USA is not decoration here. It is the jurisdiction that grades the claim.

Start with the word every product in this category puts on its home page. Evidence-based is a defined term, and the definition is at 20 U.S.C. 7801(21)(A). It names three levels - strong evidence from an experimental study, moderate evidence from a quasi-experimental study, promising evidence from a correlational study with statistical controls for selection bias - each requiring a statistically significant effect on improving student outcomes. Then it adds a fourth test and gives it no name at all. The tier numbers everyone uses, and the phrase demonstrates a rationale, come from Department of Education guidance and from its grant regulations. The statute does not contain the word tier.¹ The unnamed fourth test is two requirements joined by the word and: a rationale based on high-quality research findings, and ongoing efforts to examine the effects of the intervention. A logic model on its own does not satisfy it. The Department's own regulatory definition at 34 C.F.R. 77.1 describes the logic model and the citations behind it and drops the ongoing-evaluation element, so a product can satisfy the grant regulation and still not meet the statute.²

Two more precision points, and both are where money changes hands. The first: a school improvement plan must include evidence-based interventions with no restriction on level, while money spent under section 1003 is restricted by statute to the top three. The restriction attaches to the funding, not to the plan, and the two are quoted as one rule constantly.³ The second: the thresholds nobody can recite correctly. A sample of at least 350, a multi-site requirement and a population overlap condition all exist, and none of them is in the statute. They are in the Department's grant regulations, they apply only to the strong and moderate levels, and the two levels are separated by a single conjunction - strong requires a sample overlapping the populations and settings proposed to receive the component, moderate requires populations or settings. The promising level has no sample floor, no multi-site requirement and no overlap clause anywhere.⁴ What does not exist at all is approval. There is no ESSA approval process, no certification, no federal registry of approved educational technology, and the evidence clearinghouse rates studies rather than blessing products. The Department's own current guidance says in terms that it "does not address the specific role of evidence in each program the Department administers," so it is not even the authority for which stream needs which level. The statute is.⁵

The second regime is children's privacy, and its rulebook contains a sentence about this product. Section 312.7 of the Children's Online Privacy Protection Rule prohibits an operator from conditioning a child's participation in a game, the offering of a prize, or another activity on the child disclosing more personal information than is reasonably necessary. Child means an individual under thirteen. The civil penalty is $53,088 per violation, a figure the Commission confirmed unchanged for 2026 four days before this page was written. The 2025 amendments took effect on June 23, 2025 with a compliance date of April 22, 2026, and both dates are now behind us. And the test for whether a service is directed to children reads like a game design brief: subject matter, visual content, use of animated characters or child-oriented activities and incentives, music, the age of models, and empirical evidence about who actually uses it.⁶ The arrangement that the entire classroom-software model runs on - a school consenting on a parent's behalf - has never been written into that rule. It appears in a preamble from 1999, in a frequently-asked-questions page, and in a policy statement from May 2022. The Commission proposed to codify it in January 2024 and then wrote, in the 2025 final rule, that it "is not finalizing the proposed amendments to the Rule related to ed tech and the role of schools at this time," because the Department of Education had signaled its own rulemaking on the education records regulations. Those regulations have not been amended either, and the separate route a vendor operates under there - the school official exception - has three conjunctive conditions, one of which requires the vendor to be under the direct control of the school with respect to the use and maintenance of education records.⁷

Two enforcement actions settle what that means in practice. In June 2023 a learning platform was charged both with violating the privacy rule and, separately, with an unfair practice under the Federal Trade Commission Act for "unfairly requiring Schools and teachers to comply with the COPPA Rule on its behalf without providing adequate information or support to meet the Rule's requirements." The civil penalty was six million dollars, entered and suspended. The Commission had already put it in plain words in 2022: the responsibility for compliance is on businesses, not schools or parents, and agreements must reflect that.⁸ And in January 2016 a company selling more than forty brain-training games was charged over four claimed benefits, the second of which was improving performance in school. A fifty million dollar judgment was entered; two million was actually paid and the remainder suspended on financial condition. The substantiation the order requires is not peer review - it is human clinical testing that is randomized, adequately controlled, and blinded to the maximum extent practicable. Clearing the statute's promising level with a single correlational study is no defense at all to that standard, because the two regimes do not align.⁹ The fourth regime is accessibility. Public entities including school districts must bring web content to WCAG 2.1 Level AA, and on April 20, 2026 the Department of Justice extended both compliance dates by a year, to April 26, 2027 for entities serving 50,000 or more and April 26, 2028 for everyone else. The obligation covers content a public entity makes available "directly or through contractual, licensing, or other arrangements," which is the clause that turns a rule about government into a condition in a purchase order. Section 508 is a different law, it binds federal agencies rather than districts, and it still incorporates WCAG 2.0.¹⁰

Now the evidence, which is thinner than the marketing. The federal evidence clearinghouse has published exactly one intervention report on a commercial product that is substantially a math game. It is from December 2013, it rests on a single study that met the standards without reservations, that study covered 557 students in three schools in one city, the rating was potentially positive effects, the extent of evidence was described as small, and the improvement index was four percentile points.¹¹ We read the full list of 557 named interventions on September 19, 2026 and found no intervention report for any of eleven widely used math products. That is a statement about what the list contains, not a finding about those products, and the page says so on the card.¹² The clearinghouse's own tier counts are worth seeing drawn: 224 at strong, 164 at moderate, 652 at promising, and two at the fourth level. Within mathematics specifically, 12 of 139 intervention reports carry a tier at all.¹³ And the strongest practice guide in the subject, the March 2021 guide for students struggling with mathematics in the elementary grades, makes six recommendations - systematic instruction, mathematical language, representations, number lines, word problems, timed activities - and not one of them is about games. Nor is any of the five in the 2013 guide for young children.¹⁴ The federal government does fund this category by name: the Department's small business research program has made 258 first-phase and 99 second-phase awards since 2002, up to $1.25 million across both, and games head its own list of what those awards have built. One documented line runs all the way through to a clearinghouse rating.¹⁵

Here is what argues hardest against the premise, printed on the page selling it. A congressionally mandated randomized evaluation, run for the Department across 33 districts, 132 schools, 439 teachers and 9,424 students, tested six mathematics software products across two cohorts and reported effect sizes from minus 0.01 to 0.06. None was statistically significant. Its own sentence: the evaluation "found no significant difference in student achievement between the classrooms that used the technology products and the classrooms that did not use the technology products, in any of the four groups, in either the first or second year of use." The one product in the whole study that did show a significant positive effect was in fourth-grade reading.¹⁶ The best-designed synthesis of digital games and learning reports a pooled effect of 0.33 across 57 studies and 209 effect sizes - and when its authors kept only the four studies that met every one of their study-design quality filters, those four pooled to 0.02 with a confidence interval running from minus 0.72 to 0.76. Their own caveat, which belongs beside the number, is that the result must be interpreted with extreme caution given how few studies remained. The same paper found no evidence of publication bias, which makes the collapse harder to dismiss rather than easier.¹⁷ And the commercial premise itself has a null behind it: the largest synthesis to test whether serious games are more motivating than conventional instruction found learning at d = 0.29 and retention at d = 0.36, both significant, and motivation at d = 0.26, which was not.¹⁸

One last thing, because it explains why every number a buyer will be shown in this category is unauditable. There is no federal industry classification for educational software or educational games. The code for software publishing covers every publisher in the country under one heading and its definition never uses the words educational, game, learning, school or children. Nothing sits beneath it. What can be said from federal sources is that United States software publishing recorded about $453.5 billion in revenue in 2022 across roughly 18,425 establishments and just over a million employees, and that no federal source measures what share of that is educational, or mathematics, or a game.¹⁹ Four regulators, one intervention report, a federal trial that found nothing, and no way to size the market from public data. Somebody has to read all of that before building here, and that is the business this name is for.