# How Checkbox estimates risk

> How the "risk if skipped" level on each item is worked out from public enforcement records and your own subscriber counts, where the figures come from, and what they cannot tell you.

Every subscription business has to weigh two things: following the automatic-renewal laws, and staying in business.
Risk depends on your exposure; Checkbox shows yours. To help you weigh it, Checkbox shows a **risk if skipped** level next to each item you could choose to skip. The levels are
**very low**, **low**, **medium** and **high**, and each comes with a one-line reason. The level never changes what the law
requires. An item you skip is shown as "Risk accepted", never as passing.

> **Not legal advice.** The levels are Checkbox's estimates from public records and your subscriber counts. They are not
> legal advice, they do not guarantee anything, and no lawyer has reviewed them. Talk to a lawyer before you rely on them.

## What goes into a level

Each level combines two things: what skipping a requirement has cost other businesses, and how much of your business the
requirement reaches.

**What it has cost others** comes from Checkbox's legal catalog and its litigation research:

- **Who can enforce the law.** Some state laws are enforced only by the state attorney general. Others let subscribers
  sue, either under the renewal law itself or under the state's general consumer protection law. Some of those set fixed
  or multiplied damages per violation. Each law page lists these remedies with the statute
  ([state laws](https://checkbox.site/docs/learn/state-automatic-renewal-laws)).
- **Enforcement actually reported.** California has repeated demand letters, including to small brands (any public
  checkout can be checked remotely in minutes), as well as class actions and actions by its district attorneys' task
  force. The New York Attorney General has brought cases. The FTC enforces federal law (ROSCA): reported FTC cases
  involve large and mid-size companies, and civil penalties are counted per instance. An Illinois Consumer Fraud Act class action has
  settled. An Oregon class action (against YouTube) was dismissed because the terms were disclosed conspicuously. The
  District of Columbia joined the FTC's pending case against Uber. For Virginia, defense firms forecast suits under the
  2026 amendments, but no reported case was found yet. For the other laws, Checkbox's research did not find reported
  cases. That is all it means: it does not mean "safe".
- **How often this kind of failure is pursued.** The three most-pursued failures are renewal terms that are hidden or far
  from the button, cancellation friction, and missing consent. Reminder emails are a smaller category, but it is growing.

**How much of your business it reaches** comes from your exposure:

- **Live Stripe.** With live Stripe connected, Checkbox reads aggregate counts through the read-only connection. It reads
  active and trialing subscriptions, the share billed yearly, counts per billing state or country, and the dispute rate
  over the last 90 days (disputes divided by successful charges). The location is the customer's billing address,
  otherwise the default card's billing address or country. The counts are refreshed daily and when you connect Stripe.
  **Only counts are stored: never names, emails or addresses.** Test-mode data is never used, because it says nothing about
  real subscribers.
- **Your estimate.** Without live Stripe data, Checkbox asks you one optional question, once: "About how many subscribers
  do you have, and how many are in California?" The bands are fewer than 100, 100 to 1,000, 1,000 to 10,000, and more
  than 10,000. The California share is low (under about 8%), medium (8 to 20%) or high (over 20%). Skipping the question
  never blocks anything.
- **Nothing known.** The levels are then general estimates, and the plan says so.

## The rules

The model is deliberately small, so that every level can be explained in one line.

1. Nobody affected (none of your subscribers live where the law applies): **very low**.
2. Five or fewer subscribers affected, no active enforcement reported for that law, and no set or multiplied damages
   for private suits: **very low**.

Rules 1 and 2 apply only to exact counts. When Checkbox cannot place some subscribers by state (a card country with no
state, or no location at all), or it read only part of your subscriptions, a state's count is a lower bound: the line
says "at least N", and neither rule applies.
3. Otherwise, Checkbox adds up points for each jurisdiction the item covers, and the highest one counts:

| Factor | Points |
|---|---|
| Subscribers affected | under 10: 0 · 10 to 99: 1 · 100 to 999: 2 · 1,000 to 9,999: 3 · 10,000 or more: 4 (unknown: 1) |
| Who can enforce | public agencies only: 0 · subscribers can sue: 1 · subscribers can sue for set or multiplied damages: 2 |
| Enforcement reported | none found: 0 · documented: 1 · active, including against small businesses: 2 |
| Kind of failure | one of the three most-pursued (terms near the button, cancellation, consent): 1 · otherwise: 0 |
| Dispute rate | above 0.75% over 90 days, on at least 100 charges: 1 (fewer charges: no point, with a note that one or two disputes move a small rate a lot) |

A total of 0 to 2 is **very low**, 3 to 4 is **low**, 5 to 6 is **medium**, and 7 or more is **high**.

With only your estimate, federal law uses your subscriber band. California uses the band lowered by your California
share, and any other single state is assumed to hold a minority of your subscribers.

Two examples:

- **A small service with 2 subscribers in one state**, for a Your-call item only that state asks for: the 2 affected
  subscribers keep the score low unless that state's remedies or enforcement raise it.
- **A service with 20,000 subscribers, 40% in California**, for California's yearly reminder: the 8,000 California
  subscribers, private suits and active enforcement make it **medium**.

## Disputes and your Stripe account

For many small services, the larger practical risk is not a lawsuit but chargebacks. Stripe says that "the credit card
processing industry standard recognizes dispute activity above 0.75% as excessive", and card networks run monitoring
programs that can fine businesses with high dispute activity
([Stripe: Measuring disputes](https://docs.stripe.com/disputes/measuring)). Stripe may review accounts with high dispute
rates. It also recommends a clear refund and cancellation policy that customers agree to
([Stripe: Preventing disputes](https://docs.stripe.com/disputes/prevention/best-practices)). Below about 100 payments a
month, one or two disputes can move your rate a lot, so Checkbox counts a dispute rate as high only over at least 100
charges in 90 days. Clear renewal terms at sign-up, easy online cancellation and the required reminders help prevent
disputes. That is one reason these three items cannot be accepted as a risk.

## Sources

- Checkbox's litigation research, summarized from statutes, court orders, agency releases and settlement trackers. It is
  desk research and no lawyer has reviewed it. Highlights: California demand letters and task-force settlements (mostly
  $0.4M to $7.5M), New York Attorney General actions (1-800-Flowers, $375K), FTC actions under ROSCA (Amazon, $2.5B; Chegg,
  $7.5M), an Illinois class settlement (FloSports, $1.55M), the dismissed Oregon class action against YouTube, and the
  District of Columbia joining the FTC's case against Uber.
- The remedy provisions of each law in the catalog, quoted from the statute: [state laws](https://checkbox.site/docs/learn/state-automatic-renewal-laws),
  [ROSCA](https://checkbox.site/docs/learn/rosca), [California](https://checkbox.site/docs/learn/california-automatic-renewal-law),
  [New York](https://checkbox.site/docs/learn/new-york-automatic-renewal-law).
- Stripe's documentation on [measuring disputes](https://docs.stripe.com/disputes/measuring).

Found something wrong? Send it with `send_feedback` or from the dashboard, and it goes into the
[corrections log](https://checkbox.site/docs/corrections) once fixed.

## In the API

- `GET /v1/sites/{site}/exposure` returns the counts, `summary.line` (the sentence to show people) and, while nothing is
  known, `question`. The same object is `exposure` on the [compliance plan](https://checkbox.site/docs/guides/compliance-plan).
- `POST /v1/sites/{site}/exposure` records your estimate with `{ "subscribers": "100to1k", "california_share": "low" }`,
  or `{ "declined": true }`. Over MCP, use `call_api` with `operation_id: "updateSiteExposure"`. An estimate sent with an
  API key is shown as "estimate entered by your agent" (`estimate_by: "agent"`), and it never lowers a risk level below
  what no estimate gives.
