de-amplify_
← all distillations

distillation · the evidence ledger

MDL 3047: the federal social-media addiction multidistrict litigation

In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, MDL No. 3047, N.D. Cal., Judge Yvonne Gonzalez Rogers; defendants Meta, Google/Alphabet, ByteDance, Snap

This is the raw, evidence-tiered claim ledger behind the MDL 3047 case file. It is published for transparency: the source of truth, with its tiers intact, current as of 2026-08-28.

how to read the tiers

Each claim is tagged by the strength of the evidence behind it, not by how certain it sounds.

  • [ESTABLISHED]: stated in a primary source (a court order or filing, the docket, an official release) or, for a hearing, said on the official record. A fact about what was filed or said, not necessarily adjudicated true.
  • [OBSERVED]: carried by secondary coverage, or by a document read into the record.
  • [ASSUMED]: the distiller's inference, flagged as such.
  • Provenance flags ([interested-party], [advocacy-witness], [single-witness], [lawmaker characterization]) mark a claim that rests on one interested voice: reported, not adjudicated.

Tier key (this record's calibration, per the parent plan): ESTABLISHED means stated in a primary source (a court order or filing, the docket, an official government release) or in the project's verified baseline with independent corroboration; OBSERVED means carried by secondary coverage or inferred from it; ASSUMED means the drafter's inference, flagged as such. This authority-based calibration is the plan's deliberate adaptation for legal records, not the distillation skill's agreement-based multi-source tiers.

Reader summary

This is the big federal case bundling thousands of lawsuits that accuse Meta (Facebook, Instagram), Google/Alphabet (YouTube), ByteDance (TikTok), and Snap (Snapchat) of building their apps to be addictive to kids and teens. All of those cases are being managed together in one federal court in Northern California, in front of U.S. District Judge Yvonne Gonzalez Rogers, under the label "MDL 3047." As of August 3, 2026, 3,137 individual lawsuits were pending in that bundle. In an order filed June 29, 2026 (announced June 30), the judge refused to throw out the claims brought by a group of state attorneys general, and those claims went to trial. That trial opened in Oakland on August 18, 2026, against Meta alone, and it never reached a verdict. On August 26, 2026, four days into evidence, the parties filed a settlement they had negotiated and the judge entered it the same day as a consent judgment: a final judgment, signed, with both sides waiving appeal. The trial was vacated and the jury discharged.

What Meta agreed to has two halves. The money is up to about $17.2 billion over ten years across 51 states and territories, but roughly 30 percent of it is contingent: Meta only owes it if Snap, TikTok and YouTube become bound to the same kind of teen time limits and pay comparable amounts, and if that never happens the money reverts to Meta. The product changes run on the same logic. Teens get a default two-hour daily limit across Instagram and Facebook (the agreement's covered platforms; messaging and Meta's other products are excluded) and a default overnight block from midnight to 6 a.m., which a parent (not the teen) can loosen; the stronger version of the overnight block, 10 p.m. to 7 a.m., switches on only if the other three companies adopt equivalent rules. Teens also stop seeing like counts by default, lose access to cosmetic-surgery filters, and must be offered a non-personalized feed. An independent auditor reports publicly, but only for the first five years of the ten-year judgment.

Two things the settlement is not. It records no admission of liability, and it says in terms that nothing in it establishes a standard of care or serves as precedent anywhere outside the settling states. And it does not end MDL 3047: the personal-injury and school-district cases continue, against Meta and against Google, TikTok and Snap, who are not parties to this judgment. What it does end, beyond this case, is the states' own parallel lawsuits in their own courts: the agreement lists fifteen state-court cases it resolves, including Tennessee's Nashville trial, which was in its fifth week of evidence when the settlement stopped it. (The case background is current as of July 16, 2026 and the settlement as of August 28, 2026; litigation like this changes almost weekly.)

Note: This document covers ONLY the federal case, MDL 3047. Separate state-court cases (including a California state proceeding and a New Mexico case, both of which produced their own jury verdicts) are deliberately kept out to avoid mixing up the facts.

Claims (the ledger)

(a) Parties & court

  1. [ESTABLISHED] The case is formally titled In re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, MDL No. 3047, and appears on the federal Judicial Panel on Multidistrict Litigation (JPML) docket list. Source: JPML Pending MDL Dockets report, July 1, 2026, page 1.

  2. [ESTABLISHED] The MDL is assigned to U.S. District Judge Yvonne Gonzalez Rogers in the Northern District of California (district code "CAN"). Source: JPML report, July 1, 2026, page 1; corroborated by the California AG release and the CourtListener docket.

  3. [ESTABLISHED] As of the July 1, 2026 JPML report, MDL 3047 had 2,893 actions "now pending," with a historical total of 3,068 actions ever filed into the docket. The June 1, 2026 report showed 2,664 pending and 2,839 historical, so the count rose by exactly 229 in a month. Source: JPML reports, July 1 and June 1, 2026, page 1 of each.

  4. [ESTABLISHED] The lead consolidated state-AG case carries the individual number 4:23-cv-05448-YGR (People of the State of California v. Meta Platforms, Inc., filed October 24, 2023); the MDL master docket is 4:22-md-03047 (opened October 6, 2022). Source: CourtListener dockets for both numbers; the order caption carries both.

  5. [OBSERVED] The named corporate defendants across the MDL are Meta (Facebook/Instagram), Google/Alphabet (YouTube), ByteDance (TikTok), and Snap (Snapchat). Secondary coverage of the settled bellwether names the defendant set as "Meta, Snap, TikTok, and YouTube." Source: verified baseline; defendant set corroborated by Law Commentary and coverage of the Kentucky bellwether settlement.

  6. [OBSERVED] The state attorneys general are litigating as a coalition; coverage puts the group bringing federal children's-privacy (COPPA) claims to the August trial at 29 states, including California, New Jersey, Kentucky, and Colorado. (The official California AG release refers to "attorneys general nationwide" but does not state a number.) Source: Top Class Actions; JURIST; Engadget; Law.com headline ("29 States Head to Trial"); California AG release (no count). The four consumer-protection states are a subset of the 29; see Tensions (resolved).

  1. [ESTABLISHED] The core theory is product design: plaintiffs allege the companies intentionally built app features to keep young users coming back, and the June 29 order rests its denial of summary judgment on factual disputes over exactly that theory. Court coverage lists the specific features at issue as "infinite scroll, autoplay, notifications and other engagement-focused tools" (feature list as quoted in coverage of the order, not re-verified against the order text). Source: verified baseline; the order (Dkt 3214); the feature list per Top Class Actions' account of the order.

  2. [OBSERVED] A central factual fight is whether "social media addiction" is a real, definable condition. The judge found that "more recent research and statements from the American Psychiatric Association create a factual dispute regarding the existence and characteristics of social media addiction that must be resolved at trial." Meta argues social-media addiction is not an established psychiatric diagnosis; the American Psychiatric Association's position is that the absence of a DSM-5-TR listing does not mean the condition does not exist. Source: Top Class Actions (quoting the order); Engadget.

(c) The June 29, 2026 ruling (announced June 30)

  1. [ESTABLISHED] In an order filed June 29, 2026 (a Monday; announced and widely covered June 30), the court denied Meta's motion for summary judgment as to the state attorneys general's claims, meaning those claims survive and proceed toward trial rather than being thrown out. The California AG describes the motion as "fully denied." Source: the order itself, Dkt 3214 (AG-case Dkt 440), docket stamp "Filed 06/29/26"; Reuters ("decision late Monday") via Claims Journal; California AG press release (dated June 30).

  2. [ESTABLISHED] The denial rested on material factual disputes, including whether the platforms are addictive and whether Meta misrepresented that, which a court cannot resolve on summary judgment and must send to trial. Source: the order text ("material disputes of fact exist"); verified baseline; Top Class Actions' account.

  3. [ESTABLISHED] On one narrower COPPA point, the court gave the states a partial win: it found there is "no factual dispute that Meta did not comply with the statute's notice and parental consent requirements." The California AG frames this as the court agreeing that Meta "did not obtain parental consent in a manner sufficient to satisfy" COPPA. Source: the order, pp. 1-2 ("no dispute exists that Meta has not, in fact, complied"; "summary judgment is GRANTED" on that element); California AG press release.

  4. [ESTABLISHED] This COPPA point is a compliance finding, not a finding of liability. Whether COPPA even applies to the services (questions like child-directed status and the company's knowledge of under-13 users) remains disputed and is left for trial. Source: the order, pp. 1-2 ("material disputes of fact exist as to whether Meta is required to comply"; "a material dispute of fact exists as to whether Meta has actual knowledge").

  5. [ESTABLISHED] In the same order, the court found that Meta's own documents could support the plaintiffs' theory that Meta's time-restriction (screen-time-limit) tools were a "public relations stunt," deployed while the company allegedly knew more time on the platform was linked to worse outcomes for teens. This is cited evidence that could support the theory, still disputed, not a finding of fact. Source: the order, Dkt 3214, p. 23, verified in the RECAP PDF on 2026-07-16. (Audit trail: this claim was originally tiered OBSERVED because its only known carrier, a Law.com article, returned HTTP 403; the 2026-07-16 audit verified the phrase in the order itself. The original flag was an access failure, not a disconfirmation.)

(d) The August trial + follow-on

Read with section (k): the August trial ended on 2026-08-26 in an entered consent judgment. The claims below record what it was and how it was structured, which the settlement does not change; whether the February 2027 follow-on survives the settlement is an open question in Tensions.

  1. [ESTABLISHED, and superseded on tense by section (k)] The first trial was held in Oakland, California, and ended on 2026-08-26 in a consent judgment rather than a verdict. Jury selection began August 12, 2026; opening statements were delivered on August 18, 2026, and evidence began the same day. Source: California AG press release for the scheduled dates; the court's own Civil Minutes for 2026-08-18 (Dkt 550) for the fact that trial was held, which record a session of 5 hours 17 minutes before Judge Gonzalez Rogers, continued to August 19.

  2. [OBSERVED] The August trial combines two sets of claims: the 29 states' consolidated federal COPPA claims and consumer-protection claims from four of those states: California, Colorado, Kentucky, and New Jersey. The defendant at this trial is Meta alone (Meta Platforms, Inc. and Instagram, LLC); Google, TikTok, and Snap are MDL defendants in the separate personal-injury and school-district tracks, not in the AG trial. Source: JURIST; Engadget; Law Commentary; Top Class Actions; the Meta-only point per the order caption (California et al. v. Meta Platforms, Inc.) and the trial-motion filers on the docket (primary).

  3. [ESTABLISHED] The trial is structured to use an advisory jury. Meta had demanded a jury, then moved to withdraw that demand (motion filed March 3, 2026, Dkt 2807, heard April 15, 2026); the judge agreed and decided to empanel an advisory jury to hear the evidence and give its view while she retains the ultimate findings. She is expected to decide the federal COPPA claims herself and use the advisory jury chiefly on the state consumer-protection claims. The advisory jury's exact scope (for example, on civil penalties) was still disputed between the parties as of mid-July. Source: MDL docket (Dkt 2807); the joint pretrial statement, Dkt 470 ("the advisory jury that the Court has decided to empanel"); Law Commentary; Law360/MLex headlines. (The granting ruling appears to be a minute order not available in RECAP full text.)

  4. [OBSERVED] A second trial for 14 additional states' own state-law claims is scheduled for February 2027. Do not conflate it with a different February 2027 event in the same MDL: the first school-district bellwether (a Tucson case) is also expected then, per Meta's own filing. Source: Engadget; Reuters via Insurance Journal; Meta's trial-setting opposition, Dkt 3259 (the Tucson point). (The scheduling order itself was not located.)

17a. [OBSERVED, and it explains an arithmetic this record has carried unexplained since July] Four states left the multistate coalition in January 2025: Michigan, Georgia, Missouri and North Dakota. That is what takes the 33 joint federal filers of October 2023 to the 29 at the August 2026 trial, a drop this ledger previously recorded only as what coverage reported. Michigan's Attorney General's office attributes the withdrawal to discovery burden, saying the court "approved most of Meta's voluminous discovery demands upon the States and their agencies (such as MDHHS), demands with which the State of Michigan could not possibly comply, and especially so within the very short timeframe so ordered by the Court", and a filing estimated the narrowed request at roughly 700,000 documents. Source: Michigan Public, 2026-08-12, quoting the Attorney General's office (secondary, and the only account located). Corroborated structurally by the primary: the Ninth Circuit's caption in No. 24-7032 omits Michigan and Missouri entirely and lists North Dakota and Georgia as plain "Plaintiffs", separated from the "Plaintiffs - Appellees" block, which is the pattern four pre-appeal withdrawals would leave. Tiered OBSERVED rather than ESTABLISHED because the withdrawal filings themselves were not located; see Tensions.

  1. [OBSERVED] Counting all venues together, more than 40 state attorneys general have brought claims in the overall litigation: 33 states filed the joint federal complaint on October 24, 2023, Florida filed its own federal suit the same day (the source of "34" in some counts), and additional attorneys general filed in state courts, for roughly 42 total. The August-2026 and February-2027 federal trials divide claims, not states: since 29 plus 14 exceeds the federal-filer count, the 14 necessarily overlap the 29. Source: California AG October 2023 release ("bipartisan coalition of 33"); JURIST ("more than 40 state AGs"); NBC/TechCrunch (October 2023 counts). (No accessible source enumerates the exact 29 or 14 rosters; flagged in Tensions.)

(e) Money / exposure

  1. [ESTABLISHED] In its own court filing, Meta characterized the four states' potential penalty exposure as up to about $1.4 trillion ("the AGs' unsupported $1.4 trillion-dollar claim"). Source: Meta's penalty submission, Dkt 455, filed July 6, 2026 (RECAP, fetched 2026-07-16); corroborated by JURIST, Engadget, Law Commentary.

  2. [ESTABLISHED] The $1.4 trillion figure comes from multiplying a very large count of qualifying young users by the maximum per-violation statutory penalty under each applicable law. Meta's filing describes the states' method as counting penalties for "every single month in which a teen uses Meta's platforms for more than a half-hour", and names the exhibit it comes from: the AGs' Remedy Chart 2, "Time Spent Penalty Calculation for Teen Users", which Meta says "shows the number of teen users with monthly instances over 0.5 hours". Source: Dkt 455 (primary, cached 2026-08-18); JURIST and Law Commentary for the method framing. (Corrected 2026-08-18, on the first run after Dkt 455 was cached: this claim previously gave the exhibit's title as Monthly Instances of Teen Users Exceeding ... Over 0.5 Hours, which is a rendering from the docket listing and is not what the filing calls it. It is set as a literal here rather than as a quotation, because it is not a quotation of anything: that is the whole finding. Same class as claim 37: a docket caption is not text in a document.)

  3. [ESTABLISHED] Meta's filing argues the demanded remedies are "entirely unmoored" from the claimed deceptive statements or unfair practices, and says of the resulting number: "A sanction of that size has no analog in the history of consumer protection enforcement." (Capitalisation corrected 2026-08-18: the ledger had lowercased the filing's sentence-initial A to fit its own sentence, the same defect as the EU ledger's "Can be easily dismissed" and equally invisible by eye.) Meta also contends each state-law penalty must be tied to a separate, affirmative act of wrongdoing and raises due-process concerns. Source: Dkt 455, pp. 1-2 ("entirely unmoored") and p. 12 ("no analog"), verified in the RECAP PDF on 2026-07-16; JURIST (the separate-act and due-process arguments). See Discrepancies: the phrase "extraordinary and unjustified," previously quoted here as Meta's words, is an outlet's unquoted paraphrase and appears nowhere in the filing.

  4. [ASSUMED] The $1.4 trillion figure is a theoretical statutory ceiling, the largest number arithmetically possible under the penalty formula, not a demand any court has awarded or is likely to award in full. This is the drafter's characterization of how statutory-maximum penalty figures work, offered for reader context; the exact number the states will actually seek at trial, and any amount a court might award, are not established in the record reviewed here.

(e.1) What the states' own filing says (anchored to Dkt 473, which is cached)

The claims above quote Meta's filing (Dkt 455); the three below quote the states' reply (Dkt 473). Both are cached and the whole of section (e) is machine-verified, so this subsection is no longer doing the guard work it was created for. It stays because the split it marks is the one the $200 billion correction turns on: what Meta says about the states' ask is not what the states' own filing says. (This intro originally read that Dkt 455 was not cached, which was true for the hours between the subsection's creation and the cache landing later on 2026-08-18, and then sat false on this page: a note that explains a workaround needs re-reading when the workaround is removed.)

22a. [ESTABLISHED, and it corrects the press] The states' own submission does not state a monetary demand. In their Reply Regarding Penalty and Disgorgement Charts (Dkt 473, filed 2026-07-13) the AGs say they "will present their final requests for monetary relief at trial", selecting remedies supported by the proof, and that they "do not seek double recovery". As of that filing no total had been stated. Source: Dkt 473 (primary, fetched 2026-08-18).

22b. [ESTABLISHED] The monetary ask has two components, not one: civil penalties and disgorgement of ill-gotten gains. The AGs state that although multiple claims would entitle them to disgorgement, they "only seek to disgorge ill-gotten gains once". Most coverage of this trial says "damages", which is a third thing. Source: Dkt 473.

22c. [ESTABLISHED, and this is a trap] In Dkt 473 the figure $200 billion is Meta's 2025 revenue, cited to Meta's own Form 10-K at 61 (Jan. 28, 2026), and used to argue the scale at which deterrence has to operate. It is not a demand. Multiple outlets covering the opening reported roughly $200 billion as the amount the states seek. Source: Dkt 473. See the Tensions entry for what remains unresolved.

(e.2) What the attorneys general said in public on the day, and how it was obtained

Tier note before the claims, because it is doing real work. The four claims below rest on a machine transcription, cached as sources/ag-press-conference-2026-08-18-asr.txt and quarantined from npm run check:quotes by filename. They are tiered OBSERVED and they carry no quoted span, deliberately: nothing here is set in quotation marks and attributed to a named person, because the text in hand is a machine's rendering of an acoustic signal rather than the words the speaker used. That is the same treatment the Dutch judgments get in bits-of-freedom-v-meta, for the same reason. Where these claims touch the money question they agree with Dkt 473, which is primary and cached, so their function is corroboration from the plaintiffs' own mouths rather than new authority.

Three transcription systems sharing no pipeline (AssemblyAI on the PBS audio, PBS's own uploaded track, YouTube's automatic captions on the Fox News recording) render the load-bearing sentences the same way, with per-word confidence above 0.99 on every figure. A single pass would not have been enough to move the Tension below.

22d. [OBSERVED] At a press conference held after the first day of trial on 2026-08-18, a Reuters reporter put the discrepancy directly to the attorneys general at 32:24: both $1.4 trillion and $200 billion had been circulating, why the wide range, and did the states have a number in mind. The confusion this record has been documenting was live in the press room on day one, asked of the plaintiffs by a reporter, which is itself evidence that the coverage had not settled it. Source: press-conference ASR (CAPTION-ASR, cached), PBS NewsHour recording at 32:24.

22e. [OBSERVED, and it answers the coverage] California's attorney general replied that the $1.4 trillion figure is Meta's, that Meta put it out to the press as the highest number arithmetically derivable from data both parties had submitted, and that it is not something the states asked for or are seeking. He said twice, in plain terms, that the states are not asking for $1.4 trillion, once at 33:25 and again at 34:34. Asked at 34:38 whether the states had produced the figure, he said it is Meta's estimate, built from a count of young users multiplied by the penalties available under the unfair-competition laws at issue. Source: press-conference ASR at 32:34, 33:25, 34:34, 34:38.

22f. [OBSERVED] He described the ask as civil penalties, restitution and disgorgement, and explicitly not damages, saying so more than once in the same answer, and said the states leave the final amount to the judge, who has vast discretion. This independently corroborates claim 22b, which reached the two-component structure from the states' own filing, and it corrects the word most coverage uses. Source: press-conference ASR at 33:12 to 33:25.

22g. [OBSERVED, and it closes the trap in claim 22c] He described $200 billion as Meta's revenue per year, in a passage arguing that Meta is not the party being financially imposed upon. That is the same reading of that figure the states' own filing gives at Dkt 473, reached independently: $200 billion is the defendant's revenue, not the plaintiffs' demand, now stated by the plaintiffs in public as well as in a filing. Source: press-conference ASR at 34:14; compare claim 22c (Dkt 473, primary).

22h. [ESTABLISHED, and it is a fact about sourcing rather than about this case] The office that held the press conference published a truncated recording of it. The California Department of Justice channel carries 23:26, ending on an invitation to take questions, with the entire Q&A absent; third-party recordings of the same event run to 52:20, and the exchange in claims 22d to 22g falls in the 29 minutes the official version omits. An official channel is not automatically the fullest record of an official event. Source: durations of the five recordings, verified 2026-08-20 and listed in the cache header.

(f) Settlements

  1. [OBSERVED] All four defendants settled the first federal school-district bellwether (brought by the Breathitt County School District in Kentucky) before it reached a jury, in May 2026: Meta settled May 21, 2026, a few weeks before a planned June trial, and the co-defendants settled earlier the same month. The settlement agreements, obtained by Reuters through a public-records request (the plaintiff is a public entity), total about $27 million: Meta $9 million, Snap $8 million, ByteDance $8 million, Alphabet $2.01 million. The terms were not disclosed in court, but the amounts are documented from the agreements themselves, not estimates. Source: Reuters via Yahoo Finance; WKYT; Law Commentary; verified baseline (corrected: the baseline and an earlier version of this document called it a "personal-injury" bellwether; it was the first school-district bellwether).

  2. [OBSERVED] The bellwether settlement involved no admission of liability and produced no publicly disclosed, enforceable design standard; the disclosed agreement terms are monetary. Source: verified baseline; Reuters via Yahoo Finance (terms as obtained).

(g) Status as recorded on 2026-07-16 (superseded on the trial question by section (i))

  1. [ESTABLISHED, superseded on outcome by section (k)] The state attorneys general's claims survived summary judgment (order filed June 29, 2026) and went to trial: the first trial (29-state COPPA plus 4-state consumer protection, against Meta alone) began with jury selection on August 12, 2026 and opening statements on August 18, 2026, in Oakland. The MDL's final pretrial conference was set for July 17, 2026. Source: the order (Dkt 3214); California AG release (dates); Dkt 470 (pretrial conference); section (i) for the court's own record of the opening day. (Tense corrected 2026-08-18: the trial-flip pass corrected claim 14 and left this one, so the published ledger asserted both that openings had been delivered and that the case was headed to trial.)

  2. [OBSERVED, and it survives the settlement] No defendant has been found liable in MDL 3047 as of this document's date. The states' claims against Meta were resolved by a consent judgment that records no admission of liability (claim 54), not by a finding; liability on the remaining tracks is still undecided. Source: reasonable synthesis of the June 29 ruling (denial of summary judgment leaves liability for trial) and the compliance-not-liability nature of the COPPA finding.

(h) Pretrial rulings and the trial schedule, confirmed on the docket (added 2026-07-25)

  1. [ESTABLISHED] Meta asked the court to stay the August trial and was refused. Pretrial Order No. 6, ECF 3284, filed July 20, 2026, states at paragraph 10: "Meta's request to stay the trial in this case is DENIED." (Docket PDF fetched and read 2026-07-25: https://storage.courtlistener.com/recap/gov.uscourts.cand.401490/gov.uscourts.cand.401490.3284.0.pdf.)

  2. [ESTABLISHED] The trial dates this record has been carrying from secondary coverage are now confirmed by the court's own orders. Pretrial Order No. 3 (ECF 3212, filed June 29, 2026): "openings and evidence shall commence on Tuesday, August 18, 2026." Case Management Order No. 36 (ECF 3295, filed July 23, 2026) captions the proceeding with a stacked "Bellwether Trials:" heading over the two dates August 12, 2026 and February 3, 2027 (the caption is a block, not a sentence, so this record describes it rather than quoting a slash-joined reconstruction of it, which is what it did until 2026-08-28), which also confirms the February 2027 follow-on. Jury mechanics are on schedule: juror questionnaires "likely to be complete on August 4, 2026" and hardship stipulations due noon August 9, 2026. (Docket, fetched 2026-07-25.)

  3. [ESTABLISHED] The advisory-jury structure is unchanged and still expressly contested, which vindicates this record's earlier hedge rather than resolving it. The Joint Trial Stipulations (ECF 3258, July 13, granted by PTO 6 paragraph 12) record that Meta "does not waive, and expressly preserves, any objection to the submission of remedies-related issues to an advisory jury or any other argument concerning the proper role of the advisory jury with respect to remedies." (Docket, fetched 2026-07-25.)

  4. [ESTABLISHED] Further pretrial rulings this record did not previously carry: Pretrial Order No. 4 (ECF 3239, July 6) denied seven motions in limine and granted one in part; Pretrial Order No. 5 (ECF 3244, July 10); and ECF 3269 (July 16) denied Meta's motion to exclude Carl Saba while granting in part and denying in part the plaintiffs' motion to exclude Justin McCrary. The final pretrial conference was held July 17, 2026. The docket is current through July 24, 2026 (latest entry ECF 3299). (Docket, fetched 2026-07-25.)

  5. [ESTABLISHED] The MDL's defendant set is expanding past the original four, on paper. Los Angeles Unified School District v. Meta et al., No. 4:26-cv-02463 (N.D. Cal., filed 2026-03-20), names X Corp., Discord Inc., and Roblox Corporation alongside Meta, Google/YouTube, ByteDance/TikTok and Snap. CMO 36 (July 23, 2026) orders counsel to meet and confer and file a joint status update by August 17, 2026 addressing "the number of cases against X Corp., Discord, and Roblox" and "how to handle these cases moving forward." (Docket, fetched 2026-07-25.) That status report was filed and is now read: see claim 31a.

31a. [ESTABLISHED, and it answers claim 31's open question in a direction the site should not overstate] The Joint Status Report Pursuant to CMO 36 (MDL ECF 3407, filed 2026-08-17) gives the numbers. Discord is "presently named in 23 constituent cases brought by four law firms" (twelve personal-injury and eleven local-government or school-district cases); Roblox in 13 (five personal-injury, eight local-government or school-district), to be represented across the MDL by Mayer Brown; and X in 6, "brought by two law firms", plus one further action naming Twitter. But the three are barely in this litigation, and the report says so plainly. Discord and Roblox "were not named as a party to the Master Complaint in the MDL, and discovery has proceeded without Discord's and Roblox's participation"; they "do not have access to any materials on MDL Centrality and have received none of the discovery materials in these cases." Their cases have been under stipulated stays since an order of 2024-11-18, extended by further stipulations on 2026-04-14 and 2026-07-02. X, for its part, argues that "Maintaining claims against X in this MDL would not promote efficiency", noting that it "first became a defendant in March 2026" and "has not participated in common discovery or dispositive-motion briefing, and is not a defendant in any bellwether", and moved to sever and transfer on a forum-selection clause. (The filing joins those clauses with em-dashes; this record quotes the two fragments rather than silently replacing the dashes with a comma to fit the house style, which is what an earlier draft of this claim did.) The court's own words, quoted in the report from the July 17, 2026 transcript, are the most useful sentence in it: "X has never been a part of this case. I don't understand why it shouldn't be severed…. [T]his MDL has not involved X even if the allegations are similar." Source: MDL ECF 3407 (primary, cached 2026-08-28). So the honest reading is that three new names appear on complaints while none of the three is in a bellwether, two are stayed and outside discovery entirely, and the judge has questioned on the record whether the third belongs. The site said this was a thing to watch; what it should now say is what the watching found.

  1. [ESTABLISHED, and superseded by claim 32a] The JPML count this record cited was 2,893 actions pending / 3,068 total historical as of the July 1, 2026 report, verified in the PDF itself. (Checked 2026-07-25.) The sentence that stood here until 2026-08-28, that no August report existed, is corrected below; it was wrong, and the way it was wrong is the point.

32a. [ESTABLISHED, correcting claim 32 on 2026-08-28] The current count is 3,137 actions pending / 3,312 total historical, as of the August 3, 2026 JPML report, a rise of exactly 244 pending in the month since July 1. Source: Pending MDL Dockets by Actions Pending, report date 8/3/2026, page 1 (PRIMARY; fetched and cached 2026-08-28).

How the earlier claim went wrong, because it is a reusable trap, and how the first explanation of it was also wrong. This record asserted on 2026-07-25, and again on 2026-08-28, that no August report existed, on the strength of an HTTP 404 from a guessed filename: the June and July files are dated the 1st, so August-1-2026 was tried. JPML dated the August report August 3 (the 1st was a Saturday), so the guess could not have hit. Worse, the 2026-08-28 re-probe reported 404 for June and July as well, which should have stopped the conclusion immediately: the positive control failed and the absence was believed anyway.

For a few hours on 2026-08-28 this claim blamed two mechanisms, that JPML rotates old reports off the site and that the host requires a Referer header. Both were wrong, and both were diagnosed from the same broken query they were meant to explain. The real mechanism, isolated the same day by holding one variable at a time: the JPML host returns HTTP 404 to requests whose User-Agent is the bare token Mozilla/5.0, and 200 to a full browser string or to no User-Agent at all, reproducibly across three trials. The failing probes had all used the bare token; the succeeding ones happened to use a full string, and the Referer header sent alongside got the credit. Nothing was rotated: the June and July URLs serve their PDFs today, their MDL 3047 rows re-verify against claims 3 and 32 (2,664/2,839 and 2,893/3,068), and both are now cached in sources/ so this ledger's JPML lineage no longer cites documents it does not hold. A 404 for a bot-scented user-agent is a nastier trap than a 403, because a 403 says you were refused and a 404 says the thing does not exist, which is exactly the false absence this repo's search discipline exists to catch. The operational rules that survive: read the index at /pending-mdls-0 for the current filename instead of guessing, and send a full user-agent string when probing court infrastructure. Isolation matrix run 2026-08-28: three trials of bare-token vs full-string vs no user-agent, 404/200/200 on every trial.

(i) The trial itself, day one

The rest of the trial, and how it ended, are in section (k).

  1. [ESTABLISHED] Trial was held on 2026-08-18 before Judge Gonzalez Rogers, a session of 5 hours 17 minutes, and was continued to August 19. The court's minutes caption the matter State Attorneys General v. Meta Platforms, Inc. Source: Civil Minutes, Dkt 550 (primary, fetched 2026-08-18).

  2. [ESTABLISHED] The states' first witness was Arturo Bejar, a former Meta safety engineer. The court's minutes record that he was the witness called by the plaintiff on the opening day. Source: Dkt 550. Note that this is the same witness whose November 2023 Senate testimony this project already distils separately; the site should not conflate the two appearances.

  3. [ESTABLISHED] Counsel of record on the opening day included Megan O'Neill for the plaintiff states and Paul Schmidt for Meta. Source: Dkt 550. This settles a spelling: one outlet in the 2026-08-18 coverage rendered the deputy attorney general's first name with an h. The court's own minutes do not.

  4. [ESTABLISHED] A Stipulation and Updated Order Governing Trial Protocol was entered on 2026-08-18, following the Court's guidance at a pre-trial hearing on 2026-08-13 and in Pretrial Order No. 8 (filed 2026-08-16). It governs "the first Multistate Attorneys General (AGs) trial starting in August 2026". The word "first" is the court's. Source: Dkt 549 and Dkt 534 (primary, both fetched 2026-08-18).

  5. [ASSUMED, deliberately not upgraded] Party filings in this case use phase language: motions in limine in June were captioned as directed at an advisory jury phase of trial, and Meta filed a supplemental witness list for a remedies phase on 2026-08-17. Those phrases are docket-entry captions written by the filing parties, not text in any document cached here, which is why they are described rather than quoted. No court order reached in the 2026-08-18 pass establishes a bifurcated trial structure. Neither the Trial Protocol Order nor Pretrial Order No. 8 mentions phases or bifurcation; Pretrial Order No. 8 uses "bifurcated" only of the exchange of witness binders. Recorded as unresolved rather than published as structure. A party's caption is not a holding.

(j) The appellate track: the Ninth Circuit, 2026-08-10

Located 2026-08-18. This record had carried an appellate disposition as a single-sourced lead from one newspaper for weeks, described there as concerning Meta's First Amendment defence. The opinion is on the trial docket at entry 541, it is published, and it is not about the First Amendment.

  1. [ESTABLISHED] On 2026-08-10 the Ninth Circuit issued a published opinion in People of the State of California v. Meta Platforms, Inc., No. 24-7032, and dismissed Meta's and TikTok's appeals, along with the plaintiffs' conditional cross-appeals, for lack of appellate jurisdiction. Argued and submitted 2026-01-06 in San Francisco; panel of Circuit Judges Nguyen and Bennett with District Judge Matsumoto sitting by designation; opinion by Judge Nguyen. The disposition line is one word: "DISMISSED." Source: the opinion (primary, cached 2026-08-18).

  2. [ESTABLISHED, and it is the holding that matters] The court held that Section 230 provides "a defense to liability, not immunity from suit", so an order denying that defence "can be effectively reviewed on appeal from a final judgment" and fails the third Cohen factor. It rejected Meta's argument that immunity from suit should be implied where the statute does not provide it. Source: the opinion's summary and part II. This is a merits holding about what Section 230 is, delivered in a case the court declined to hear on the merits, and being published it binds the circuit.

  3. [ESTABLISHED, and it corrects a lead this project was carrying] The opinion records that "Meta does not identify any constitutional interests at stake", which is why the panel reasoned that any source of immunity from suit had to be Section 230 itself. A characterisation of this disposition as a First Amendment ruling is not supported by the opinion. Whether a separate appellate ruling on Meta's First Amendment arguments exists was not established in the 2026-08-18 pass; what is established is that this one is not it. Source: the opinion at part II.1.

  4. [ESTABLISHED] The trial survived a second stay attempt, this one at the appellate level. A footnote records that "Meta's emergency motion to stay trial pending resolution of this appeal (docket entry no. 190 in case no. 24-7032) is denied as moot". This is distinct from the district court's refusal in Pretrial Order No. 6 on 2026-07-20, which this record already carries at claim 27. Source: the opinion, footnote 4.

  5. [ESTABLISHED] The other two Cohen factors also failed. The district court's ruling did not conclusively determine the question, because that court "indicated its willingness to revisit the issue of section 230 immunity at a later stage of the proceedings"; and it did not resolve an issue separate from the merits, because deciding Section 230 immunity "necessarily involved reviewing the content and context of the factual allegations in the plaintiffs' complaint". Source: the opinion's summary.

  6. [ESTABLISHED, and it is a different appeal in the same courthouse] On 2026-08-17, the day before this trial opened, the Ninth Circuit denied rehearing en banc in Jane Doe 1 and Jane Doe 2 v. Meta Platforms, Inc., No. 24-1672, on appeal from D.C. No. 4:22-cv-00051-YGR, the Northern District of California at Oakland: the same district and the same district judge as this MDL. The vote is the substance. "Judge R. Nelson votes to deny the petition for rehearing en banc, and Judges W. Fletcher and Berzon recommend granting the petition for panel rehearing en banc." The order then records that "The full court has been advised of the petition and no judge has requested a vote on whether to rehear this matter en banc", so the petition failed for want of a call for a vote rather than on a contested one. Source: the order (primary, cached 2026-08-20). Why it sits in this ledger and not only in the policy paper: claims 39 to 42 establish what Section 230 is in this circuit (a defence to liability, not immunity from suit) in the appeal arising from this MDL, and Doe 1 is the circuit's holding on what Section 230 covers (that recommending content is publishing conduct). They are two halves of the Section 230 law this trial is being conducted under, decided a week apart, and a reader of this record who met only one of them would have half the picture.

Added 2026-08-28. The trial recorded as under way in sections (h) and (i) ended in its second week, not in a verdict but in a judgment the parties wrote themselves and the court entered the same day. Every claim below comes from a document on the docket, cached in sources/ and machine-verified. The money in particular is taken from the agreement's own Exhibit B rather than from any of the five different totals that ran in coverage, which is the point of this section.

  1. [ESTABLISHED] On 2026-08-26 the parties filed a Joint Motion to Enter Consent Judgment (Dkt 572), attaching the fully executed settlement agreement as Exhibit 1 (Dkt 572-1, 130 pages). The court took it up the same day in a session of 49 minutes: "The Court met with the parties regarding the Joint Motion to Enter Consent Judgment. Jury Trial is suspended." Source: Civil Minutes for 2026-08-26 (AG-case Dkt 574, MDL ECF 3449) and the motion itself (primary, cached 2026-08-28).

  2. [ESTABLISHED, and this is the posture that matters] The court granted the motion and entered the consent judgment the same day. Trial Order No. 3 (Dkt 575, MDL ECF 3450) states: "Upon filing, judgment shall be entered, the trial is deemed vacated, and the jury shall be advised that they are each discharged immediately." The judgment itself (Dkt 576, MDL ECF 3451) is signed by Chief Judge Yvonne Gonzalez Rogers and dated August 26, 2026, finds "there is no just reason for delay", enters "as a final judgment under Fed. R. Civ. P. 54 and 58", and directs that "The Clerk is ordered to enter this Judgment forthwith." Source: Dkt 575 and Dkt 576 (primary, both cached 2026-08-28).

45a. [ESTABLISHED, and it is a correction to the official channel rather than to the press] The California Attorney General's release of the same day describes a proposed settlement that "remains subject to court approval through entry of a consent judgment". That was accurate at filing and was overtaken within hours. A press release describes the posture at the moment it was written, and an entered judgment outranks it. The site must state the posture from Dkt 576, not from the release. Source: the release (official, fetched 2026-08-28) against Dkt 576. This is the same class as the June 29 / June 30 order-date correction in Discrepancies, with the roles reversed: there the primary was later than coverage, here the primary is later than the official announcement.

  1. [ESTABLISHED] The settlement resolves the claims of 51 states, commonwealths and territories, which is the row count of Exhibit B and matches the "bipartisan coalition of 51 attorneys general" in the California release. That is a wider group than the 29 states whose claims were being tried: Exhibit B carries a payment line for states that were not plaintiffs at this trial. The release's own enumeration reconciles exactly: five announcing attorneys general (California, Colorado, Tennessee, Kentucky, New Jersey) plus 46 named as joining. Three absences from Exhibit B are worth recording: Florida (which filed its own federal case in October 2023), Texas, and New Mexico, whose separate addiction case already went to final judgment on 2026-08-06 and is therefore consistent with its absence here. Source: Dkt 572-1, Exhibit B; California AG release (both cached).

46a. [ESTABLISHED, and it closes a loop this ledger opened at claim 17a] The four states that left the coalition in January 2025 are settling states. Michigan, Georgia, Missouri and North Dakota, whose withdrawal over discovery burden explains the drop from 33 joint filers to the 29 at trial, each carry a payment row in Exhibit B and each appear in the California release's roster of joining attorneys general. So the four that could not afford to prosecute the case are paid out of its resolution. Source: Dkt 572-1, Exhibit B; the California AG release of 2026-08-26 (both cached 2026-08-28). Whether that was negotiated for them or follows automatically from the release's scope is not established here.

  1. [ESTABLISHED, and it settles a number that ran five different ways in coverage] Exhibit B states the totals directly. Each figure in its state rows is per installment, over ten installments, which the exhibit's own footnote says in terms: "Figures are the payment amounts per installment, not the total amounts over the full payment term." The totals row reads $1,165,662,174.56 in Guaranteed Installment Payments, $502,402,600.77 in Contingency Installment Payments, and $16,680,647,753.21 as the maximum across all installments. Two further sums sit outside that table: a $75 million Cost Fund Payment for state costs, and a Cambridge Settlement Amount defined as $459,293,017.80. The arithmetic maximum is therefore about $17.21 billion, of which the Exhibit B maximum alone is about $16.68 billion and the guaranteed portion about $11.66 billion. Source: Dkt 572-1, Sections I.N, VI.B and VI.E and Exhibit B.

  2. [ESTABLISHED, and it is the most underreported term in the agreement] About 30 percent of the money is contingent, and the condition is what Meta's competitors do. The Contingency Installment Payments ($5,024,026,007.70 over ten installments, 30.1 percent of the Exhibit B maximum) are owed only once a "Contingent Monetary Payment Trigger" occurs in a given state, which requires both industry-wide adoption of the contingent time-management terms and that "all Core Industry Members with annual profits above $10 billion are subject to a monetary obligation to such Settling State" of comparable size. "Core Industry Members" is a defined term and means exactly three companies: the agreement says it "shall mean Snap, TikTok, and YouTube, for so long as the product or service is available to US Teen Users." If the trigger is never met, the agreement is explicit about where the money goes: those payments "shall be permanently forfeited by such Settling State and retained by Meta." Source: Dkt 572-1, Sections I.U, I.W and VI.D.

  3. [ESTABLISHED, and it is the same condition applied to the remedy] The brake is written in two phases, and the stronger phase is contingent on the same industry-wide adoption. Phase I applies for five years and is unconditional: a default night access mode "from 12 a.m. to 6 a.m.", and a default daily limit "of 2 hours per day" measured cumulatively across the covered platforms. Phase II applies only "whenever, in the 10 years following the Effective Date, Industry-Wide Adoption has occurred" and remains in effect for the obligation in question, and widens the night mode to "from 10 p.m. to 7 a.m." "Industry-Wide Adoption" in turn requires Snap, TikTok and YouTube to be bound to substantively equivalent obligations by settlement, by statute, or by audited voluntary undertaking. Source: Dkt 572-1, Sections I.DD and II.B.1 to II.B.3. For this project this is the finding of the whole document: the settlement concedes the brake is buildable and then makes its stronger setting conditional on competitors not defecting, which is the collective-action problem the policy paper argues a written standard exists to solve.

49a. [ESTABLISHED, and it bounds every product term above and below] The phrase Meta's apps, as this record has been using it, means Instagram and Facebook and not the rest of Meta. The defined term every injunctive obligation attaches to is "Meta SMP", covering "the existing and future web and app versions of Instagram" and of Facebook, plus future products meeting a definition, "excluding, in each case, any and all (A) direct messaging features linked to or included in those platforms, such as Instagram Direct or Messenger". WhatsApp, Messenger and Meta's other products are outside the brake entirely, and time in messaging does not count against the daily limit. Reader-facing prose on this site should name Instagram and Facebook rather than gesturing at Meta's apps generally. Source: Dkt 572-1, Section I.II.

  1. [ESTABLISHED] The Phase I terms are written the way this project's standard asks a brake to be written, and the language is worth recording precisely because it is enforceable rather than aspirational. The default "cannot be modified to a less restrictive setting without approval from a Supervising Parent". Meta must limit adjacent surfaces so that teens "cannot access broader Meta SMP features (beyond the content in a single message) and effectively circumvent" the limit. And when the brake engages, Meta's apps "shall not recommend or otherwise suggest that the Teen User utilize a different Meta SMP or engage in messaging", which is an explicit ban on the sideways nudge. Source: Dkt 572-1, Sections II.B.2.a and II.B.3.a.

  2. [ESTABLISHED] Other injunctive terms, all default-on for teen users unless noted. Numbers of likes and reactions: Meta's apps "will, by default, disable Teen Users from seeing numbers of likes or reactions", modifiable only by a supervising parent. Cosmetic procedure filters are disabled for teens outright. A non-personalized feed must be offered as a selectable default home feed within four months, with a prompt every 90 days, and the agreement forbids a thumb on that scale: Meta "shall not preselect or prioritize accepting or rejecting this switch for the Teen User and shall present this choice independent of other setting options". Note the asymmetry that the site should not blur: the time limits and the likes counter are defaults, while the non-personalized feed is an option. Source: Dkt 572-1, Sections II.C and II.D.

  3. [ESTABLISHED] Compliance is audited, but not for the whole term. An Independent Auditor reports to a State Committee on implementation and must "prepare and make public an executive summary of each Final Report". The auditor's term "shall begin two (2) months following the Effective Date and continue until 120 days following the fifth Final Report", while the obligations of the judgment "shall expire 10 years from the Effective Date". So the audited window is roughly the first five years of a ten-year judgment, and the later years are enforced by the court's retained jurisdiction rather than by a standing auditor. Source: Dkt 572-1, Sections III.C, III.G and III.J; Dkt 576, Section IX.B.

  4. [ESTABLISHED] The Effective Date is defined as "the first business day after which the Court in the MDL Action has entered the Consent Judgment", which makes it 2026-08-27, and the Compliance Date is six months after that. Dates in the injunctive terms run from those two, not from the announcement. Source: Dkt 572-1, Sections I.Y and I.R.

  5. [ESTABLISHED, and the site must not overstate what this settles] The judgment records no finding of wrongdoing and expressly disclaims being a standard. It states that the agreement and judgment "do not constitute an admission by Defendant of any liability, wrongdoing, or violation of any local, state, federal, or international law", and, in the sentence that matters most for how this project may describe it, that "Nothing in this Consent Judgment or the Agreement shall be construed to apply to, establish a standard of care for, or serve as precedent in any non-participating U.S. state or any international jurisdiction whatsoever." The obligations are contractual and limited to the settling states. Source: Dkt 576, Section IX.C.

  6. [ESTABLISHED] What is not released. The release carves out criminal liability, securities, tax, antitrust and environmental liability, and two categories that matter here directly: "Any claims of private individuals for any types of monetary or injunctive relief", and claims by school districts and other governmental-unit plaintiffs whose cases are in JCCP 5255 or the MDL. The personal-injury and school-district tracks of MDL 3047 survive this judgment, as do the other three defendants, who are not parties to it. What is carved in rather than out is at claim 61: the fifteen satellite state-court AG cases the release does reach. Source: Dkt 572-1, Section IV.C.1.

  7. [ESTABLISHED] Both sides waived appeal. The judgment provides that "Each Party waives all rights to appeal this Final Judgment and all orders to date arising from the Action" and requires every party to withdraw pending appeals within five business days, with a carve-out for the portions of any appellate matter that arise from other actions. That disposes of Meta's remaining appellate track in this case, including anything left over from the Ninth Circuit disposition at claims 38 to 42. Source: Dkt 576, Section XI.B.

  8. [ESTABLISHED] The full trial record, now that it is closed. Evidence was heard on four days: 2026-08-18 (5 hours 17 minutes), 2026-08-19 (5 hours 13 minutes), 2026-08-24 (5 hours 27 minutes) and 2026-08-25 (5 hours 22 minutes). 2026-08-20 was a status conference of 1 hour 34 minutes on "trial filings, protocols and e-discovery issues", after which the matter was continued to the Monday. The plaintiff states called five witnesses in all: Arturo Bejar, Dr. Jean Twenge, George Volichenko, Francesco Fogu and Adam Mosseri. Source: Civil Minutes for each day (AG-case Dkt 550, 560, 561, 571, 573, 574).

  9. [ESTABLISHED] Mark Zuckerberg did not testify at this trial. The states' case was interrupted by the settlement four days into evidence, and no civil minute entry records him as a witness. This closes an open question this ledger carried from the opening day, and the answer is that the trial ended before it was reached, not that either side resolved it. Source: the civil minutes for every trial day, which name every witness called.

  10. [OBSERVED, and it is a caution about secondary coverage of a number] Coverage of the settlement on 2026-08-26 and 2026-08-27 reported the total as $16.68 billion, $16.7 billion, $17 billion, $17.1 billion and $18 billion, in outlets of comparable standing, and one report described the coalition as 29 states while another said 47 states plus four territories and a third said 51 attorneys general. The Exhibit B figures at claim 47 reconcile most of the spread: $16.68 billion is the exhibit's own maximum, and roughly $17.2 billion is that plus the cost fund and the Cambridge payment. No published account located in this pass separated the guaranteed money from the contingent money, which is the distinction that determines what Meta actually pays. Source: contemporaneous coverage, surveyed 2026-08-28; reconciliation against Dkt 572-1, Exhibit B.

  11. [ESTABLISHED, corrected 2026-08-28 on the same day's audit] The Cambridge Settlement Amount buys releases of Cambridge Analytica-era claims from 48 of the 51 settling states, allocated by Exhibit E, whose 48 rows sum to the defined $459,293,017.80 exactly. An earlier version of this claim said the settlement resolves the four Cambridge Analytica-era state complaints outright, and that is not what the document does. The agreement enumerates four pending Cambridge complaints as the subject-matter template: People v. Meta Platforms, Inc., No. 25-631678 (Cal. Super. Ct., S.F. Cnty.); State of New Mexico ex rel. Torrez v. Facebook, Inc., No. D-101-CV-2021-00132; District of Columbia v. Facebook, Inc., No. 2018-CA-008715-B; and People of the State of Illinois v. Facebook, Inc., No. 2018 CH 03863. But the "Cambridge Settling States" list excludes California, the District of Columbia and American Samoa, and does not include New Mexico at all, and only Cambridge Settling States are Cambridge Releasors. So of the four enumerated complaints, only Illinois's is held by a releasing state; the California, D.C. and New Mexico Cambridge cases are not released by this agreement, and nothing in it establishes what becomes of them. The New Mexico one is in any event a different case from the addiction case in this project's New Mexico ledger (No. D-101-CV-2023-02838), and neither New Mexico case is part of this settlement. Source: Dkt 572-1, Sections I.L, I.M, I.N, I.O and IV.B, and Exhibit E (48 rows verified to sum to the defined amount).

  12. [ESTABLISHED, and it is the part of the scope this record missed on its first pass] The settlement reaches past this case into the states' own courts. "Actions" is defined as "the MDL Action and the Satellite AG Actions", and "Satellite AG Actions" is an enumerated list of fifteen state-court cases in fourteen jurisdictions: Arkansas (Polk County), the District of Columbia, Iowa (Polk County), Massachusetts (Suffolk County), Mississippi (Hinds County), Montana (Lewis and Clark County), three Nevada cases (Clark County), New Hampshire (Merrimack County), Oklahoma (Osage County), Puerto Rico (San Juan), Tennessee ("State of Tennessee v. Meta Platforms, Inc., et al., Chancery Court, Twentieth Judicial District, Davidson County, No. 23-1364-IV (Tenn.)"), Utah (Salt Lake County), and Vermont (Chittenden County). The agreement makes those cases end in two steps, on two different clocks, which is worth separating because they are easy to conflate. First, under Section X.G, "Promptly, and in no event more than 24 hours after the execution hereof", Meta and the settling states must "jointly notify the courts in each of the Actions" of the settlement and "jointly request that each court vacate all deadlines, including trial dates". Then, under Section X.I, they "shall jointly move in each Satellite AG Action within ten (10) business days" of that notice "for the entry of the Consent Judgment fully and finally resolving the Releasors' Claims". A vacated calendar is not a resolved case; the second step is the one that ends it. The Tennessee entry is the case this project covers as its own separate proceeding, whose Nashville trial was in its fifth week of evidence on the day this agreement was executed; see the tennessee-v-meta ledger, section (g), for that side of it. Source: Dkt 572-1, Sections I.A, I.ZZ and X.I.

61a. [ESTABLISHED as to the list and the one docket read; the access survey is the drafter's, dated 2026-08-28] The fifteen satellite cases, and whether anyone can actually watch them. The agreement gives each a caption and a number, which makes them checkable in principle. Whether they are checkable in practice varies by state, and this was surveyed rather than assumed:

JurisdictionCase numberPublic docket access, probed 2026-08-28
Arkansas, Polk Cnty. Circuit57CV-23-47CourtConnect returns HTTP 200 and an empty JavaScript shell; a plain fetch sees no case data at all
District of Columbia Superior2023-CAB-006550eAccess search endpoint returned HTTP 500; no case-level query was reached
Iowa, Polk Cnty. DistrictEQCE092822not probed
Massachusetts, Suffolk Superior2384CV02397only the portal landing page was loaded; the case itself was not queried
Mississippi, Hinds Cnty. Chancery25CH1:23-cv-01205not probed
Montana, Lewis & Clark Cnty.BDV-25-2024-0000797not probed
Nevada, Clark Cnty. District (three cases)A-24-886110-B, A-24-886115-B, A-24-886120-Bprobe redirected (HTTP 302)
New Hampshire, Merrimack Superior217-2023-CV-00594not probed
Oklahoma, Osage Cnty. DistrictCJ-2023-180fully open, server-rendered, no login: the whole docket is one HTTP GET
Puerto Rico, San JuanSJ2024CV11569not probed
Tennessee, Davidson Cnty. Chancery23-1364-IVregister free, documents behind a paid subscription (see the tennessee-v-meta ledger)
Utah, Salt Lake Cnty. Third District230908060only the XChange landing page was loaded; the case itself was not queried
Vermont, Chittenden Superior23-CV-04453the portal root timed out; the case itself was not queried

The survey's own limits, stated because a table like this invites over-reading. Only three rows record a real case-level query (Arkansas, Nevada, Oklahoma; Tennessee's access was established separately in its own ledger). Five were not probed at all, and four more were touched only at a portal's front door, which proves nothing about the case behind it; a not probed row is not a paywalled one, and the earlier version of this project's position, that these are mostly paywalled and not worth tracking, was an assertion with one data point behind it and is retired here. What the probing does establish is that at least one satellite docket is completely open, which is enough to disprove the general claim, and that HTTP 200 is not evidence of access for this family of sites: Arkansas returns a successful empty shell against a direct case URL, the same false-success shape this project already documented for the European Commission's press corner. Source: the case list from Dkt 572-1, Section I.ZZ; the access results from probes run 2026-08-28 and recorded here rather than in prose.

61b. [ESTABLISHED, and it is the satellite mechanism working, observed on a primary] Oklahoma's docket shows the settlement arriving. The Osage County docket for CJ-2023-180 carries, on 2026-08-26, the entry "JOINT MOTION TO VACATE ALL CASE DEADLINES", the same day the federal consent judgment was entered, alongside a "STIPULATED AMENDED PROTECTIVE ORDER" the day before. That is the first of the two steps at claim 61, the 24-hour obligation under Section X.G, not the ten-business-day one under X.I. An earlier version of this claim called it the ten-day obligation, which conflated the two clocks: what Oklahoma did on the day was ask its court to clear the calendar, and the motion to enter a consent judgment there is the next thing to look for. Either way it is visible in a state court that charges nothing to look. What cannot be done there is quotation: OSCN serves its docket entries as text but its underlying documents as TIFF images with no text layer (a pdftotext extraction of the motion returns zero characters against a passing control), so those documents are pointers to a place in a record, never a text to quote, the same treatment this project gives ASR transcripts and the Dutch judgments. Source: OSCN docket export, cached 2026-08-28; OSCN's own header states the information is not an official record, which is why the entry titles are cited and nothing is quoted from the documents.

  1. [ESTABLISHED, and it is the quietest term in the agreement about the loudest count at trial] The deception theory gets a forward-looking injunction, and it is the least enforceable term in the document. Section II.E.4 provides: "Meta is enjoined from making false, misleading, or deceptive representations regarding the effect or efficacy of safety features for Teen Users on Meta SMPs." Then the gates: it "applies exclusively to representations made by Meta on or after the Effective Date"; it "may be enforced exclusively with the consent of a majority of the State Committee, and only if the enforcing state provides Meta with at least 30 days notice of the alleged violation, and an opportunity to confer and propose a corrective action plan"; and compliance with it "shall not be subject to the independent audit under Section III". The claim that carried both trials (that Meta misrepresented its safety tools) is answered prospectively by an injunction no single state can enforce alone and the auditor is instructed not to examine. Source: Dkt 572-1, Section II.E.4.

  2. [ESTABLISHED, recorded because the site's own test runs on these definitions] Four defined terms decide how strong the product obligations actually are. "Reasonably Accessible" (the standard for finding the feed option): "viewable within three user gestures and clearly labeled, easy to notice, viewable without scrolling, and discoverable in an intuitive location within the Meta SMP". "Clear[ly] and Conspicuous[ly]" (the standard for the prompt): "difficult to miss (i.e., easily noticeable), and easily understandable by users". "Non‑Personalized Feed": "a Meta SMP feed for which content is populated by accounts the Teen User follows or has friended, displayed in chronological order", with a carve-out for integrity, security and legal moderation that "shall not constitute personalization". And "Longform Content": "a piece of video or audio content with a duration of at least 22 minutes" not artificially extended, which matters because the daily limit's own text says it "shall not include time spent watching Longform Content, engaging in messaging, or accessing settings". A teen's hour inside a 22-minute-plus video counts as zero against the two-hour limit. (Note the extraction wrinkle: the defined term uses a non-breaking hyphen, U+2011, where the operative text uses an ordinary one, so a naive search for the definition by its own name misses it.) Source: Dkt 572-1, Sections I.Q, I.EE, I.OO, I.VV and II.B.3.a.i.

Tensions / open questions

  • Whether Mark Zuckerberg testifies at this trial. CLOSED 2026-08-28, and the answer is that the question expired. He did not. The trial ended on 2026-08-26, four days into evidence, and no civil minute entry names him as a witness; the states called Bejar, Twenge, Volichenko, Fogu and Mosseri and got no further. The expectation recorded here on the opening day (Meta's counsel saying in opening statement that he would testify, the judge indicating the same) was never tested. He did testify in person in the California state bellwether on 2026-02-18 (see california-state-bellwethers claim 26), which remains the only occasion on which he has answered for these design decisions under oath in a trial. Source: the civil minutes for every trial day (claim 57). Note what this closure is worth: a settlement four days into evidence is also a settlement before the chief executive took the stand, and this record should not imply the two are unrelated when it cannot show that they are.

  • What Meta's age gate actually does when a user enters an under-13 birthdate (opened 2026-08-20). This is the sharpest factual dispute this record has found about a specific product mechanic, and the COPPA claims at the centre of this trial turn on it. Colorado's attorney general said in public that entering a birthdate under 13 does not remove the user: the product prompts them to enter a different date. In sworn testimony in the California bellwether six months earlier, Meta's chief executive said enforcement of the age limits is very difficult and that the company removes users found to be underage. Those are not the same account. One describes a product that actively coaches a child past the gate; the other describes a company struggling to catch children who lie. Both cannot be the whole truth of the same screen, and the difference is exactly the kind this project exists to resolve, because a control that redirects a truthful answer is not a weak brake but an inverted one. Sources: press-conference ASR at 10:37 (CAPTION-ASR, quarantined) and multi-outlet coverage of the 2026-02-18 testimony (secondary). Neither is a primary. Resolvable from the trial record, from Meta's own product, or from the COPPA exhibits.

  • Are the "four states" a subset of the "29 states"? RESOLVED (2026-07-16): subset. Law.com's headline counts the first trial at 29 states total; Reuters describes the COPPA claims plus consumer-protection claims from CA/CO/KY/NJ; all four are among the original 33 federal filers. No source lists the four as additional to the 29. Caveat: the subset reading is a well-supported inference across sources; no single source states it in one sentence.

  • "Terms formally undisclosed" vs. specific settlement dollars. RESOLVED (2026-07-16). Reuters obtained the actual settlement agreements through a public-records request, so the per-defendant amounts (claim 23) are documented terms, not reported estimates. "Not disclosed in court" remains true.

  • "Bellwether" is used for different things. The settled bellwether was the first school-district test case (Breathitt County). The August trial is the state attorneys general's trial. A first school-district bellwether trial (Tucson) is expected February 2027. All get called "bellwether" in coverage; they are different tracks within the same MDL and should not be conflated.

  • The "$1.4 trillion" is a ceiling, not a verdict. It is Meta's own framing of the maximum arithmetic exposure. What the states will actually argue for, and what a court could award, are unresolved.

  • The exact rosters of the 29 (August) and 14 (February 2027) states are not enumerated in any accessible source. Narrowed 2026-08-20: the departures are now named (Michigan, Georgia, Missouri, North Dakota, January 2025, claim 17a), which accounts for 33 minus 4 equals 29 without enumerating who remains.

  • The four withdrawal filings have not been read (opened 2026-08-20). Claim 17a rests on the Michigan Attorney General's office as quoted by one outlet, plus structural corroboration from the Ninth Circuit caption. A search of the MDL docket for a January 2025 notice of dismissal did not locate one in this pass, and that is a statement about the query rather than about the docket: the search API samples documents rather than returning a full docket, so absence here is not evidence. Reading the filings would take claim 17a to ESTABLISHED and would settle whether the other three states gave the same reason.

  • The Section 230 narrowing is ANCHORED as of 2026-08-18, though not to the orders themselves. This was opened 2026-07-24 because the hub described earlier MDL rulings as narrowing the case on Section 230 grounds with nothing in this ledger tying that to a document. The Ninth Circuit's published opinion now describes those rulings directly, which is a better anchor than the coverage and a weaker one than the orders: on the first track, the district court granted Meta's motion in part and denied it in part, "ruling that section 230 bars some of the claims and only to the extent that they concern allegedly defective platform features targeting Meta's role as a publisher of third-party content", and it then denied Meta's motion to certify that order for interlocutory appeal under 1292(b); on the fourth track and most of the second, it "generally denied" the motions and let the claims proceed while finding that Section 230 provides "a fairly significant limitation" on them. What is still missing is the order numbers and dates, which an appellate opinion does not supply. The hub may now state the narrowing as a holding described by the Ninth Circuit rather than as background. See claims 38 to 42.

  • The states' own penalty submission has been read. CLOSED 2026-08-18, and the answer was not the expected one. Dkt 473 was fetched and read. It states no monetary total: the AGs say they will present final requests for monetary relief at trial. The only "$200 billion" in it is Meta's 2025 revenue, cited to Meta's Form 10-K. See claims 22a to 22c.

  • What the states will actually seek. LARGELY ANSWERED 2026-08-20, and the answer is that they will not name a number (opened 2026-08-18). Four outlets covering the opening reported roughly $200 billion as the states' ask, two of them placing the statement at a hearing in the week of August 10 and one attributing it to a court filing. No filing reached states it as a demand, and the states' own filing uses that number for the defendant's revenue. Two readings were left open: the states stated it separately at an August hearing, whose transcript (Dkt 540) is not mirrored in the free archive; or the coverage read a revenue figure as a demand.

    The second reading is now strongly supported, from the plaintiffs in public. At the press conference on the day of opening statements a Reuters reporter put both figures to the attorneys general and asked whether the states had a number. California's attorney general said the states are not asking for $1.4 trillion, that the figure is Meta's own arithmetic maximum, that $200 billion is Meta's annual revenue, and that the states leave the final amount to the judge's discretion. See claims 22d to 22g. So the honest answer to "what will the states seek" is that they have declined to state a total, which is the same thing Dkt 473 says.

    What is still open, and why this is not marked closed. Those claims rest on a machine transcription and are tiered OBSERVED; ASR is not a verbatim record and cannot establish exact words. Dkt 540 remains unmirrored, so if a figure was stated at the August hearing this record still cannot see it. And a public statement about what a party is not seeking is not a filing: the states' actual trial request, when made, is the thing that settles this. The site should still publish no figure as the states' demand, which is what it already does.

  • Whether the trial was bifurcated is now unanswerable, and stays recorded rather than closed (opened 2026-08-18). See claim 37. The trial ended before any remedies phase could exist, so the docket will never settle what the structure would have been. It is left here because the underlying lesson is the durable part: a party's docket-entry caption is not a holding, and this record declined to publish one as structure.

  • Does the February 2027 follow-on trial survive the settlement? NARROWED 2026-08-28, and mostly answered by the judgment's own terms. Claim 17 records a second trial for 14 further states' own state-law claims. The entered judgment states that "all claims by the States in this Action are fully and finally resolved through entry of this Consent Judgment", and the released states include every state that was to be tried, so the states' February 2027 trial has nothing left to try. What holds the February 3, 2027 date now is the school-district bellwether track, which the release expressly carves out: coverage identifies the cases as the Tucson Unified School District and the Charleston County School District, the second of which this record had not previously named. Still open: no scheduling order confirming the school-district pairing has been read here (the Tucson expectation is claim 17's, from Meta's own filing; Charleston is coverage only), and no order formally vacating the states' follow-on has been located, so the mechanics are inferred from the release's scope rather than from a docket entry. Narrowed further 2026-08-28 by ECF 3407 (claim 31a): X states it "is not a defendant in any bellwether", and Discord's and Roblox's cases are stayed outside discovery, so whatever is tried in February 2027 is tried against the original four. Source: Dkt 576, Section XI.A; MDL Update and JTNY trial-calendar coverage, August 2026 (secondary, for the two district names).

  • Whether the fourteen non-Tennessee satellite cases actually get their consent judgments entered. OPEN (2026-08-28), and partly checkable for free. Claim 61 records the ten-business-day obligation, which runs to roughly 2026-09-10. Claim 61b shows Oklahoma performing it. The other thirteen were surveyed for access (claim 61a) and most are either behind a subscription, behind a JavaScript shell, or unprobed. This is a Tension rather than an issue because it belongs to this proceeding's evidentiary record, and it has a real trigger rather than a reminder. Trigger: refetch https://www.oscn.net/dockets/GetCaseInformation.aspx?db=osage&number=CJ-2023-180 and look for an entry disposing of the case; Oklahoma is the cheapest proxy for whether the mechanism is completing generally.

  • Whether the four withdrawn states were paid as part of a negotiated re-entry or by operation of the release. OPEN (2026-08-28). See claim 46a. Michigan, Georgia, Missouri and North Dakota left over discovery burden and are settling states with Exhibit B rows. The agreement does not narrate how they got there. Answering it would say something about whether withdrawing from a case like this actually costs a state anything.

Sources

← the MDL 3047 case file

all evidence ledgers