How these numbers are made

Where the numbers come from

Five read-only sources, all of them the KwickPOS network’s own operational data. Nothing is bought, scraped, surveyed or modelled from someone else’s published figures.

A restaurant counts as live, and therefore counts at all, only if it is trading, is a real food business, and is not a demo, test, dealer or non-food account. The filter is a single expression shared byte-for-byte with KwickEat, the consumer directory built from the same store master, so the two sites can never disagree about which restaurants exist.

When a figure is published, and when it is withheld

Every cell the builder computes is tested against a fixed threshold before it is allowed to be a number on a page. A cell that fails is stored as withheld and renders as “not enough data”. It is never interpolated from its neighbours, carried forward from yesterday, or shown as a zero.

A reading publishes when

the day has at least 30 restaurants and 500 orders behind it in that geography

A price index publishes when

at least 30 matching menu items are listed by at least 15 restaurants in that geography

A metro exists when

it has at least 30 live restaurants. Below that it is not offered as a geography at all — 21 metros are defined, 7 currently clear it

A store count publishes when

the store threshold alone is met: there are no orders behind a count of restaurants, so the order threshold does not apply to it

A rating publishes when

at least 30 restaurants in the geography have 50 or more Google reviews. A restaurant with fewer reviews than that is not counted as rated at all

A cuisine appears when

at least 5 restaurants in the geography cook it — a bucket of two is a step toward naming them

Of the 177,937 cells this site has computed, 144,440 are withheld for want of sample and are not shown as numbers anywhere, including in the API.

Outliers, and the dropped count

Two bounds are applied before any median is taken. They exist for data errors — a decimal point in the wrong place, a test transaction, a catering invoice rung through the till — not to tidy up real business. Anything outside them is excluded from the statistic and counted, so the exclusions are visible rather than silent.

Online order amount

kept when it is between $1.00 and $1,000.00; anything outside is excluded and counted in dropped

In-store ticket

a restaurant’s own average ticket for the day (its dollar total ÷ its order count) is kept when it is between $2.00 and $500.00; outside that the restaurant’s whole day is excluded from the median and counted in dropped

Impossible dates

rows dated in the future are excluded outright — they are data-entry artefacts, not trading

What dropped counts

the number of rows the bounds above removed from that one cell — orders for an online reading, restaurant-days for an in-store reading. It is a count of rows, not of dollars, and it is never subtracted from the sample size — n is what the figure was actually computed over, after the bounds

13,428 rows have been dropped by these bounds across every national cell this site has computed, over 479 days.

Coverage: how to tell a thin day

A holiday, an outage or a slow Tuesday can leave a day with far fewer restaurants reporting than usual. Such a day is real and is published, but comparing against it would be misleading, so every cell carries a coverage figure and the site refuses to draw a change from a thin one.

Coverage is

the day’s restaurant count divided by the median restaurant count over the previous 28 published days of the same metric and geography

Computed only when

at least 7 earlier published days exist to compare against. Before that it is stored as NULL, which means unknown — not zero, and not “fine”. A NULL never counts as low coverage and never suppresses anything

A day is flagged thin when

coverage is below 50.0% of its own recent normal. The figure is still shown, badged, and is excluded from every day-over-day and year-over-year comparison on the site

The change guard

A pipeline failure can look exactly like news. So a headline national reading that moves more than 30.0% in a single day is not published that night: the page keeps showing the last good value with a notice saying it is being held, and a warning goes to the operations channel for a human to look at.

Guarded readings

Online order ticket, Online tip rate and In-store ticket, plus every price index

Trigger

a day-over-day move of more than 30.0% against the previous published day

It clears itself when

the new level holds: once three consecutive held days sit within 10.0% of each other, the move is a real step rather than a glitch, and the reading resumes publishing from the day the step began. A guard that needed a human to clear it would turn a real market move into a permanent gap

Geography, and what a “metro” is here

National and state figures are exact: a restaurant is in the state its store record says it is in. Metros are an approximation, and this is the weakest link in the geography.

How a metro is assigned

by the first three digits of the restaurant’s mailing ZIP code, against a fixed list of ZIP3 prefixes per metro. 21 metros are defined this way

What that gets wrong

a ZIP3 prefix is not a Census metropolitan statistical area. Some outer suburbs fall outside the prefixes listed for their metro and are counted only in the state; a prefix that straddles a metro boundary puts every restaurant in it on one side. A restaurant whose ZIP is missing or unmapped is counted nationally and by state, and in no metro at all

Why not use MSAs

because the store master carries a mailing address, not a Census geography, and inventing a crosswalk that is precise to the census tract from a mailing ZIP would add false precision, not remove it. The rule is stated plainly instead, so a reader can decide what it is good for

Currently publishing

7 metros and 18 states have at least one published figure. A state or metro with none has no page

National-only readings

a reading that has never cleared its threshold for any metro on any day is not a reading having a bad day — the sample is not there to support it locally, and it never will be without more restaurants. Its page says that in words, with the largest sample a metro actually reached, rather than printing a table whose every row reads “not enough data”. Currently Online tip rate is national-only. So are 10 of the 13 dish indexes. The test is made against the data on every page load, so a reading that starts clearing the threshold locally gets its table back with nothing edited

Every reading, defined

These are printed from the same catalogue the tiles, the charts and the JSON API read, so the definition here is the one under the number.

Hour buckets are recorded on the point-of-sale platform’s clock (US Central), not each restaurant’s local time; a page that prints an hour says so.

Every dish index’s matching rules

A price index is the median of what restaurants list a dish at on their own menus — the asking price, before any discount, tax, tip or delivery fee, and regardless of whether anyone ordered it. Items are matched by name with the regular expressions below, run against the lower-cased item name in the order shown; the first rule that matches wins. These are printed from the builder’s own rule table.

One exclusion runs before all of them: a kids’ or children’s menu item is a smaller, cheaper product than the dish it shares a name with, so it never counts toward any index — /\bkids?\b|\bkid'?s\b|\bchild(ren)?\b|\bjunior\b|\bjr\b/.

13 indexes. A dish is matched on the item’s name alone: the menus carry no structured dish taxonomy, so a restaurant that calls its pho “house special soup” is not in the pho index, and nothing pretends otherwise. Cuisine is classified from the restaurant’s own menu by the same classifier the consumer directory uses.

The calendar

The one part of this site that a person wrote rather than a job computed. It holds 109 entries on 90 days of the year, each carrying the primary page it was verified against — cdtfa.ca.gov, comptroller.texas.gov, dor.georgia.gov, floridarevenue.com, ipt.fifa.com, irs.gov, media.nfl.com, mlb.com, nationaldaycalendar.com and tax.ny.gov. The loader refuses any entry without a working source URL, so an unsourced date cannot be rendered even if someone adds one to the file. The remaining 276 days of the year are deliberately blank. The calendar.

Build schedule, revisions and history

Rebuilt

every night, after the menu index rebuilds. Last run 2026-09-23 19:22 Central.

Revisions

the last three days are recomputed every night, because an order can be completed after midnight. Days older than that are never rewritten

History

479 days with something published, from June 1, 2025 to September 22, 2026. Price indexes have history only from the day they were first computed: the menu index is a snapshot with no past, so backdating menu prices would be inventing them

Year-over-year

shown only where 365 days of published history actually exist for that metric and geography, and never against a thin day

Rounding

money to the cent, shares to one decimal place, ratios to one decimal place, counts to no decimal places. Rounding happens at display time; the stored value and the JSON API carry the full precision the computation produced

Taking these numbers

Free to cite with a link. Every reading has a JSON endpoint that returns published cells only, and an embeddable chart. Terms · About Kwick365 · For restaurants

• JSON https://kwick365.com/api/v1/reading/online-ticket/?geo=us
• Chart <iframe src="https://kwick365.com/embed/online-ticket/us/" width="560" height="360" style="border:0" title="Online order ticket — Kwick365"></iframe>
• Contact hello@kwick365.com