In 1997, I put on my cycling clothes and rode about two miles to 35 West Wacker Drive in Chicago with the first expert-witness search I had ever done. I had printed the candidates' resumes on linen stock and bound them into a volume I called a Knowledge Portfolio. I thought it was a handsome piece of work. I also thought a messenger would cost too much.
The lobby was full of marble and sculpture. I made it as far as the front desk. "Messengers need to sign in at the loading dock," the guard told me. The next day I came back for a meeting with the lawyer at Winston & Strawn who had asked for the search. I wore a suit and tie whose colors and style I had borrowed from a photograph of the Duke of Windsor. This time the same guard offered cold sparkling water. "Yesterday a courier, today a courtier," I wrote in The Scholars' Treasure. My qualifications had improved remarkably overnight.
I thought of that guard when Roblox, the game platform, began asking players to show it their faces. The company had a serious reason for doing so. It wanted to know how old the people in its chat rooms were. Once the check was in place, Roblox could say what share of the accounts playing on a given day had passed it. I wondered what that share could tell us about the accounts that had been asked for a face and might never show up again.
The answer, after a good deal of arithmetic, is not much. We built two futures for Roblox. In one the business grows again. In the other it shrinks. By June 2027 about 80 percent of daily activity comes from checked accounts in both, 78.3 percent in the growing one and 80.0 percent in the shrinking one. Yet the two sit about $514 million apart in a single quarter's bookings, the money players pay in. The shrinking company reports the higher share. Roblox's shares have lost more than half their value since it set the dates for the check, so that is an expensive thing for a percentage not to tell you.
I came to the story through a shareholder complaint, Police & Fire Retirement System of the City of Detroit v. Roblox Corporation et al. I read filings for fun. Helping trial lawyers find expert witnesses does things to a man's reading habits. Case Chase, the tool my assistant and I built to read new complaints, sent this one to Vista's research desk on October 19, 2026. We take no view of the case. We waited for Roblox's October 29 results, and the question we took to them was our own. Would players get used to the new requirement, or had Roblox changed the reasons some of them came in the first place?
I come to video games with a history. As a boy in New Jersey in the 1970s and 1980s, I loved going to arcades with my brother Andy, and I collected rolls of quarters so that I could stay as long as possible. The quarters went into machines that somebody else had built. That was the arrangement, and I held up my end of it faithfully. At fourteen I attended Atari Computer Camp in East Stroudsburg, Pennsylvania, and learned to program video games myself. I spent much of my early teens playing the famous games of the 1980s and even designing games of my own. I would never have imagined how games would evolve into today's Roblox.
Roblox lets people make games and other people play them. It supplies the tools, a program called Roblox Studio, and its users do the building. The arrangement of my boyhood has been turned around. Now some of the players build the machines. David Baszucki and Erik Cassel developed Roblox in 2004 and released it to the public in 2006, and Baszucki runs it today as chief executive. A friend can meet you in one game, talk with you by text or voice, and go with you to the next. In the third quarter of 2026, Roblox averaged 138 million active accounts a day. Of the daily users who have passed the age check, by Roblox's count, more than two in three are under eighteen. Children are part of the reason the place is so valuable and the reason the company chose to do something about who could talk to whom.
The quarters have changed too. Players buy Robux, Roblox's own currency, with ordinary money and spend it on things like clothes for an avatar or a pass to a level. A creator whose game earns Robux can exchange them for cash. In 2025 Roblox paid creators about $1.5 billion that way, up 63 percent from the year before, and more than 23,500 of them were paid. No arcade ever offered me those terms.
The money players pay in during a quarter is what Roblox calls bookings. Bookings tell us when sales take place, and they came to $6.8 billion in 2025. Revenue, the figure on the income statement, is mostly bookings spread over the roughly 27 months Roblox estimates a paying user stays. The company became publicly traded on March 10, 2021, through a direct listing of its Class A shares on the New York Stock Exchange.
Children, strangers and chat in one place raise a question of safety. In July 2024 Bloomberg Businessweek published an investigation reporting that since 2018 police in the United States had arrested at least 24 people accused of abducting or abusing victims they had met or groomed using Roblox. That is Bloomberg's reporting, not the finding of any court. The same day, Roblox's chief safety officer, Matt Kaufman, wrote that "safety issues are not widespread or systemic on Roblox", though "even one incident involving the health or safety of children is horrible".
That October a short seller, Hindenburg Research, published a report alleging that Roblox overstated the number of people on its platform and had compromised child safety for the sake of growth. Hindenburg disclosed in the report that it held a short position in Roblox's shares, a bet that pays if the price falls. Roblox rejected the report the same day and said in a statement: "The financial claims made by Hindenburg are misleading."
Governments followed. Louisiana's attorney general sued in August 2025, and other states, with Los Angeles County and San Diego County, have sued since. In the main they allege that Roblox misrepresented the safety of its platform and failed to protect children. Roblox disputes the states' allegations. Families have sued too. They allege that children were harmed by predators who targeted them through the platform. Roblox's quarterly report says the company "intends to defend itself vigorously in all such matters". None of these claims against Roblox has been proven in court, and we say nothing here about how any of them will or should come out.
Roblox's other answer was to change the product. It announced a timetable on November 18, 2025, following a plan it had set out in September, for requiring an age check to use chat. Players could use a camera for a facial age estimate or, from age thirteen, show a government ID. Enforcement began in Australia, New Zealand and the Netherlands in the first week of December and went worldwide in early January 2026. A player could still enter a game without checking. Using chat required the extra step. A conversation now had a price, and the price was a face.
Roblox did not take this step in a vacuum. Britain's Online Safety Act has required, since July 2025, that services children are likely to use put highly effective age checks in front of the most harmful content. In March 2026 Ofcom, the British regulator, named Roblox among the six services children use most and told them all to do more. It noted that Roblox had "put age checks in to restrict adults from contacting children they don't know." Australia's law barring children under sixteen from social media took effect on December 10, 2025. It excluded online gaming, and the Australian regulator's list of restricted services left Roblox off. Laws like these are pushing more platforms with young users toward a door of their own, which is why the answer here matters beyond one company. Whether laws like these are wise is a question for legislatures and the public, and this inquiry takes no side on it.
For an investor, the controversy has also reached the financial statements. Roblox has signed settlements with states, including six that its September 2026 quarterly report names, and says it is negotiating with certain other states and territories. It recorded legal settlement expenses of $57 million, $34 million and $49 million in the first three quarters of 2026, $140 million in all. Roblox leaves those expenses out of its adjusted earnings. They have their own line in the reconciliation and do not enter bookings. The players who met the check and did not come back have no line anywhere. Any effect of the age check on the number of players who return and spend money has to be worked out another way.
What has happened since can be said in a few numbers. Roblox's October letter reported $1,697 million of bookings in the summer quarter of 2026, down 12 percent from a year earlier. Daily users were down 9 percent and hours were down 15 percent, to nearly 34 billion. The shares had closed at $101.27 on November 17, 2025, the day before the timetable was announced. They closed at $45.45 on November 3, 2026, about 55 percent lower. The price tells us the stakes. It gives us no reliable way to assign the loss to the age check.
The first numbers sounded encouraging. By the time Roblox reported in February, 45 percent of its daily users had checked. The company described a mid-single-digit drag on engagement growth and a low-single-digit drag on bookings growth, with the lost hours largely from less engaged users. In April, Roblox said the headwinds were greater than expected. It believed the check had restricted chat for unchecked accounts, thinned it for those who had checked, and slowed new-user acquisition. The reported share had reached 51 percent.
By June the share was up to 57 percent. Bookings had come in at $1,557 million, the low end of guidance, and average daily users had fallen from 132 million to 123 million. Roblox said retention of existing users remained stable relative to historical periods. It also said it had more work to do to bring on-platform communication above pre-check levels. September brought a higher share again, 64 percent overall and more than 80 percent among users aged thirteen to seventeen. Daily users rose to 138 million, up 12 percent from June.
A percentage can go up because more accounts have checked or because fewer unchecked accounts show up to be counted. The spring is a good place to see the difference. Multiply the checked share by average daily users and you have an estimate of checked activity. What is left is unchecked. From March to June, the estimated checked count went up less than three million while the unchecked count fell almost twelve million. Total daily users fell nine million. In summer the checked count rose from about 70 million to 88 million while the whole platform gained 15 million. The estimated unchecked count slipped about three million. The share advanced six points in the spring and seven in the summer.
| Quarter | Daily users | Checked share | Checked, millions | Unchecked, millions |
|---|---|---|---|---|
| March 2026 | 132 million | 51% | 67.3 | 64.7 |
| June 2026 | 123 million | 57% | 70.1 | 52.9 |
| September 2026 | 138 million | 64% | 88.3 | 49.7 |
What lay behind those similar-looking gains was quite different. In the spring the share rose mostly because unchecked activity fell. In the summer it rose mostly because checked activity grew.
There are two limitations to that multiplication. The first is the unit. Roblox's daily active user, or DAU, is an account that logged in "on a given calendar day", averaged across the quarter. The company says plainly that its DAUs "are not a measure of unique individuals". An increase of 18 million in checked DAUs could come from the same checked accounts playing more often. It could also include returning accounts, newcomers who checked right away, or previous holdouts who finally did. The published figures do not give us the flows.
The second limitation is the calendar. Roblox measures the checked share over the last seven days of the quarter and averages DAUs over the whole quarter. Move end-of-quarter DAUs 5 percent either way and the little spring gain could turn into a loss. The summer gain still falls between about 10 million and 26 million. The assumed 5 percent gives us a way to see how the answer moves. The actual gap between the reporting windows is unknown.
September gives us a little more to work with. Daily accounts that sent a chat message or connected to voice rose 24 percent from June, which the letter called "double our total platform DAU growth rate." In-game chat messages rose 13 percent. The first figure includes voice users and the second counts text messages in games. An increase in voice use could change the relationship between those figures. I would like to know whether conversation is returning and in what form. The two measures give us the beginnings of an answer.
There were other reasons to expect the summer figures to move. The year before had hits in it, Grow a Garden, a game of planting and selling crops, and Steal a Brainrot, in which players raid one another's bases for cartoon meme characters, among them, so every comparison with 2025 starts from a high mark. Russia blocked Roblox in December 2025 and restored access in June 2026. Roblox changed its Recommended for You algorithm in spring and refined it in summer, and credited those refinements with better retention and spending. In June it began assigning unchecked users to more restricted accounts with gentler games and chat off. At the end of May it stopped letting one game sell another game's passes. And school let out. Roblox's fourth-quarter guidance allows for the kind of seasonal decline in users that has followed past summers.
Roblox also ran more than 200 paid campaigns in summer, with marketing to adults up nearly threefold. Adults accounted for 27 percent of checked activity in June and 29 percent in September. By the same multiplication, adults contributed about 6.7 million of the 18 million increase, more than a third. Checked adult activity grew about 35 percent in summer, against roughly 8 percent in spring. The age mix gives us a clue to what changed. It leaves open how much of the increase came from the campaigns.
At Vista we use a model named for the Russian mathematician Andrey Markov. A Markov chain describes movement between states using a set of probabilities for what happens next. The entry for it in our Decision Library is written around one sentence: the retention number is stable; the transitions are not. We have two pools of daily activity, checked and unchecked. Activity can arrive in either, move from unchecked to checked, or lapse out of either. In the model a check is permanent and a lapsed account that comes back counts as a fresh arrival. Paying is something an account can do in either pool on a given day. The filings supply neither the flows nor a rate at which real accounts move between those states. The exercise treats an average daily user as a unit of activity, as if an account played every day. The most willing holdouts may check early, so a constant conversion rate is already a generous simplification.
Give Roblox two possible futures. Neither is a forecast. In the first, which we call the reset scenario, 14 percent of unchecked activity turns checked each quarter and 10 percent lapses. New activity arrives at 13 percent of the base, and the share of daily activity that pays returns to its 2025 average by mid-2027. In the lower scenario, 5 percent of unchecked activity turns checked and 25 percent lapses. New activity arrives at 9 percent of the base and the payer share holds at its September 2026 level.
In both scenarios, three quarters of new activity checks on arrival and checked activity lapses at 10 percent a quarter. Both take the summer lift out of the September starting point, using the 2025 ratio of fourth-quarter to third-quarter DAUs. The conversion and lapse rates are our choices, picked so that the model can rebuild both of the quarter-to-quarter changes Roblox has reported. The inflows and the other settings are assumptions too.
By June 2027, the reset scenario gives us about 143 million daily users, 112 million of them checked. The lower scenario gives us about 111 million, 89 million checked. Divide each pair, before rounding. In the growing business, 78.3 percent have checked. In the shrinking one, 80.0 percent have checked. If you could see only the percentage, the shrinking company would look a little further along.
The reason is in the denominator. Unchecked activity that falls away raises the share just as surely as unchecked activity that turns checked. It does so by less for each million, but in the shrinking future far more of it falls away. The title of the Beatles' 1967 song "Hello, Goodbye" fits Roblox's checked share. A newcomer who checks on arrival and a holdout who quits both push it up, and the share cannot hear the difference.
The scenarios differ by about 32.5 million daily users. About 17 million of the difference comes from what happens to unchecked activity and 15 million from how much new activity arrives. Let the scenarios run and the checked shares settle at about 89.6 and 90.5 percent, close to the 90 percent Roblox has named as its long-term goal for age checks. Those limits depend in particular on the assumption that three quarters of new activity checks on arrival. You can get remarkably close to the same percentage with a great many fewer daily users.
The September 2026 quarterly report, the Form 10-Q, rounds daily payers to about 1.8 million and the year-earlier comparison to 2.3 million. Read those literally and the decline is 22 percent. Open the 2025 filing and the comparison quarter has about 2,250,000 daily payers. Fifty thousand daily payers seem to have joined that quarter in the year after it ended. Nobody bought anything. The figure was rounded up, and the rounding did Roblox no favors.
Divide bookings by the days in the quarter and by the bookings per paying user per day, which Roblox states to the cent. The estimates come out at about 2.25 million for the summer of 2025, 1.89 million in the first quarter of 2026, 1.64 million in the second and 1.85 million in the third. The summer-to-summer decline is about 18 percent, where the rounded figures said 22. Payers did fall twice as fast as users.
The daily payer share, payers divided by users, fell from about 1.5 percent in summer 2025 to about 1.3 percent by June 2026. It stayed there in summer. September's 1.34 percent is above the 1.28 percent of summer 2024 and below the 2025 average of about 1.41 percent. Daily payers spent about $9.98 per paying day in the September quarter, up 7.5 percent from a year earlier. A smaller group can spend more per person when some of its lightest spenders disappear or pay less often. Monthly unique payers were also down 16 percent even as average bookings per monthly unique payer rose 5 percent.
Even the small rise over the summer, from about 1.335 to 1.339 percent, deserves care. It is inside the rounding. It came in the quarter when checked adult activity grew fastest. Roblox says its U.S. adults monetize 50 percent or more above users under eighteen, an estimate it extrapolates from checked users to unchecked ones. That is a spending figure. It leaves the frequency of paying open. We tried the numbers. If a checked adult pays on a given day only a tenth to a sixth more often than everyone else, the change in mix could account for the whole of the apparent hold. A strong summer tells us something about the summer.
Now take the payer assumptions back to the two scenarios. By the June 2027 quarter the reset scenario has about 2.03 million daily payers and the lower one about 1.48 million. Keep the June 2026 rate of $10.42 per paying day and a 91-day quarter. Multiply the payers by the rate and by the days. Bookings come to about $1,920 million in one case and $1,410 million in the other, $514 million apart before rounding the totals. That gap is about a third of the $1,557 million players paid in during the June quarter of 2026. Against that quarter, one case is growth of about 23 percent and the other a decline of about 10 percent. The same eighty percent on an investor's screen can describe businesses going in opposite directions.
At $45.45 a share, Roblox's market capitalization was about $32.5 billion. Management spoke in April and July of a long-term goal of 20 percent annual growth or more. The October letter did not repeat it. Its fourth-quarter guidance implies full-year 2026 bookings growth of roughly 1 to 3 percent, with fourth-quarter bookings 10 to 15 percent below the previous year. Depending on which quarter you start with, the two-year annualized rate ranges from 18 to 32 percent, carrying both the 2025 boom and the changes of 2026. Anyone who owns the shares for that goal is betting that 2026 was a step down and not a new slope.
Read backward, the price requires free cash flow to grow 6.14 percent a year for ten years, at a 9 percent discount rate, when stock compensation is left out. Count the stock compensation and it requires 23.92 percent a year. Whether the price already assumes the company's 20 percent goal depends on how the stock compensation is counted. The working is in "What the price assumes" in the record below.
To tell a step from a slope, we set down a test in advance, with the evidence still to come. We froze it on October 31, 2026, and amended it on November 11. In February 2027 the next letter should give us a December checked share and the quarterly DAUs to put beside September's 88.32 million estimate. If overall activity falls by the same seasonal ratio as in 2025, a checked share of roughly 67 percent would keep the estimated count at the September level. Our reset scenario puts checked activity near 95 million in December, with 70.3 percent checked. The lower scenario puts checked activity near 87 million even as the share rises to about 71 percent.
It would take only a 1.8 percent change in the December estimate to erase the lower scenario's dip. That is why we amended the rule. As first frozen, it let a dip that small bar the top verdict whatever bookings and payers did. The February count is now an early reading beside the main score.
The deciding figures arrive with the June 2027 quarter, around August. A strong rebound requires bookings growth of at least 20 percent from June 2026 and an increase in daily paying users from the estimated 1.642 million baseline. Weak growth means bookings growth below 10 percent or a decline in daily payers. Anything else that can be scored reads cannot tell. The frozen record gives the precise rules for rounding, missing figures and the conditions that void the test. Russia's return will be noted beside the result. Both June quarters are under the chat gate. The comparison tells us about business performance over the interval. The effect of the gate itself remains open. A rebound can coexist with a lasting cost from the gate if growth elsewhere makes up for it.
Our assessment is short. Within the 5 percent sensitivity we ran on the timing of the two measures, the summer's evidence rules out a share that rose only because unchecked activity fell. Estimated checked activity rose by about 18 million in the summer, while Roblox added 15 million daily active users overall. Even after allowing for the difference in reporting windows, our sensitivity check still showed a gain of at least 10 million in checked activity. Beyond that it favors neither future. The summer was flattered by Russia's return, school holidays, paid campaigns, the June default into restricted accounts and recommendation changes the company credits with better retention and spending. And a payer share that holds in a strong quarter has not yet held in a weak one.
Asked for a number, we give one, as our judgment and not a computation. Conditional on a scorable result, it is about 25 percent strong rebound, 45 percent weak growth and 30 percent cannot tell. That is our estimate of the test, not a forecast of Roblox's share price or of the business beyond the test's two measures, bookings and daily payers. The November amendment moved five points from cannot tell to strong rebound, since the checked count, noisy as it is, could no longer veto it.
Our lower scenario scored weak growth in every variant we ran. The reset scenario scores strong rebound at the center and cannot tell when the payer share stays flat. Halfway between the two, bookings grow about 6 percent and score weak growth. We would rather be told cannot tell than be told what we hoped. The chance of a strong rebound is low because the test is strict about it and Roblox's own guidance leaves a long climb to 20 percent. It is not low because the summer's evidence favors the lower scenario.
The filings leave three questions for people who have watched a platform put up a gate. I would ask a former growth or retention lead how many of the affected accounts return within two to four quarters and what they spend when they do. I would ask somebody who has run age assurance how completion changes after a month, three months and a year, and who tends to stay unchecked. And I would ask a former Roblox creator who saw the January change in several games whether the old players came back and paid as they had before. The answers that would change our view are set out with the questions in "Primary research" below.
The next time a platform reports a rising share of users who passed an age check, ask how many active accounts sit behind that share. The same arithmetic applies to the share of users who pay for a free service. It can rise because the users who never paid stopped coming. The unemployment rate has the same feature. Someone who stops looking for work disappears from both the unemployed count and the labor force, so the rate can fall though nobody was hired. The comic book publishers of 1954 adopted a code with a seal of approval, and 28 of 31 of them had joined by December. The seal told them nothing about the readers who had stopped buying. That story is in the record below. A smaller denominator can make for a fine announcement.
At 35 West Wacker, I had someone upstairs to see, so I came back. Roblox's filings can help us count the days on which an account came back. I would like to ask the player who lost chat whether there was still somebody worth coming back to see.
What follows is the test as frozen, a dated reading added after an outside review, the amendment that followed a second review, the reverse valuation as a companion analysis, the model workings the body summarizes, and the fuller record of the Comics Code. The record calls the two scenarios the reset path and the lower path.
Here is what we froze on October 31, 2026, in a program with a digital fingerprint that refuses to run if any band, word or date is changed: d4da3b308e16c0a2c7f8e8367ed26debd269e6567048b3e6e8eff97614f6769d. The hypothesis is that Roblox's verified age gate on chat, global since January 2026, is a one-time reset that Roblox grows through and not a lasting lower path. It is scored from Roblox's shareholder letter and quarterly report for the second quarter of 2027, expected around August 2027, and assessed by September 30, 2027. Three measures.
The first is bookings. Bookings in the second quarter of 2027 against the $1,557 million of the second quarter of 2026, each taken from the quarterly report's reconciliation of revenue to bookings. Both quarters will have lived under the chat gate, so whatever it cut from the level drops out of the comparison and what remains is growth, with known exceptions: Russia was blocked for most of the second quarter of 2026 and will be open for all of 2027's, worth nearly three points of growth in daily users by the company's own figure for the headwind, which we record beside the score without moving the lines; and the base quarter was also held down by the recommendation experiments and holds only one month of the restricted accounts for unchecked users and of the end of cross-game pass sales, which the scored quarter holds in full. Reset if bookings are up 20 percent or more, the long-term rate Roblox has said it is focused on delivering, which sits inside the range of the last two years. Lower path if growth is below 10 percent, half the company's floor and below every two-year rate in the record, gate included. Between the two, the measure cannot tell.
The second is payers. Daily paying users in the second quarter of 2027 against the second quarter of 2026, each worked out, as above, from bookings divided by the days in the quarter and by the stated daily spending per paying user. For June 2026 that is about 1.642 million payers, within the roundings of the published figures. Payers grow or decline only when the two ranges do not overlap; otherwise they are flat.
The third is the count. The checked share times average daily users, as Roblox reports both, must rise from each quarter to the next: June to September 2026, which it already has, from 70 million to 88 million; then September to December 2026; December to March 2027; and March to June 2027. The December step is the one the seasons work against, because Roblox's own guidance assumes the usual seasonal fall in users from the third quarter to the fourth: if users fall only by the ratio of 2025, the count beats September's 88 million once the share reaches about 67 percent, a gain of about three points against the six and seven of the spring and summer. Along our reset path the count clears that bar, even if the summer's lift was twice what 2025's seasons suggest. Along our lower path users fall by more than the season takes, so the count falls in December, to about 87 million, while the share rises to about 71 percent: exactly the case in which the share would be telling investors the wrong story and the count the right one.
The verdicts. Reset if bookings grow 20 percent or more, payers grow and the count rises every quarter. Lower path if bookings grow less than 10 percent, or payers decline. Cannot tell for anything else, with every measure disclosed. Void in two cases. The first is a hit on the scale of 2025: any quarter from the fourth of 2026 through the second of 2027 with daily users or hours more than 50 percent above the same quarter a year earlier, a rule that would have fired in the second, third and fourth quarters of 2025, when hours ran 58, 91 and 88 percent above the year before, and in no other quarter from the start of 2025 through the third quarter of 2026. The second is Roblox's letter or report for the second quarter of 2027 naming a change to its discovery or recommendation algorithm as a main driver of that quarter's year-over-year change in bookings, users or payers. The Recommended for You changes of 2026 and Russia's return are recorded beside the score as known confounders, since they are already in the base, and do not void it; a bookings reading within about three points above either line is reported with the note that Russia's return could account for the margin. The test goes untested if Roblox stops disclosing the age-checked share, or stops giving daily payers in any form; if only the spending per payer disappears, the payer figure Roblox states is used, with 50,000 either side. A missing number is never a pass and never a fail. Legal settlement expenses are in no measure.
| Measure | Baseline, Q2 2026 | Reset | Lower path | Cannot tell |
|---|---|---|---|---|
| Bookings growth, Q2 2027 against Q2 2026 | $1,557 million | 20% or more | below 10% | 10% to under 20% |
| Daily payers, Q2 2027 against Q2 2026, unrounded | 1.642 million | grow | decline | flat |
| Checked count rising every quarter, Q2 2026 to Q2 2027 | 70.1 million at June 2026; 88.3 million at September | yes, in all four steps | any fall counts against reset | the other measures decide |
Two properties of the test should be said plainly so that its score is not over-read. Our lower path reads lower path in every variant we ran, including the kinder ones. Our reset path reads reset at the center, and it reads cannot tell when the payer share merely stays flat instead of recovering, with bookings up 16.9 percent, when spending per payer slips 5 percent, or when the summer of 2026 flattered the September count by a tenth; a path halfway between the two reads lower path, with bookings up about 6 percent. Swap only the holdouts' rates or only the newcomer flow between the paths and bookings growth comes out below 10 percent, between a rise of 2.4 percent and one of 9.6: a reset needs the holdouts, the newcomers and the payers all to go its way. So the bands lean toward cannot tell for a true reset and toward lower path for a mixed one, and in no case we ran did they produce a false reset. A third property belongs beside those two: the test scores the road Roblox is on after the gate, not the gate's share of it, so growth under 10 percent for a reason the voids do not catch, a longer fade of the 2025 games among them, would read lower path too, and so would a mixed road on which users and bookings still grow. We would rather be told cannot tell than be told what we hoped. Today the count has risen once, from June to September; everything else waits, and the test stands untested until August 2027.
A reading added on November 6, 2026. After an outside review of this inquiry, and after we read Roblox's supplemental deck for the third quarter, three things are recorded here without moving a band, a word or a date of the test above. First, the count measure is a measure of checked daily activity, since Roblox's daily active users are account activity on a calendar day, averaged over the period, and "are not a measure of unique individuals"; a rise in the count mixes new checks, lapses and returns with how often checked accounts log in, so a rising count will be reported as rising checked activity and never as a number of people who checked, and the test does not claim to measure conversions. Second, the count's two inputs are measured over different windows, the share over the quarter's last seven days and the users over the whole quarter, and the 5 percent exercise in the body is an assumed sensitivity, not a bound on the error; a December count within that band of September's 88.32 million, which includes the lower path's own 86.77 million, does not distinguish the paths, and any verdict on the count measure that turns on a difference of that size will be reported as inside the measurement uncertainty. Third, the test scores the trajectory of the business after the gate and not the gate's contribution to it, so a reset verdict is consistent with a lasting cost from checking that other things outweighed, and a lower path verdict with causes the voids do not name.
Amendment 1, dated November 11, 2026. Made after a second outside review and before any new result: we checked EDGAR that day, and since its third-quarter results of October 29, 2026 the only filings about Roblox were insiders' and one holder's reports of their shares. The checked count no longer decides the verdict. Its readings, the first of them the December count in the February 2027 letter, are reported beside the score as an early indicator and never decide it. The verdict is decided by bookings in the second quarter of 2027 against the second quarter of 2026 and by daily payers alone, with the 20 and 10 percent lines and the payer rule as frozen. The verdicts are named for what one quarter's bookings and payers can show: strong rebound, where the frozen text says reset, if bookings grow 20 percent or more and payers grow; weak growth, where it says lower path, if bookings grow less than 10 percent or payers decline; and cannot tell, as before, for anything else. A one-time reset and a lasting lower path remain the question holders face and the readings a verdict may support, never the verdict itself. The voids, Russia's note beside the score, the rule for a missing number and the exclusion of legal settlements stand as frozen, except that a missing count can no longer leave the test untested. Why: a change of 1.8 percent in the December estimate would erase our weaker scenario's dip in checked daily activity, from 88.32 million to 86.77 million, well inside the 5 percent sensitivity we ourselves assume, yet under the frozen rule that dip would bar the top verdict even with bookings up 25 percent and payers growing. The test frozen on October 31, 2026 and the reading of November 6, 2026 stand above, unchanged; the program still reproduces the frozen test's fingerprint, carries this amendment's own, and stops if a character of either moves.
Its digital fingerprint, computed in the program over the paragraph above exactly as printed: 0a4f8326f2620e831a1f01814c54ec8f798151df0d7361ccb902ec568b8785a0.
| Measure | Baseline, Q2 2026 | Strong rebound | Weak growth | Cannot tell |
|---|---|---|---|---|
| Bookings growth, Q2 2027 against Q2 2026 | $1,557 million | 20% or more | below 10% | 10% to under 20% |
| Daily payers, Q2 2027 against Q2 2026, unrounded | 1.642 million | grow | decline | flat |
| Checked count each quarter, Q2 2026 to Q2 2027, an early indicator | 70.1 million at June 2026; 88.3 million at September | reported beside the score | reported beside the score | reported beside the score |
Strong rebound needs both of its cells; weak growth needs either of its cells; anything else is cannot tell.
What the price assumes. We name no value here and no target, by Vista's rule. What we can do is read the price backward for the growth it requires, and say in words what a holder must believe to own Roblox on the company's own terms. At $45.45 a share Roblox's market capitalization, $32,529.2 million, is 20.75 times its free cash flow as we measure it for the twelve months to September 2026, $1,567.6 million. By that measure the price requires free cash flow to grow 6.14 percent a year for ten years at a 9 percent discount rate, and bookings the same if margins hold: far less than the company's goal of 20 percent or more, and a decade at the goal would offer a return of 16.94 percent a year. Free cash flow, though, is counted before stock compensation, $1,153.4 million in the trailing year, which Roblox pays in new shares rather than cash. Counted after it, the base shrinks to $414.2 million and the price requires 23.92 percent a year at 9 percent, more than the goal at every rate from 8 to 12 percent, and a decade at the goal would offer 7.62 percent a year. At discount rates of 8 and 10 percent the growth required is 3.88 and 8.21 percent a year before stock pay and 21.13 and 26.48 after it. Whether today's price assumes the company's goal turns on how stock pay is counted: before it, much less than the goal; after it, the goal and more.
Two cautions belong beside those figures. The trailing year is the base at its most generous. The third-quarter letter guides free cash flow for the fourth quarter of 2026 below zero, at minus $190 million to minus $105 million, "which includes the capital expenditures timing shift from Q3’26 to Q4’26". That puts the year 2026, on the company's own definition, at $1,064 million to $1,149 million, 26.4 to 31.8 percent below the trailing year's $1,561.1 million on the same definition, a base that requires 10.24 to 11.26 percent a year at 9 percent before stock pay and leaves no positive base at all after it. And "margins held" is a choice, not a record: free cash flow was 14.71 percent of bookings in 2024, 19.97 percent in 2025 and 21.75 percent in the trailing year, and after stock pay it was below zero in 2024, 3.33 percent in 2025 and 5.75 percent in the trailing year.
Set our two paths against the price, each one year of bookings growth and an illustration only. The reset path's first year, bookings up 23.4 percent, runs far ahead of what the price asks of free cash flow and close to what it asks after stock pay; the lower path's, down 9.6 percent, falls short of both. And even after a reset year the price, counted after stock pay, still asks 23.99 percent a year of the nine years that follow at 9 percent.
A word about what stays out. Roblox's third-quarter letter excludes from its Adjusted EBITDA, its own measure of adjusted earnings, $49 million of legal settlement expenses, $140 million for the year so far, over youth consumer protection and digital safety matters with states and territories. We have checked, line by line for six quarters, that none of it enters bookings, which are revenue plus the change in deferred revenue, the money taken in and not yet counted as revenue, plus a small adjustment; the settlements appear only in the reconciliation of Adjusted EBITDA. Nothing in this inquiry measures them, and nothing in it says how the matters behind them will or should come out.
What the counts cannot say. The counts cannot say how many holdouts converted: a daily user is a day of an account's activity, and an 18 million rise in average daily users could be 18 million accounts that play every day or 36 million that play every other day.
The Comics Code, in more detail. In the spring of 1954 the psychiatrist Fredric Wertham published Seduction of the Innocent, a book arguing that comic books were turning children toward crime, on evidence that a later scholar, Carol Tilley, found he had overstated and in places manipulated. A Senate subcommittee held hearings in New York that April and June; its eventual report did not blame comics for crime, and it recommended that the industry tone down its content voluntarily. The newly formed Comics Magazine Association of America adopted its code on October 26, 1954, and sealed each approved book with the words "Approved by the Comics Code Authority." The code purged sex and violence and forbade the words horror and terror in a title; the wholesalers agreed to handle only comics bearing the seal, and in its first two months the code's reviewers read 440 comics and rejected 126 stories outright. The publishers that had joined by December put out about three in four of the comics published. One industry history counts 364 comic titles in print in July 1954 and 293 by May 1956, puts monthly sales at a 1953 high of about 90 million copies and 50 million by the summer of 1955, and has EC Comics giving up comic books after ten months under the code. Nyberg warns against the fan folklore that the seal alone crippled the business: television arrived in the same years, and in 1957 the American News Company, the largest distributor of magazines and comics, abruptly quit the business and left many publishers without a route to the newsstand. Whether the gate kept the customers out, or television did, or the distributor's collapse did, is a question the comics industry never had the data to answer, and the parallel to a platform that cannot see its own lapsed users, in a summer that also brought Russia back and more than 200 paid campaigns, is close.
Chat, in more detail. In the third quarter the ratio of in-game text messages to daily users who used text or voice fell 8.87 percent (1.13 divided by 1.24, less one) while messages per user were flat within the roundings, at about 11.6 a day. The checked count's own growth runs from 22.95 to 29.07 percent on the roundings of the published figures alone, so chatting users per checked user ran from a fall of about 4 percent to a rise of about 1, and further either way on the date basis; the quarter reads as neither a pass nor a fail of the canary. This ratio uses the combined text-or-voice denominator. The letter gives no separate count of text senders.
Every Roblox figure in this inquiry outside the history of the company and the controversy, told in "Rolls of quarters", "The price of a conversation" and "How Roblox got here", is computed in code from Roblox's own shareholder letters, Forms 10-Q and Form 10-K for the quarters from the third of 2025 through the third of 2026, with the letters for the first two quarters of 2025 read for the void's record, and with each input traced to its passage in the inquiry's working files; the program carries 518 assertions, each checked figure against the printed one or a second computation. The checked and unchecked counts multiply the age-checked share Roblox reports for the last seven days of each quarter by the quarter's average daily users; they are estimates of checked and unchecked daily activity, since Roblox defines a daily active user as an account that logged in on a calendar day and reports the average over the period, and they are approximate for the difference in windows; the model carries the effect of quarter-end users running 5 percent above or below the average, as a sensitivity and not a bound, and of rounding in the published share and user figures, which moves each count by about a million. The reverse valuation is in the equity form: free cash flow after interest, operating cash flow less purchases of property and equipment over the four quarters to September 2026, set against the market capitalization, the November 3 close times the shares outstanding on the quarterly report's cover, solved for the growth that ten years at that rate followed by 2.5 percent a year forever would need to justify the price, at discount rates of 8 to 12 percent, once before stock compensation and once after it, with the enterprise form run as a cross-check and the year 2026 on the company's fourth-quarter guidance run as a check of the base. The adult part of the summer's checked gain applies the letters' adult share of checked users to the checked counts, with the published percentages rounded as ranges; the payer rate by age that would account for the payer share's hold is an assumption tried at several values, since Roblox gives none. The chat figures are the letter's own growth rates divided one by another, and its 1.6 billion daily messages divided by daily users. The split of the gap between the paths reruns the paths with one flow at a time swapped, averaged over both orders. A check is treated as permanent, though Roblox re-checks some accounts whose behavior does not fit their stated age. Daily payers are bookings divided by the days in the quarter and by the stated average daily bookings per daily paying user, which reproduces the 10-Qs' rounded figures in every quarter; the fourth quarter of 2025 is derived from the full year less the first nine months and marked so; the 2025 quarterly user counts for the year-earlier comparisons come from Roblox's printed growth rates with the roundings carried as ranges. The two scenario paths use constant conversion and lapse rates chosen so that both disclosed transitions can be rebuilt with no flow below zero, treating each daily user as an account, and stated in the text, new-user inflows of 13 and 9 percent of daily users a quarter with three in four checking on arrival (assumptions, not fits), a payer share recovering to the 2025 average or held at the September 2026 level, spending per payer held at June 2026's $10.42, and the summer lift removed from the September count by the 2025 third-to-fourth-quarter user ratio, our choice; they are illustrations and forecast nothing. The frozen test's bands, words and dates are constants in the program with a fingerprint that stops it if any is changed; the thresholds were tested against the bookings record of 2024 through the fourth-quarter guidance before freezing, and one was narrowed: the void for an algorithm change now requires Roblox to name such a change as a main driver of the scored quarter, because the broader wording would have voided the test at birth. The test was frozen on October 31, 2026 and re-frozen the same evening, after the fact-check, to remove a contradiction in its words about a missing payer figure, to add the second quarter of 2025 to the void's record, and to record Russia beside the score; no band, date or verdict moved, and the first freeze's fingerprint is kept in the program and reproduced from the second. The reading of November 6, 2026 is a note beside the test and changes nothing in it. Amendment 1, dated November 11, 2026 and printed after the test, changes the verdict rule: the checked count becomes an early indicator, bookings and payers alone decide with the frozen lines and payer rule, and the verdicts are named strong rebound, weak growth and cannot tell; its text has a fingerprint of its own, printed with it, and the program stops if a character of it or of the frozen test moves. Bookings are confirmed quarter by quarter as revenue plus the change in deferred revenue plus other, and the legal settlement expenses appear only in the reconciliation of Adjusted EBITDA. The history of the company and the controversy rests on Roblox's filings, including its Forms 10-K for 2023 and 2024, on its newsroom and on the published record named in Sources, with the founding year and Erik Cassel's part taken from secondary sources because the annual report, which calls David Baszucki its founder, states neither; its figures are quoted from those filings, not computed, and every allegation in it is reported as its maker's, with Roblox's position beside it. The historical facts about the Comics Code were checked against the sources named below, and the comics sales and title counts carried from a single secondary source are marked as such in the text. The share prices in "The price of a conversation", "Two dates to watch" and "What the price assumes" are not from Roblox's filings: they are daily closes from Massive market data, the same closes printed on public market sites, and no other figure from any licensed data feed is used anywhere. The calculations are in model.py and charts.py.
The audience was young. Roblox's annual reports put 42 percent of its daily users under 13 in 2023 and about 40 percent in 2024. Those were ages the users had typed in for themselves, which is why Roblox stopped reporting the figure when it began checking faces. The early data in the letter for the fourth quarter of 2025 showed a younger base than users had been reporting about themselves: of those checked, 35 percent were under 13, 38 percent were 13 to 17 and 27 percent were adults. The same letter said that Roblox "was initially created to serve the younger demographic" and is today "the largest dedicated gaming platform for users aged 13 and under."
Roblox rejected Hindenburg's report the same day, in a release headed "Roblox Refutes Misleading Claims in Hindenburg Report," and said in a statement: "The financial claims made by Hindenburg are misleading. The authors are short sellers and have an agenda irrespective of the substance of Roblox's business model and results."
In the main the states and counties allege, under state consumer protection laws, that Roblox misrepresented the safety of its platform and failed to protect children. The quarterly report says the Federal Trade Commission, Pennsylvania and Utah notified Roblox in October 2026 that they are considering claims of their own. In December 2025 the cases families had filed in federal courts were gathered before a single judge in California for pretrial proceedings, and Roblox and the other defendants have said they will ask that many of them be sent to arbitration; the quarterly report's statement that Roblox will defend itself vigorously covers these suits as well as the states'. None of these claims against Roblox has been proven in court, and we say nothing here about how any of them will or should come out.
The product changes came in steps. In November 2024 Roblox gave parents new controls and set new limits on how children under 13 can communicate. In July 2025 it announced age estimation. And in November 2025 it announced the gate this inquiry studies, describing itself as "the first online gaming platform to require facial age checks to access chat." Roblox presented the gate as a new safety standard for its industry, and it came while the suits were multiplying.
Britain's Online Safety Act carries fines of up to 10 percent of worldwide revenue or £18 million, whichever is greater. The legal settlement expenses concern what Roblox's investor materials call youth-related consumer protection and digital safety matters with states and territories.
None of what follows is a recommendation. Each is a hypothesis with the evidence beside it and the fact that would kill it.
Roblox, the count against the share. The hypothesis, bolder than the paths themselves, is that Roblox is on one of the two and not on a road between them, and that the checked count will give the first public reading of which in the February 2027 letter, six months before the test can. A count and a share that both rise is what the reset path predicts, and a road between the paths would show it too; it is the lower path alone that pulls the two apart. Falsified if the February count is above 88 million and the August 2027 verdict is not strong rebound, or if the count is below 88 million and the verdict is not weak growth; the first is what a road halfway between the two paths would produce, and the second what a December dip followed by bookings up 10 percent or more, with payers not falling, would produce. A February count inside the assumed date-basis sensitivity around 88 million will be scored this way but reported as inside the measurement uncertainty. Untested if Roblox stops reporting the share.
Roblox, chat as the canary. The hypothesis is that the company's own measure of users who chat or join voice will grow at least as fast as the checked count from the fourth quarter of 2026 through the second of 2027, which would say that checked activity is turning into conversation, whatever lies behind its growth. In the third quarter that measure grew 24 percent, a little slower than the checked count's 26 percent and twice the whole base's 12; the gap is inside the roundings of the published figures, so the third quarter reads as neither a pass nor a fail. The measure counts daily accounts that use text or voice. In-game text messages rose 13 percent while accounts using text or voice rose 24 percent. That gives us a lower ratio of text messages to the combined population, with the amount each text sender wrote still unknown. It can tell whether checked users take up chat, not whether conversation recovers, the company's own stated aim, and it cannot say which path, since chat outgrows the base on both paths because the share rises on both. Falsified if chatting users grow more slowly than the checked count, by more than the roundings and the date basis allow, in two consecutive quarters from the fourth quarter of 2026 on. Untested if Roblox never reports the chat figure again.
The transitions themselves. Roblox discloses enough to estimate two pools of daily activity each quarter and none of the five flows between them: how many holdouts converted, how many newcomers arrived and checked, how many checked accounts lapsed. A single table in a shareholder letter, new users by whether they checked on arrival and existing users by whether they checked in the quarter, would turn the question this inquiry can only frame into one it could answer.
Accounts, not logins. Every user figure Roblox publishes is activity averaged over a period. A count of accounts, or of unique people, by check status, even once a year, would say how much of a rise in checked activity was new faces and how much was old faces on more days.
Churn. Roblox gives no figure for the share of its active accounts that stop playing in a quarter. The 27-month life of a paying user is the only anchor the filings offer, and payers are likely to stay longer than the average player.
The regions. The quarterly report says hours and bookings are tracked by region, and the revenue table splits the world into four blocks with Australia and New Zealand folded into Asia-Pacific. The three countries where the gate went up a month early are the natural early test for the whole question; Roblox has given their age-check rates, and nothing public separates their users or bookings.
The payers by check status. Nothing Roblox has filed says whether a checked user is more or less likely to spend than an unchecked one. The payer share is an average over both pools; if each paid at a fixed rate, two quarters of data would solve for the two rates, and we ran that arithmetic on the spring and on the summer. The answers contradict each other, so the rates moved, for reasons the letters themselves name, and the question stays open.
The supplemental materials. Roblox posts additional quarterly materials on its investor site, and we read both the quarter's investor data sheet and the 41-page supplemental deck. Neither holds a checked count or a quarter-end user figure; every user figure in them is a quarterly average, so they do not fix the date basis. The deck repeats the definition the 10-Qs give: a daily active user is an account that logged in on a calendar day, averaged over the period, and the company notes that "Age-checked metrics are not comparable to historical periods that relied on self-reported data."
Likely, we can get some of it from primary research, information gathered by asking the people who know first hand, which is what Vista is all about.
Three kinds of people have seen what Roblox's filings cannot show.
We would ask, as a matter of general past practice: when a platform adds friction to a core social feature, what share of the affected cohort returns within two to four quarters, and do the verified users spend at the old rate? We would also ask how much of a summer's rise in daily activity is the same accounts logging in more often, and how a platform tells that from new arrivals.
The answer that would change our view: "A third or more of the affected cohort never came back and the verified users spent less than before." That would say Roblox's lower payer count is a lasting new path. "Most came back within four quarters at the old rate" would say it is a one-time reset.
We would ask, in general: what completion rates do mandatory age checks reach at one month, three months and a year, and which users never complete them?
The answer that would change our view: "Completion leveled off near two thirds within two quarters and the users who never complete are mostly low spenders." That would say Roblox's 64 percent is close to a ceiling, and a ceiling of holdouts who were never going to pay is a reset in all but name. "Completion keeps climbing for a year and many non-completers are regular payers" would say the drop is still unfolding.
We would ask, as a matter of general past practice: after the age check requirement, what happened over the following quarters to players who had not checked compared with those who had: did they return, convert or drift away, and did checked players pay at the old rate?
The answer that would change our view: "Unchecked players drifted away within a few quarters and checked players spent less than before." That would say the lower path is lasting. "Unchecked players kept playing without chat and spent almost nothing, and checked players spent at the old rate" would say it is a reset.
Every conversation would draw on past, general experience only: no current employees of Roblox, nothing confidential, and every Advisor, as Vista calls the experts it interviews, screened for conflicts before a word is said. For a client engagement, Vista would combine this inquiry with those interviews: two or three Advisors, structured written conversations, and one brief that says where they agree, where they split, and what would change the answer.
Because the share rises in two ways: when unchecked activity turns checked, and when unchecked activity simply stops coming. In two futures we built for Roblox, one growing and one shrinking, the checked share of daily activity reaches 78.3 and 80.0 percent by June 2027, yet the two sit about $514 million apart in a single quarter's bookings. The shrinking company reports the higher share. The futures are illustrations built on stated assumptions, not forecasts.
No. Multiplying the checked share by average daily users gives an estimate of checked activity. From March to June 2026 the estimated checked count rose less than three million while the unchecked count fell almost twelve million. In the summer the checked count rose from about 70 million to 88 million while the whole platform gained 15 million. These are estimates of daily activity, not counts of people, and the two inputs are measured over different windows.
From Roblox's report for the June 2027 quarter, expected around August 2027. Under the test frozen on October 31, 2026 and amended on November 11, a strong rebound needs bookings growth of at least 20 percent over the $1,557 million of the June 2026 quarter and an increase in daily paying users from the estimated 1.642 million; weak growth means bookings growth below 10 percent or a decline in daily payers; anything else is cannot tell. As our judgment, not a computation, and conditional on a scorable result: about 25 percent strong rebound, 45 percent weak growth and 30 percent cannot tell.
This inquiry is research, not investment advice. It states what the filings and models show under labeled assumptions; nothing here is a recommendation to buy, sell or hold any security, and the decision belongs to the reader. Russ W. Rosenzweig, Vista's founder, owns shares of Meta Platforms and holds no options on any company named here; he owns no shares of Roblox, Snap or Reddit. Holdings through mutual funds and exchange-traded funds are not counted. How this inquiry was made: Written with the help of AI tools. The facts were gathered from Roblox's shareholder letters, filings and investor materials on sec.gov and its investor site, from the regulators' and archives' own pages, from the public record of Roblox's history and of the child-safety controversy named in Sources, and from the published record of the Comics Code; the Markov chain, the payer figures and the frozen test were built and every figure computed in code; each quotation was checked against its source; and the draft was fact-checked, given two outside reviews and read for plainness before I read it. I read it and listened to it before it was published. Why I write with AI. Why I publish these inquiries.