A roadmap is a promise with a date on it — Roblox's RDC26 post commits to prompt-authored NPC behaviour, generated scene layouts and install-free play in a browser, all three dated Late 2026 (Roblox, 11 Sep 2026)

Why it matters (the medium, not the myth). The Story Sweep asks what the medium can newly do, and the honest answer from this document is nothing yet — which is worth carding precisely because the temptation is to read a roadmap as a capability. What the document actually establishes is that a UGC platform which reports "123 million users" playing daily has committed, in its own voice and on a published date, to four things that would change what a story-maker can attempt there: character behaviour authored by prompt and reusable as an object, scene layouts generated rather than placed, distribution by bare link with no install, and worlds that persist what participants make inside them. Each carries a delivery window. None carries a shipped date.

That distinction is the card. This garden's own lesson ledger records it as L44, written after a spec revision was mistaken for a shipped release: "a version string is an intention until the tag exists." A roadmap bullet is the same object with a friendlier face. So this card claims a dated commitment and the platform metrics printed alongside it, and it claims no capability at all.

The four items that matter to the medium, quoted whole — five bullets, because the offline commitment is carried separately. Posted to the Roblox Developer Forum by the staff account skywise84 (NickT), timestamped "September 11, 2026, 10:30pm" on the face of the post and 2026-09-11T22:30:06+00:00 in its metadata, filed under the section its metadata names as og:article:section: Announcements, tagged rdc and featured, headlined "RDC26: What We Announced".

• "NPC Dynamic Behavior (Late 2026): Populate games with dynamic NPCs through prompts in Studio. You can edit, version, tune, constrain, and reuse it across as many other NPCs as you want."
• "Scene Generation (Late 2026): Prototype faster by generating a layout of a scene."
• "Play on the Web (Late 2026): Distribute your game on the web. Share your game's Details Page URL in a YouTube description, on X, or in a Discord, and anyone who clicks can start playing immediately in Chrome, with no app or install required." and, as a separate bullet, "Play offline (Mid 2027): Enable offline solo play alongside a live, multiplayer mode."
• "In-Game Creation Persistence (Early 2027): Persist assets that were created within their game, similar to how game data can be saved to a data store."

Three more are worth recording because they touch how a participant is met rather than what is built: "Early testing (Late 2026): Invite up to 10 age-checked friends of similar ages to test your game early, before you publish."; "Developer Push Notification Support (Late 2026): Turn on push notifications for opted-in players to alert when it's their turn, enabling new cases like async turn-based play."; and "New Default Movement (Early 2027): Use new default avatar movement that includes motion matching and root motion."

Three further Late-2026 commitments are quoted here because the falsifier below tests them and a claim should not make its first appearance in a falsifier: "Asset Moderation (Late 2026): Get instant, detailed moderation feedback on assets during development so you can fix issues before publishing."; "Player Support (Late 2026): View and manage support requests submitted by your players."; and "Observability Platform (Late 2026): Trace client sessions and logs, get instant troubleshooting insights from Studio Build, and ship fixes directly inside Studio and Creator Hub."

The post's own framing of its completeness, quoted because it caps every claim above: "What we revealed at RDC is just a portion of the roadmap we plan to deliver over the next year. We'll share a full list of updates on our roadmap next week."

The prompt item is the one with a genuinely new shape, and the reason is the second sentence. Prompt-to-content on this platform is not new to the cabinet — roblox-build-prompt-to-publish-alpha (verified, 30 July 2026) records it arriving for whole games, and quotes Roblox's own cap on the claim, which travels here rather than being paraphrased away: a single prompt building a full editable 3D scene is "coming soon", not shipped. What is different in this bullet is that the output is described as a manipulable, versioned, constrained, reusable object — "edit, version, tune, constrain, and reuse it" — rather than a one-shot generation. A character's behaviour that can be version-controlled and constrained is closer to a script than to a prompt result, and constraint is the word a story-maker should notice: an NPC that can be bounded is an NPC that can be trusted to hold a beat. Whether the shipped feature does any of that is exactly what the falsifier tests, because today this is one sentence in a bullet list.

The platform metrics, quoted, with their own footnotes. *"Because of your hard work, on average, last quarter, 123 million users (up 10% year-over-year)1 came to play on Roblox every single day. Game visits are also up 55% from last year2, and creators have earned approximately $1.7 billion through DevEx, up about 50% since last year3. And it's not just the established hits carrying the platform: in August, 28% of games in the top 100 by hours, and 33% of games in the top 1000 were launched in the preceding 90 days.4"*

The footnotes do not line up with each other and the card says so rather than smoothing it. Footnote 1 reads "As of the quarter ending June 30, 2026". Footnote 2 reads "Last twelve months ("LTM") ended June 30, 2026 vs. LTM ended June 20, 2025" — the two end-dates are a year and ten days apart, so the comparison is offset by ten days from a clean year-over-year and the two twelve-month windows are therefore not the same span; that is the platform's own text, not a transcription slip here. Footnote 3 reads "Creator earnings through DevEx during the last twelve months ended June 20, 2026". And footnote 4 defines "New" as "the top 10/100/1000 games by hours engaged for the month of August 2026 and launched in the 90 days before the month start", while the sentence it annotates cites only the top 100 and top 1000 — the top-10 figure is defined and not printed.

A discrepancy between the live text and a reply's quotation of it, recorded rather than resolved. A reply on the same page, from the account catinpsyop (astriddd), timestamped "September 11, 2026, 10:54pm" — twenty-four minutes after the post — contains two quote blocks attributed to NickT. The first reads "Roblox Everywhere (Late 2026): Distribute your game as a standalone app and on the web. Share your game's Details Page URL in a YouTube description, on X, or in a Discord, and anyone who clicks can start playing immediately in Chrome, with no app or install required." The second reads "Next year, you'll be able to enable offline solo play."

Three differences from the live text, stated precisely: the feature is named "Roblox Everywhere" in the reply and "Play on the Web" in the live text; the phrase "as a standalone app and" is in the reply and absent from the live text; and the offline commitment is "Next year" in the reply and "Mid 2027" in the live text. The two documents agree that offline play is a separate statement.

Two explanations fit and this card does not choose between them: the post was edited after publication, or the reply's quote blocks do not reproduce what the post said. The observation is recorded because the difference is not cosmetic — "standalone app" is a distribution claim the live text no longer makes, and the reply's own commentary turns on that phrase ("'standalone app' is interesting"). Nothing on this card rests on the quoted variant; the live text governs throughout.

A provenance question this card's tier does not price, named rather than buried. If the post was edited after publication, then the text read on 15 September is not the text published on 11 September, and every quotation above is dated to the reading rather than to the publication. The tier is verified because the primary is a dated, first-party, staff-posted document and the claim made — that Roblox committed to these items — is fully evidenced by it. But a reader should know that on this particular page the garden has direct evidence of at least one textual variant in circulation, and that the discipline of quoting from the face of a document assumes the face holds still.

What is stated versus what is measured. Every roadmap item is a stated intention with a delivery window and no shipped date. The five footnoted metrics are stated past facts, unaudited and company-published. The publishing changes are stated too, with one number attached to a forecast rather than a result: "Adjusting our approach to content violations, changes we expect to reduce bans by nearly 40% for well-intentioned creators" — expect, and "well-intentioned" is undefined on the page. The concrete, checkable commitment in that section is the clause "we're lowering the Highly Engaged Player threshold from 250 to 100 this November", and it is the earliest dated commitment anywhere on the page.

Derived figures, enumerated. Three are computed from the primary, and nothing else on this card is arithmetic on the primary's figures. Two other categories of number appear on this card and are named rather than enumerated: counts of the primary's own contents (the eleven quoted roadmap bullets, its four footnotes, the three differences from the live text, the two explanations offered for them), which are stated inline at the point of use; and counts about this run and this cabinet, which follow.

1. The draw-distance claim reconciles with its own numbers: the page says "Minimum draw distance to 500 studs" gives "more than double today's draw distance of 200" — 500 ÷ 200 is 2.5, so "more than double" is the page's own conservative phrasing of its own figures, recorded rather than restated as a finding.
2. The reply's timestamp is twenty-four minutes after the post's, 10:54pm against 10:30pm, both read off the page.
3. The earliest dated-window roadmap item is roughly three and a half months out from the post's date, taking "Late 2026" at its latest as 31 December 2026; the latest, "Mid 2027", is roughly nine to ten months out. Both intervals inherit the page's imprecision and neither is a forecast. Note that the November publishing commitment quoted above is earlier than either, at roughly two and a half months, and that one item on the page — "Pre-roll video ad formats (Coming soon)" — carries no window at all.

Counts about this run and this cabinet, computed here and named as such: footnote 2's two window end-dates are a year and ten days apart; the falsifier below has five limbs, of which four can fail with the primary unchanged; this is the second first-party page in this run's window whose text is in question, the other being DeepSeek's changelog on deepseek-v4-pro-deprecation-reversed, where a sentence sits under a date it cannot have been written under — a different shape from this one, since no earlier dated record of that page's text exists either; and a grep of content/signals/*.md for roadmap, re-run after this card was finished rather than extended from memory (L84), returns seven files: deepseek-v4-1-flash-routes-v4-pro-traffic, eu-ai-act-digital-omnibus-in-force, mcp-2026-roadmap-agent-identity, visa-stablecoin-platform-open-usd, this card, roblox-wallet-replaces-devex-usd-payouts and steam-frame-reservations-open — the last two of which match only because they cross-reference this card's slug. No priority claim about the cabinet is made on this card.

So we should, for The Call (a design constraint, not canon). The item worth designing against is "Play on the Web", because its shape is the closest on this list to the work's own. The Call's threshold — an email returns a permanent, never-reused number — is owned canon and is labelled as canon in this clause, not a verified fact, and it is named here only because the structural point depends on it: a threshold whose cost of crossing is one click and no install is a different object from one that requires an app. Roblox proposes to make that true for whole worlds by Late 2026. The useful move now is to write the beat that a bare link could carry and keep it platform-agnostic, so that if this ships the work can meet it and if it slips the work loses nothing.

The second constraint is narrower and lands on a card the cabinet already holds. roblox-animation-packs-creator-authored-locomotion (verified, 9 September 2026) concluded, in its own words pasted here before this card leans on them, that a work must decide "whether a participant arrives in their own body language or in the work's — because on this platform that is now a setting the work controls, and the default hands it to the participant." "New Default Movement (Early 2027)" adds a third state to that binary: the platform's own default movement is itself described as changing, to "motion matching and root motion." The bullet's verb is "Use", which is consistent with a creator-side option as well as with a platform-wide flip, and this card does not decide which. Either way, a beat that reads how someone walks may in early 2027 be reading a different baseline than the one it was designed against, whether or not the participant bought anything. Nothing here is a claim about The Call's story.

What would prove it wrong (falsifier). Five limbs; the first is a provenance re-check and is labelled as one (L83), and four can fail with the primary unchanged.

• The facts — provenance re-check, not a test of the claim — fail if the devforum post at the cited URL does not carry the 11 September 2026 timestamp, the skywise84 staff attribution, the eleven quoted roadmap bullets with their delivery windows, the five footnoted metrics (123 million, 55%, $1.7 billion, 28%, 33%) with their four footnotes, or the "just a portion of the roadmap" sentence.
• The calendar — the whole card fails, in the direction that matters, if the items dated "Late 2026" are not shipped by 31 December 2026. The ones quoted above are NPC Dynamic Behavior, Scene Generation, Play on the Web, Early testing, Developer Push Notification Support, Asset Moderation, Player Support and the Observability Platform. Observable in Roblox's own Announcements and release-notes categories, which is where this candidate was found.
• The object claim — that prompt-authored NPC behaviour becomes a reusable, constrainable thing rather than a one-shot generation fails if the shipped feature produces per-NPC output that cannot be versioned, constrained or reused across NPCs. This is the limb the medium actually turns on, and the evidence for it today is one sentence.
• The distribution claim fails if Play on the Web ships gated — behind an account, a region, a device class, or a creator opt-in that most experiences do not take. The bullet promises a bare URL in Chrome and names no condition; the cabinet has already seen this platform ship a capability and its off-switch in the same post, which is the roblox-animation-packs-creator-authored-locomotion shape.
• The publishing change fails if the Highly Engaged Player threshold is not lowered from 250 to 100 in November 2026. This is the cheapest check on the card and the first one due.

Provenance (what I actually read). One document, fetched from the main thread on 15 September 2026: devforum.roblox.com/t/rdc26-what-we-announced/4865880, posted by the staff account skywise84 (NickT), timestamped "September 11, 2026, 10:30pm" on the face of the post, 2026-09-11T22:30:06+00:00 in its metadata, with og:article:section: Announcements in the same metadata block.

Every quotation attributed to the primary on this card comes from that page, enumerated: the headline; the metrics paragraph and its four footnotes; the "lowering the Highly Engaged Player threshold" clause and the content-violations bullet; the eleven roadmap bullets quoted above (NPC Dynamic Behavior, Scene Generation, Play on the Web, Play offline, In-Game Creation Persistence, Early testing, Developer Push Notification Support, New Default Movement, Asset Moderation, Player Support, Observability Platform); the "just a portion of the roadmap" closing; the draw-distance bullet; the "Pre-roll video ad formats (Coming soon)" label; and the catinpsyop reply with its two quote blocks and its "'standalone app' is interesting" fragment, which is labelled throughout as a reply rather than a Roblox statement. The post's key-takeaway block was read and is not quoted, and is named here outside the enumeration for that reason.

Quotations from other cabinet cards, two in total: one from roblox-build-prompt-to-publish-alpha (its "coming soon" cap) and one from roblox-animation-packs-creator-authored-locomotion (its action line). One quotation from this garden's lesson ledger, L44 in tasks.lessons.md. Each is named inline at the point of use and comes from those files, not from this page. The Call is named once, in the action line and in the so_we_should paragraph, and is labelled as owned canon in the same clause on both occasions (L75c); nothing about The Call is asserted as a verified fact anywhere on this card.

Not opened and not sourced to: the about.roblox.com newsroom post this page links ("RDC 2026: The World Needs More Play"), the "Making publishing easier" post, the terrain-updates post, the Roblox Wallet and Card post — carded separately as roblox-wallet-replaces-devex-usd-payouts from its own primary, which this page links to — the Restricted Content Policy, the embedded YouTube video, and the roadmap the post says will follow "next week." The single embedded image on the page is not a source for any figure here. The date was read off the document (L71).