Comparable Launches — Evidence File
A reference file, condensed from the deep research behind the public essay. Numbers are sourced; treat as approximate and re-verify before quoting externally.
Failures (root cause + lesson)
Meta Horizon Worlds. Billions spent (Reality Labs ~$6.02B operating loss in one quarter, early 2026). Leaked memos: only ~9% of user worlds ever visited by 50+ people; "an empty world is a sad world"; only researchers could survey ~514 active users; ~1 in 10 users returned after a month; MAU target cut from 500k to 200k; admitted no product-market fit; "quality lockdown." VR support shut down June 2026, pivoted to mobile. Lesson: capital cannot fill an empty world; content density is not buyable.
Decentraland / The Sandbox. Oct 2022: DappRadar showed Decentraland ~38 DAU, Sandbox ~522 DAU (unique on-chain wallets), each ~$1.3B market cap. Projects countered with ~56,697 monthly logins / ~8,000 daily (Decentraland) and 39k daily in an alpha (Sandbox). Both metrics measured the wrong thing; the gap between valuation and real engagement is the point. (Roblox in the same window: 52M DAU vs 11.3M monthly payers — most people in free worlds rarely transact.) Lesson: a valuation/token cap is not an audience; pick metrics that measure fun, not transactions.
Dreams (Media Molecule). Adored creation tool. Live service ended Sept 2023; no sustainable business model. No creator monetization, promised in-game store never shipped, PS4-only (no PC, no native PS5 at launch). Co-founder Mark Healey's biggest regret: not focusing on multiplayer; concluded the IP "works better as a game than a creative tool." Lesson: pay creators, ship multiplayer, be a game first. This is the most important single data point we have.
Crayta. Collaborative UGC platform, shut down March 2023, users lost their creations. Launched on Google Stadia; died when Stadia died. Meta-owned (acquired 2021) and still killed. Lesson: never build on a platform that can evict you.
Project Spark / Sansar / High Fidelity. Microsoft's UGC maker shut down; Linden Lab's Sansar sold off; Rosedale's High Fidelity pivoted away from worlds. Lesson: a powerful tool with no reason to return has no daily habit.
Rec Room. 150M+ lifetime players, shut down June 1 2026, never reached sustained profitability. Lesson: reach is not a business; engagement that doesn't convert to a model still dies.
Survivors (the actual unlock)
Minecraft. Procedural survival = fun for one player with zero user content, infinitely. The mod/server/map ecosystem grew on top. 350M+ copies. Unlock: never have an empty world; the game generates its own content.
Roblox. 2006 beta: ~100 people, peak concurrency 30-40. ~12-year slow burn to 80M+ DAU. Single move that mattered: let the community build the games. But economics are brutal even now (operating losses, ~6% of users buy currency). Most users only play. Unlock: be a place to play first; the creator catalog and economy are the decade-long second act.
.io browser games (Agar.io, Slither.io, Krunker). Agar.io ~5M daily players within weeks; Slither.io 68M mobile + 67M browser plays; Krunker 200M+ players. Moat is distribution itself: a link loads a game, no install. One-click onboarding, no account. Unlock + caveat: no-download virality is real but commitment is shallow; survival depends on instant fun.
Among Us. Launched 2018, near-dead for two years, nearly abandoned. Exploded Sept 2020 via a streamer ladder (Sodapoppin). Early base: itch.io feature + Korean audience. Unlock: survive long enough to be discovered; breakout can come from an unplanned channel years later.
Figma. ~3 years stealth, launched free, charged 2 years later. Aha moment: real-time multiplayer. Removed the 2-collaborator free-tier cap that was blocking the very magic that sold it. Seeded via design influencers on Twitter. Adobe offered ~$20B (deal later terminated). Unlock: find the aha, then clear every barrier to it.
Townscaper. A toy with no goals; sold ~380k on the strength of screenshots people shared unprompted; developer's Twitter following jumped during dev. Unlock: unprompted sharing pre-launch is the realest validation.
VRChat. ~160k concurrent in 2026, growing while funded rivals (Horizon, Rec Room) folded around it. Community/creator-driven, not capital-driven. Unlock: engaged creator passion outlasts money.
Territory claiming in a shared world (successful + cautionary)
Minecraft Towny / Factions. Among the most-played multiplayer modes for over a decade. You claim land chunk-by-chunk on a shared persistent world; Towny layers towns, nations, taxes, and protected resident builds, Factions layers raidable conquest. Retention comes from social investment, economic progression, and the security of owned, protected assets. Design key (Towny docs): you "can't run rampant claiming massive amounts of land without working for it," and claims must be adjacent, which keeps the world dense. Unlock: territory claiming is a durable retention engine when it is earned and rides a game; claim adjacency prevents sprawl.
Reddit r/place. A shared pixel canvas where you place one pixel at a time and communities claim, defend, and contest territory. ~1M users in 2017; ~6M users and 72M pixels in 2022. Free, equal, social, about the process and community, not ownership-as-investment. Unlock: shared-canvas territory claiming has an enormous viral ceiling when it is open and social.
Ownership psychology. Psychological ownership plus the endowment effect (owners value what is theirs ~2x; losing it triggers loss aversion) plus the IKEA effect (you value what you built). Actionable rule: deposit ownership in the user's hands in the first session, and make it feel earned rather than gifted. Unlock: an earned claim, built in session one, recruits loss aversion to do the retention work.
Cautionary: Decentraland / The Sandbox (above) and Horizon Worlds (above). The same mechanic fails two specific ways: land-as-speculation (scarce investable parcels, billion-dollar valuations on dozens of daily users) and empty-plot sprawl (everyone gets an isolated plot; Horizon had ~9% of worlds ever visited by 50+). Lesson: earn claims by playing (never sell them as investments), gate by adjacency, and keep a high-bar shared meeting ground as the default destination.
Cross-cutting patterns
- The empty-room / cold-start problem is the boss fight. Everything else is a tutorial level.
- Be a game first, platform second. Earn the platform by first being worth using with an empty catalog.
- Retention-curve-flattening is the survival metric. Registered accounts and valuations predicted nothing.
- Marketplaces die on supply (~67%). Creator retention is the vital sign.
- No-download helps acquisition and hurts commitment. The first 60 seconds carry the whole weight.
- Pay creators or lose them; ship multiplayer; don't depend on a platform that can evict you.
Key sources
- Horizon Worlds memos: kotaku.com, fortune.com, ryanschultz.com, mixed-news.com, cnbc.com (2026 shutdown)
- Decentraland/Sandbox DAU: coindesk.com, blockworks.com
- Dreams: pushsquare.com, icon-era.com
- Crayta: gamedeveloper.com
- Rec Room / VRChat 2026: uploadvr.com
- Roblox: howtheygrow.co, matthewball.co/all/roblox2024
- Among Us: esports.net, howtomarketagame.com, businessofapps.com
- Figma: review.firstround.com, productgrowth.blog
- Townscaper: gamedeveloper.com, mcvuk.com
- .io games: news.viverse.com, megacool blog
- Retention/cold-start: solsten.io, newsletter.pmcurve.com, lennysnewsletter.com (atomic network), forkoff.xyz (marketplace cold start)