Read the terms as a purchase order and the cheque is the least interesting line. Eighteen billion dollars spread across ten years averages about $1.8bn a year [17], a figure a company of Meta's size can budget against well in advance; the part that keeps costing money is the age-assurance package, which pairs Meta's own tools with third-party ones, adds regular outside audits, sets explicit goals on false-positive rates, and requires the company to re-check the ages of the friends of any account removed as under 13 [5]. Arturo Bejar, the former Meta engineering director who was the first witness at the trial this settlement cut short [12], likes that section for exactly that reason: it has measurement, independent testing and goals, and in his words Meta should be measured on effectiveness rather than effort [11]. He had testified that the company previously ran a "don't ask, don't tell" approach to under-13 users, with no goals and no metrics for checking suspected ages despite world-class infrastructure for spotting fake accounts [15].
The behavioural terms bite harder than the two-hour figure sounds. The midnight to 6 a.m. block takes six of the twenty-four hours off the table [3], so the two-hour cap [2] fills about eleven percent of the eighteen hours that remain [18], and both settings arrive switched on, removable only with a parent's permission [4]. Adults can pick up individual versions of the controls, such as time-limit reminders, but there is no switch that puts an adult account into the teen configuration [14].
What Meta did not concede is delivery. Turning off video autoplay and viewing posts in chronological order stay opt-in [8], which is why Josh Golin of the nonprofit Fairplay said he was disappointed the deal does not default off the recommendation algorithms [9], and why Bejar's objection lands where it does: two hours of a product is still two hours of whatever that product delivers [10]. Like counts get hidden [7]. Ranking stays a company decision.
This is probably wrong, but the party most constrained by this document may not be Meta. A ten-year written specification, agreed with 48 attorneys general and scoped to the United States [1][6], reads to me as the floor any smaller youth-facing platform will be asked about, without the account base across which audited age inference can be amortised. The counter-thesis, which I hold loosely, is that the terms were drafted against Meta's particular infrastructure and bind nobody else until somebody sues again, and that the clauses left opt-in show precisely what the states could not win. It goes three ways from here: the audits land showing the false-positive goals met and the package becomes cheap boilerplate that rivals copy; the goals are missed and the measurement language turns into the real enforcement hook; or US teen hours fall far enough to show up in advertising, in which case the cap cost more than the $1.8bn a year did. That third one cannot be priced from what is on the record here, and pretending otherwise would be the satisfying answer rather than the honest one.
