venue or the manufacturer, by filing a statement. |
| Collective terms on the label | [CORPUS] | The manufacturer, by naming the source of the flavoring. |
| No label exists at all | [CORPUS] | Nobody, in-store bakery, deli counter, a home kitchen. |
| No fixed recipe | [CORPUS] | Nobody, a daily special changes by definition. |
| Venue could not be resolved | [CORPUS] | The agent, by confirming which venue it meant. |
| Question raised, unanswered | [CORPUS] | The kitchen, at the table. |
| No nutrition panel exists | [CORPUS] | Nobody, a made-on-site item has none. A finding, not a failure. |
| Panel present, serving basis unstated | [CORPUS] | The manufacturer, by stating whether the figure is per serving or per 100g. |
| Panel older than the current formulation | [CORPUS] | The manufacturer, by dating the panel they publish. |
No preparation evidence published
Share of refusals
[CORPUS]
Who could close it
The venue or the manufacturer, by filing a statement.
Collective terms on the label
Share of refusals
[CORPUS]
Who could close it
The manufacturer, by naming the source of the flavoring.
No label exists at all
Share of refusals
[CORPUS]
Who could close it
Nobody, in-store bakery, deli counter, a home kitchen.
No fixed recipe
Share of refusals
[CORPUS]
Who could close it
Nobody, a daily special changes by definition.
Venue could not be resolved
Share of refusals
[CORPUS]
Who could close it
The agent, by confirming which venue it meant.
Question raised, unanswered
Share of refusals
[CORPUS]
Who could close it
The kitchen, at the table.
No nutrition panel exists
Share of refusals
[CORPUS]
Who could close it
Nobody, a made-on-site item has none. A finding, not a failure.
Panel present, serving basis unstated
Share of refusals
[CORPUS]
Who could close it
The manufacturer, by stating whether the figure is per serving or per 100g.
Panel older than the current formulation
Share of refusals
[CORPUS]
Who could close it
The manufacturer, by dating the panel they publish.
By arm, and by premise type
Where a question can be asked, refusals fall sharply. Number checks refuse far less often than allergen checks, because a panel either exists or it does not.
| Arm | Couldn't verify | Why |
|---|---|---|
| Shop · packaged | [CORPUS] | A label is a legal declaration. Most products resolve. |
| Eat out | [CORPUS] | A person is present, so questions convert into answers. |
| Cook | [CORPUS] | Preparation is the household’s own kitchen and is not ruled. |
| Order in | [CORPUS] | No kitchen reachable. An unanswered question stays unanswered. |
| Potluck | [CORPUS] | No venue, no label, no barcode. The normal case, not the failure case. |
| Packaged · numeric threshold | [CORPUS] | One source, and 446,371 panels are rulable. The lowest refusal rate in the product. |
| Restaurant dish · numeric threshold | [CORPUS] | Almost no venue publishes a panel. Structurally the highest, and honestly so. |
Shop · packaged
Couldn't verify
[CORPUS]
Why
A label is a legal declaration. Most products resolve.
Eat out
Couldn't verify
[CORPUS]
Why
A person is present, so questions convert into answers.
Cook
Couldn't verify
[CORPUS]
Why
Preparation is the household’s own kitchen and is not ruled.
Order in
Couldn't verify
[CORPUS]
Why
No kitchen reachable. An unanswered question stays unanswered.
Potluck
Couldn't verify
[CORPUS]
Why
No venue, no label, no barcode. The normal case, not the failure case.
Packaged · numeric threshold
Couldn't verify
[CORPUS]
Why
One source, and 446,371 panels are rulable. The lowest refusal rate in the product.
Restaurant dish · numeric threshold
Couldn't verify
[CORPUS]
Why
Almost no venue publishes a panel. Structurally the highest, and honestly so.
Corpus values above were measured 31 Aug 2026 against production. They are never estimates and never filled in by hand. Order-in and potluck are structural refusals, not covered by any source, not a fake zero.
[CORPUS] values are computed from production against the evidence corpus and stamped with the date they were derived. The measurement runs after the national catalog load lands, so these fill in late. They are never estimates and never filled in by hand.
We do not optimise this number down.
The obvious way to improve it is to clear things on thinner evidence. That would make the number look better and every verdict worth less. If this figure ever falls sharply without coverage rising, assume something is wrong and tell us.
What we do optimise
Coverage, not clearance. The refusal rate falls honestly only when more evidence exists.
| Metric | Now | Direction |
|---|---|---|
| Venues with a filed statement | 0 | the statements table shipped 30 Aug, empty, up from here |
| Products resolving to a label | [CORPUS] | up |
| Questions with a sufficient-answer rule | 14 of 14 | held at 100% |
| Verdicts issued without a date | 0 | structurally impossible, held at zero |
Venues with a filed statement
Now
0
Direction
the statements table shipped 30 Aug, empty, up from here
Products resolving to a label
Now
[CORPUS]
Direction
up
Questions with a sufficient-answer rule
Now
14 of 14
Direction
held at 100%
Verdicts issued without a date
Now
0
Direction
structurally impossible, held at zero
