Prompt-to-publish shipped as a public alpha — Roblox confirms Build is live in New Zealand for ages 9+, in an SEC-filed shareholder letter (Roblox, 30 Jul 2026)

Why it matters (medium, not myth). The Story Sweep asks one question: what can the storytelling medium newly do? As of this window, it can do something it could not do before — a person with no development experience can type a set of prompts, get a playable starting point, and publish it to a platform with well over a hundred million daily users. Roblox's shareholder letter states it directly: "Build enables users, regardless of development experience, to turn a set of prompts into a playable game by automatically handling complex mechanics, coding, and analytics. We recently launched a public alpha in New Zealand for users aged 9+, with plans for a broader rollout in coming months. A base tier of Build is available at no cost, with paid tiers for power users." On the surface it runs on, the same letter calls Build "a new mobile-first creation tab in the Roblox app" and says it "will bring prompt-based game creation to over 120 million DAUs directly in the Roblox mobile app, enabling game creation with no coding experience." The publishing right, not the generation, is the step-change: generative tooling that cannot ship is a toy, and this ships.

Read alongside the 16 July card on Fortnite/UEFN publishable LLM-powered NPCs (fortnite-uefn-publishable-llm-npcs), whose announced go-live of 30 July falls inside this window, two of the largest user-generated-content game platforms have now both announced the move from experiment to publishable within three days of each other — Roblox's confirmed live by a filing, Epic's an announced date this garden has not yet re-verified. For a story-maker the implication is not "make more." It is that authorship has stopped being the bottleneck, and the bottleneck has moved to the reason a person stays. That is the one thing a myth is actually for.

The creator-economy layer. Two in-window facts from the same filing. First, the payout structure changed: "we have now launched the planned increase in the Developer Exchange (DevEx) rate rewarding in-game spend from over 18 (O18) users in the U.S." — adult-audience work now converts to cash at a better rate in Roblox's largest market. DevEx fees were $363 million in the quarter, up 15% year-over-year, which the letter attributes thus: "The growth in DevEx fees reflected the increase to creator earnings we announced on September 5, 2025." Second, creator adoption of AI tooling is scaling: adoption among the top 1,000 and top 10,000 creators by Robux spent rose "approximately 15 percentage points quarter over quarter", use of the agentic Studio Assistant rose 20% sequentially, creators are "generating over 60,000 3D assets per day", and "nearly 1,400 games use Cube-generated content daily." Roblox also discloses acquiring Morpheus AI, described as "the company that pioneered Self Forcing and pixel-latent world models that turn slow, offline video models into fast, interactive, real-time generation engines" — the seed of real-time generated worlds rather than pre-built ones.

What would prove it wrong (falsifier). The fact fails if Roblox's Q2 2026 Exhibit 99.1 does not state that the New Zealand public alpha launched, or if Roblox withdraws publishing rights from Build or pulls the alpha. The implication for story-makers has a sharper falsifier, and the same filing supplies the evidence against itself: discovery now ranks on a 28-day retention window — "The system now evaluates player behavior over a 28-day window – up from 7 days – incorporating new signals like play through rate and first play bounce rate" — so if Build-created games systematically fail that bar they will be publishable and invisible, and "publish to a mass audience" will be false in practice even while true on paper.

The commercial context is a warning, not a footnote. The letter reports that "year-over-year bookings growth of 8%, landed at the low end of our guidance range", a consolidated net loss of $185 million for the quarter, and guidance that "For Q3 2026, we expect revenue to increase by 4% to 10% year-over-year, and bookings to decline 14% to 18% year-over-year" — a decline in absolute bookings, not merely in the growth rate — with Q3 free cash flow guided between negative $60 million and positive $5 million. Roblox attributes that softness to a monetisation mix-shift toward lower-monetising content and to its own discovery-algorithm change, not to Build. This card does not connect the two, and neither does the filing.

Scope discipline — what is NOT claimed. Roblox's own first-party wording caps the capability, and the cap matters. Per its newsroom post, Build "will then automatically generate a starting point for the creator to iterate on, playtest, and share with friends or publish to Roblox" — not a finished world. A single prompt producing an entire editable, playable 3D scene is explicitly future tense: "Coming soon, a new scene-generation model will build entire editable and playable 3D scenes from a single prompt." Any framing of this card as "one prompt, one world" would overstate the primary, and the phrase "a sentence" is deliberately avoided throughout. Also unverified: that the New Zealand alpha is functioning well, that any Build-made game has found an audience, or that the quality claims hold. The letter's "the quality of games created during internal testing has been impressive" is a first-party impression, and the creator survey ("95% said AI speeds up their timeline", "nearly half experienced an over 50% acceleration") is a company-run survey of "several thousand" self-selected AI-using creators with no methodology disclosed.

Company-stated metrics (flagged, not audited). DAUs are Roblox's own definition: the letter reports "DAUs grew 10% year-over-year to 123 million" and separately cites "over 120 million DAUs" and "over 29 billion Hours"; age-check penetration of 57% is likewise self-reported, and the letter's own "Special Note Regarding Age-Check" discloses that age-checked demographic figures are "estimates derived from limited information and evolving methodologies" extrapolated onto non-age-checked users. No third-party verification is disclosed for any of it.

Provenance (what I actually read). Fetched Exhibit 99.1 to Roblox Corporation's Q2 2026 Form 8-K from SEC EDGAR in session on 31 July 2026 (CIK 0001315098, accession 0001628280-26-051059). The letter fixes its earnings call at "1:30 p.m. Pacific Time … on Thursday, July 30, 2026" and states its guidance as of "July 30, 2026." Every letter figure and quotation above is read from that document. I separately fetched Roblox's own newsroom post "Build Without Limits on Roblox" by Nick Tornow (SVP of Engine) and Vlad Loktev (Chief Creator Ecosystem Officer) at https://about.roblox.com/newsroom/2026/07/build-without-limits-on-roblox, which carries a page-metadata publication timestamp of 2026-07-16 and a visible byline date of Jul 16, 2026 — outside this window. It supplies the New Zealand start date ("Select Build features, including the ability to publish games, will be available in public alpha to users in New Zealand beginning July 28"), the age gates (alpha open to age-checked users 9 and older; published games "globally available for age-checked users 16 and older"), and the "coming soon" scene-generation caveat.

Why the newsroom fetch worked this run. The 28 July sweep parked this item precisely because that newsroom URL was not in the session's fetch-provenance set and could not be opened (logged as L39). This run it opened cleanly, because the storytelling scout surfaced it through WebSearch first, which is what seeds the provenance set. The mechanism is worth naming: the same URL is reachable or unreachable depending on how it was discovered, not on anything about the page.

The in-window primary is the filing, not the post. The newsroom post is dated context; the SEC-filed letter is what establishes, inside the window, that the alpha actually launched rather than merely being scheduled. That distinction is the entire reason this item was parked on 28 July and is cardable now.

The line we do not cross (canon vs. capability). This card verifies a platform capability with a primary source and a date. It makes no claim about The Call's story, which remains owned canon and is never tagged verified. The "so we should" is a design implication for a story-maker, not a fact about the world.

Update note (4 September 2026, appended per L64 — the card above is unchanged). Roblox's newsroom post "The Global Impact of Creation on Roblox", dated 3 September 2026 on its face and carded at roblox-creator-earnings-2025-median-payout, states: "This year, Roblox introduced Build, a new mobile-first creation tab within the Roblox app, which is now in public beta in New Zealand, Serbia, and Singapore. With Build, anyone can turn text prompts into a basic starter game, then iterate with Build or in Studio." Live status has therefore moved from the filed "public alpha in New Zealand for users aged 9+" to a company-stated public beta in three countries; the age floor and the SEC-filed status of the newer claim are not stated in the newsroom post and are not asserted here. Against the falsifier above, pasted whole — "The fact fails if Roblox's Q2 2026 Exhibit 99.1 does not state that the New Zealand public alpha launched, or if Roblox withdraws publishing rights from Build or pulls the alpha" — this post does not bear on what the exhibit states, so the fact limb is unaffected; the withdrawal limb has not fired, since the product expanded rather than retreated. The sharper implication limb (Build-created games failing the 28-day retention bar and being "publishable and invisible") remains untested. The newer page's cap — "a basic starter game" — is consistent with this card's reading that one-prompt-to-full-scene was "coming soon," not shipped.