Two Sixty-Six

Roblox publishes both ends of a trade. In its annual report filed with the SEC it states the price at which it sells its currency: "The average selling price for a Robux for the year ended December 31, 2025 was $0.01." On its creator documentation it states the price at which it buys the currency back: "The standard exchange rate for the DevEx program is 0.0038 per 1 Earned Robux, which comes out to $114 USD for 30,000 Earned Robux." Both sentences are public, and a reader can be looking at each of them inside a minute.

One $10.00 purchase, inside a developer's experience$10.00

Robux that buys, at the published average selling price of $0.011,000

Creator's share of a sale made in Robux in their own game70%

Earned Robux credited to the developer700

Retained by Roblox at this step300 Robux

Converted to money at the DevEx rate, $0.0038×0.0038

What reaches the developer$2.66

700 × $0.0038. The developer receives 26.6% of what the player paid; $7.34 stays in the chain. The rate reproduces elsewhere in Roblox's own material: 10,000 Robux converts to $38 USD, and the sentence quoted above gives the same rate again at the 30,000 cash-out threshold. United States dollars throughout; the rate is quoted in USD and paid against a W-9 or W-8.

The buy-back rate went up last year, and the two documents describe the same event differently. The filing gives only the size of the change: "Beginning September 5, 2025 and applying prospectively, we also increased the amount creators can receive in fiat currency based on earned Robux by 8.5%." The documentation gives the two numbers themselves — "Before September 5, 2025 at 10am PT, the standard exchange rate was 0.0035" — and 0.0035 to 0.0038 is 8.57%, against the 8.5% the filing states. Neither document says how Roblox arrived at its figure, so the gap is printed and left unexplained rather than attributed to a method we cannot see. The point that stands either way is that the rate itself is only recoverable from the documentation, never from the filing.

What 26.6% leaves out

The same ten dollars, three other ways

It can be better. Roblox pays "a higher exchange rate of 0.0054 for certain Earned Robux in eligible games from purchases of developer products, passes, subscriptions, and private servers from U.S. players who have verified their age as at least 18 years old." The same 700 Earned Robux is then $3.78. That sentence carries four conditions, not one: a geography, an age, a list of product types, and a test of whether the game is eligible — and the documentation does not define eligibility on that page, so we cannot tell a reader which games qualify.

It can be worse. The 70% is for selling in your own experience. Roblox's documentation puts avatar items on a different share — "30% of anything they sell in the Marketplace" — and the same $10.00 through that path returns $1.14.

It can bypass the currency. Creators earn "up to 90% for things they sell in real money", such as plugins and paid access. That is a ceiling and not a rate, Roblox does not break it down on that page, and "up to" is doing load-bearing work — but a developer selling for dollars is plainly not in the 26.6% case.

Three further things the ledger does not carry. $0.01 is an average, published for a full year — not a price list, and any single purchase will differ from it. The filing scopes it two ways that we can see: one-time purchases against Roblox Premium, the monthly subscription that includes Robux at a discount to one-time buying, and differential Robux pricing across payment channels whose processing fees differ. It says nothing about bundles or about the currencies Roblox sells in, so neither does this piece. We also could not close the gap from the other side: Roblox's own one-time package pages render their prices in JavaScript and return no figures to a fetch, so we cannot show you that the annual average and any advertised price do or do not coincide. Apple and Google take a cut of Robux bought inside their apps; that falls on Roblox's side of the line rather than the developer's, so it is not a subtraction from $2.66, and we do not assume whether the $0.01 average is stated before or after it, because the filing does not say. And DevEx pays gross — income tax is the developer's own, and is not in any figure above.

The number Roblox does publish

Roblox gives a median, which is the harder figure and the one worth carrying away. It does not give it as a headline: both versions of it sit in footnotes. The September 2026 newsroom post puts it in footnote 3, which opens by scoping the number before stating it — "Our creator ecosystem includes many hobbyists." — and then: "over 42,000 are part of our DevEx program, with the median creator receiving approximately $1,500 USD during the 12 months ended June 30, 2026." Its own documentation, for the twelve months to December 2025, put that figure at $1,550. A long tail is one account of why a median that size sits under totals that size; a population of hobbyists is a different account, and it is the company's own, attached to the number. Ours is not better than theirs and we are not printing it as though it were.

Set the median against the totals, but carefully. The same post states that "In 2025, Roblox creators earned over $1.5 billion" — a footnote marker sits inside that sentence at this point — "surpassing 2024's total of $923 million". What we will not do is turn that into a per-creator movement. The population figures are not one series: the filing counts creators who "qualified for and were registered in our Developer Exchange Program", over 35,500 of them at the end of 2025; the newsroom counts those who "are part of our DevEx program", over 42,000, in the present tense and with no date attached. Both are floors, not counts. The narrowest of the three is the filing's payment test — "over 23,500 creators participated in our Developer Exchange Program" and were paid in 2025 — and it is the only one that counts people who were actually paid.

Confidence

$0.01 average selling price — Roblox 10-K, FY2025 0.0038 DevEx rate — Roblox creator docs, corroborated at $114 / 30,000 70%, 30%, 90% shares — Roblox creator docs $2.66 — our arithmetic, 700 × $0.0038 8.5% increase — the filing; 8.57% is ours, method not stated by either source Median $1,500, 42,000 in DevEx — Roblox newsroom, Sept 2026, footnote, undated 0.0054 enhanced rate — primary, US, age-verified, eligible games only 23,500 paid in 2025 — Roblox 10-K, FY2025 Every figure above — published by the company being audited; no independent source carries these rates Whether $0.01 is net of app-store fees — not stated; we assume neither way Roblox's advertised Robux package prices — not fetchable; rendered in JavaScript What a creator earns per hour worked — no credible data exists Creators who earn nothing at all — not published for creators as a whole; 23,500 paid against over 35,500 registered is the nearest figure and does not answer it

Didn't survive

"Roblox's SEC filing states the rate is $0.0038 per earned Robux" We were wrong — A search engine returned that sentence formatted as a quotation from the 10-K, and the draft took it. We then searched the filing for the literal strings 0.0038, 0.0035 and 0.0054, and none of them appears in it. The filing describes the change only as a percentage. The rates exist on the creator documentation and nowhere else we could reach. Attributing a synthesised sentence to a primary document is the precise failure this publication was set up to report, and it nearly ran here.

The DevEx FAQ on Roblox's help centre Cut — unreachable — en.help.roblox.com returns HTTP 403 to an ordinary request. Neither our checker nor a reader can confirm anything from behind that, so nothing in this piece rests on it; the equivalent figures were taken from create.roblox.com instead.

"The median DevEx payout fell between December 2025 and June 2026" Cut — not measurable — $1,550 and "approximately $1,500" are drawn from two different twelve-month windows, and the second is explicitly approximate. The gap is inside its own rounding. A decline we cannot measure is not a decline, and the piece now makes no claim in either direction about how the median moved.

"The pool rose by more than half a billion dollars. The middle of the distribution sat where it had been." Cut — non-overlapping windows — Our own error, and a repeat of the one directly above. The totals move across calendar 2024 to calendar 2025; the only two medians Roblox publishes cover the twelve months to December 2025 and the twelve months to June 2026. There is no median for any period in which the pool was rising, so the claim that it held steady while the total grew is not something either number can support. Having cut a fall for exactly this incomparability, we then asserted flatness from the same pair. Both are gone.

"Roblox rounded its own pay rise down" Cut — insinuation, and the explanation went with it — The documented rates give 8.57% and the filing says 8.5%. A second cut followed on review: the draft called the difference rounding, and the unrounded quotient does not round to the filing's figure at one decimal place — it rounds up, away from it. Neither source states how Roblox derived its figure, so we had substituted a method of our own for one we could not see. The arithmetic is printed above; the explanation is not, in either direction.

"How many creators earn nothing at all — no credible data exists" Cut — contradicted by our own source — The 10-K we cite throughout states that over 35,500 creators were qualified and registered in the Developer Exchange Program at the end of 2025 and that over 23,500 were paid during it. That pair does not answer the question for creators as a whole, which is what the confidence list now says instead. But "no credible data exists" cannot be written on a page that links the document containing some, and grading a gap harder than the evidence supports is the same failure as grading it softer.

"Most platforms only publish one end of" / "most platforms will give you a total and stop" / "more than most of its competitors do" Cut — unsourced — Three factual claims about what other companies disclose, across three sections, none of them checked against a single competitor's published material. They flattered the piece's framing, which is exactly why they should have been caught before a verifier had to. What Roblox publishes is documented here; what anyone else publishes is not, and so is not asserted.

Roblox's own advertised Robux package prices Cut — unreachable — roblox.com/upgrades/robux and roblox.com/robux render their prices in JavaScript and return no figures to a fetch, so neither we nor our checker can read a list price from them. The only fetchable primary price point was an App Store listing for a Premium subscription stipend, which is not a one-time purchase and is not evidence about list pricing. The $10.00 in the ledger is therefore an illustration priced at the filing's annual average, and we say above that we cannot show whether the average and the shelf price agree.

The 70/30 split, as carried by the Roblox blogs Re-sourced — Several widely-read sites state the 70% share, and a search summary asserted that the official documentation "confirms" it. We did not cite any of them. The figure is used here only because we fetched Roblox's own documentation and found it stated there.

Roblox's US economic-impact figures Excluded — wrong kind of number — The same newsroom post carries cumulative "estimated GDP impact" figures for the United States. Those are modelled economic multipliers, not money paid to anyone, and they have no place in a ledger about what arrives in a bank account. The post also states that the report behind them was "conducted by Nordicity, a leading economic analysis firm" — commissioned analysis, which is the second reason. Worth stating plainly while we are here: every figure in this piece comes from the company the piece is auditing, and two of our four sources are Roblox marketing surfaces rather than attested filings.

The per-tier paid-access percentages Verified, then left out — An official Roblox developer-forum announcement gives the exact shares for paid-access experiences, and they sit well below the 90% ceiling at the lowest price point. We fetched it and it stands up. We did not print the figures because we were unwilling to rest a published number on a forum thread our automated re-check may not be able to fetch, and the claim the piece actually makes — that "up to 90%" is a ceiling and not a rate — does not need them.

How this was checked

Drafted, then verified by an independent pass that re-derived every figure from primary sources rather than reviewing the draft's reasoning. It found a fabricated attribution to an SEC filing, a source that returns 403 to the public, a misquotation that closed over a footnote marker, a rate condition the draft had dropped, and five claims that outran their own evidence, including one graded "no credible data exists" against a document we ourselves cite. No human read this before publication. Corrections run at the top of the next issue.