Game-First Loop — Multiplayer Reference

Status: living doc. Owner: Oleg + Mariana. Audience: public — we publish our launch thinking in the open.

Bet 1 says we launch a game, not a platform. This doc answers the next question it forces: which game? It captures the trends actually shaping multiplayer right now (2026), sorts the winners by loop archetype, grades each archetype against our specific constraints, and lands on the loop we should ship for the alpha. Numbers are sourced and approximate; re-verify before quoting externally. Sources are listed at the end.

The filter (our constraints decide the loop)

A loop only counts if it survives all five of these:

1. Browser, no download, playable in ~5 seconds. Frictionless to enter and frictionless to leave. The loop has to be obviously worth it inside the first 60 seconds (the .io lesson).
2. Fun solo, with zero human-made content. At launch the world is ~30% AI-seeded. One person who logs in when nobody else is online must still have something to do (the Minecraft lesson).
3. Uses what we are uniquely good at. A deep, fast authoring toolset — terrain sculpting, modular snap-kits, scatter, and (to build) one-click assembly and blueprints — lets people assemble real structures by hand in seconds. AI generation is an optional accelerant on top, not the core, because on-demand generation is slow (minutes) and inconsistent. A loop that does not lean on building/assembly wastes our only real edge.
4. Persistent and shared. What you do has to stay, and it has to be visible to the next person, so individual effort becomes everyone else's discovery (the three-layer model: shared ground, unique deposits, personalized path).
5. Alive at 50, not just at 50,000. The atomic network is small. The loop cannot require a packed lobby to be fun. Daily-habit and idle hooks matter more than raw concurrency.

Trends shaping the moment (2026)

This is the part to read closely. Six trends are bending the market right now, and each one moves a lever we care about.

Trend 1 — Winner-take-more on a flat engagement base

Newzoo's 2026 PC and Console report frames the year as an "inflection point"; the sharper read is winner-take-more on flat-to-shrinking total engagement hours, even as revenue grows. A small set of evergreen, live-service titles (3+ years old) commands a disproportionate share of playtime, revenue, and culture, and is actively squeezing mid-tier AA games into downsizing and closure (Niko Partners). Two consequences for us:

• "Gateway theory is broken." Roblox and Minecraft players are not graduating to AAA. The audience matures inside the platform it started on. Whoever owns someone's first shared-world habit tends to keep them. That is an argument for being a destination, not a feature.
• Attention is the scarce resource, not players. We are not fighting for installs; we are fighting for the slot in someone's week. That rewards daily-habit loops over big-but-rare sessions.

Trend 2 — UGC went from a feature to a revenue pillar

UGC is the clearest growth engine in the industry, and the framing has shifted from "UGC as a feature" to "UGC as a revenue generator" (Niko Partners). Roblox keeps scaling, Fortnite Creative is maturing, and live-service titles are bolting on creator tools plus revenue share (Genshin Impact added a UGC mode with monetization plans). Modding is in a parallel boom: CurseForge mod downloads grew 38% to ~33B, Mod.io grew 56% to ~920M, and Overwolf creator payouts hit ~$300M (+25% YoY), nearly Fortnite-scale (Naavik State of UGC 2026).

The cautionary half: no Web3-native UGC platform has reached meaningful scale, and most mobile social/party UGC games that once showed promise have regressed. Network effects plus heavy technical investment make this a brutal category to enter and to fundraise in. This validates our exact sequencing: earn the game first, turn on the creator economy second. The money is real, but it accrues to platforms that already retain people.

Trend 3 — The generative-AI fork, and why our side of it is the durable one

This is the trend most specific to us, and it broke wide open in January 2026 when Google rolled out Project Genie (Genie 3) to AI Ultra subscribers: an 11B-parameter world model that generates interactive 3D worlds from a text or image prompt at 720p/24fps in real time. Gaming stocks dropped on the news (Roblox ~-6.6%, Nintendo ~-8.9%, CD Projekt ~-9.2%). The market read it as existential.

Then everyone read the fine print, and it is the single most useful data point in this doc. Genie worlds last 10-60 seconds before coherence degrades. There is no persistent state, no inventory, no NPC logic. It generates the appearance and feel of a world, not a functioning game. Take-Two's president put it flatly: "Genie is not a game engine." This is the essay's "lonely dead end" made literal: a private, ephemeral, single-player world with no memory and no neighbors.

The generative-AI-in-games space is splitting into three camps:

Camp	Examples	What it produces	Durability
World models	Google Genie 3, Decart Oasis	Whole real-time worlds from a prompt; ephemeral, single-player, no state	Spectacular demo, not a game; the lonely dead end
Asset + engine generation	Roblox Cube (1.8M+ objects made), World Labs Marble	Objects/structures that drop into a persistent, shared engine	Production-grade today; this is our camp
Video generation	ByteDance Seedance 2.0	Cinematics, not interactive	Adjacent, not a game loop

Cinevva sits squarely in the asset-plus-persistent-shared-engine camp, which is the one shipping real retention. Roblox's agentic Studio (single-prompt-to-playable, with a playtest loop) is the closest analog and the proof the camp works. Our wedge writes itself: while world models hand each person a beautiful world that vanishes alone, we drop AI-built structures into one shared world that stays and that other people walk into. The broader pro sentiment is also a constraint to respect: over half of industry professionals view genAI's impact negatively, so we lean on "AI builds the parts, humans design the fun" rather than "AI replaces the game."

Trend 4 — Two loop shapes are eating concurrency: cozy idle and meme-chaos

The two biggest concurrency stories in living memory are at opposite emotional poles, and both are instructive.

• Cozy idle (Grow a Garden). Built by a teenager in days, it peaked above 20M concurrent and beat Fortnite's all-time record. The loop is plant, harvest, upgrade, repeat, and crucially progress continues while you are offline, which manufactures the "log in, check, bounce" daily habit. The cozy segment overall is a real, growing market (~$1.1B in 2025, ~6% CAGR), skews adults 25-44 and ~60% women, and is driven by stress relief, not competition. "Players are not paying for graphics, they are paying for a feeling."
• Meme-chaos (Steal a Brainrot). Launched May 2025, built in ~4 months, hit ~25.8M concurrent (the highest CCU ever recorded for any game) and 68B visits by May 2026, with a single-day record of 605M visits during staged "admin war" events. The mechanic is an old tycoon loop (buy income-generating characters) plus a steal/raid layer. It did not win on mechanical novelty; it won because the steal produced emotional reactions worth filming (clips of kids reacting became organic ads and implicit tutorials), it rode the Italian-brainrot meme cycle (itself AI-generated content, now in IP litigation), and it staged spectacle events before competitors could respond.

The shared lesson under the opposite tones: a self-contained loop that fits a 10-minute window, plus a reason the session produces something shareable, beats mechanical depth. Cozy gives us the retention pattern we can actually adopt; meme-chaos gives us the virality pattern (build something worth filming) without the pay-to-win and tonal baggage we should avoid.

Trend 5 — Distribution is consolidating onto social surfaces and instant play

The no-download advantage is compounding because the surfaces that host instant play are getting huge and social. Discord opened Activities (iframe games via the Embedded App SDK, launchable in a voice channel with no install) to all developers, and the social-proof numbers are striking: 72% of users play games with friends weekly on Discord, 40% launch a game within an hour of watching a friend stream, players play 6x longer with one friend present and 8x with three, and Instant Play Quests can drive up to 4x higher new-user trial rates. The classic .io channel still works too (Agar.io hit ~5M DAU within weeks, Slither.io 68M mobile + 67M browser, Krunker 200M+, and 2026's Redcoats.io runs up to 1,000 players in one browser battle). For a browser-native product, this is the tailwind: our weakest point (no-download means low commitment) is offset by living where the social graph and the one-click trial already are.

Trend 6 — Onboarding and cold-start are now close to a solved science

The playbooks have converged, and they map almost one-to-one onto our alpha:

• Time-to-fun under 30-60 seconds. Auto-start the loop, defer sign-in and menus, show the core mechanic in the first 10 seconds. D1 benchmarks: ~30-40% casual, 40-60% mid-core, 60%+ for live-service with an existing community (we have one).
• Beat the empty lobby with a lobby/session list, not strict matchmaking. A lobby-list architecture is viable at ~10 concurrent; global skill-based matchmaking needs ~1,000+ or the queue dies. Fill gaps with clearly labeled bots, never deceptive ones.
• Seed the empty state. The explicit, current best practice is to pre-populate content in empty states to show community value immediately. That is independent confirmation of our ~30% AI-seed bet, with the same honesty caveat (label it, convert it).
• Invite friends late, after the first win. Premature social prompts feel intrusive; post-milestone invites lifted D1 ~21% and social invites 4x in a documented case. Async social (gifting, visiting) is a cheap retention loop.
• Discovery now rewards retention and growth, not raw counts. Roblox's algorithm rewards D1/D7, 7-day play-days, and invite-driven co-play, and caps the reward past ~60 min/day. Fortnite's Creator Economy 2.0 pays from a pool (40% of Item Shop net revenue; ~$352M to creators in 2024 from engagement payout alone), now adds direct In-Island Transactions (creators keep 100% through Jan 31, 2027), and explicitly downranks copycats while rewarding new-player acquisition (75% of a new player's contributions for six months). The whole industry is repricing around the exact metric we already chose as our north-star: retention.

What is actually winning right now (2026 scan)

Game	Loop archetype	Scale signal	The transferable lesson
Grow a Garden (Roblox)	Cozy idle build/tend	Peaked >20M concurrent, beat Fortnite's record; built by a teen in days	Progress continues while you are away, so the habit is "log in, check, bounce." Calm beats flashy; frequency beats session length.
Steal a Brainrot (Roblox)	Tycoon + light-PvP heist	~25.8M peak concurrent (highest CCU ever), 68B visits	Old tycoon loop plus a steal layer that produces clips worth filming. Virality came from emotion, not mechanics.
Brookhaven RP (Roblox)	Social hangout	~400-700K concurrent, 45-60+ min sessions	"An empty city full of friends beats a packed dungeon full of strangers." Social density is the retention engine.
Adopt Me! (Roblox)	Collect + trade economy	6-year economy, top lifetime earner	A trading layer between players turns other people into content.
99 Nights in the Forest (Roblox)	Co-op survival	~200K concurrent breakout	Co-op survival against a shared threat manufactures collision and shared stakes.
Tower of Hell (Roblox)	Skill obby	~55-400K concurrent	Retention from self-improvement, not reward drops. Replayable with zero new content.
Valheim (PC)	Co-op survival + base building	12M copies, 2-10 players	"The base you build together is why groups stick around." Building is the anchor, combat is the pretext.
Deep Rock Galactic (PC)	Co-op objective runs	10M copies, 48h avg playtime	Short, repeatable objective missions in procedural levels keep a tight crew returning.
Romestead (PC, 2026 EA)	Co-op survival + village mgmt	Polished EA launch	Blueprint-place-build with a crafting mini-game; recruit NPCs to automate. Building is the hook, not the dungeon.
Agar.io / Slither.io / Krunker / Redcoats.io	.io instant arena	5M DAU in weeks / 68M+ / 200M+ / up to 1,000 in one browser battle	One-click, no account, instant fun. The moat is distribution itself; commitment is shallow by design.
Skribbl / Gartic Phone / Artbitrator	Party game, private room	Artbitrator: 1-12 players, AI judges	Share a link, you have a lobby. AI as a judge/host is now a normal, expected mechanic.
Townscaper	Solo creative toy	~380K sales on screenshots alone	Beautiful building people share unprompted is its own acquisition channel.
Project Genie (Google)	AI world model	Crashed gaming stocks Jan 2026; worlds last 10-60s	The anti-pattern: a gorgeous private world with no persistence, state, or neighbors. Our wedge.

Archetypes graded for fit

Archetype	Exemplars	Why it works	Fit for Cinevva World	Verdict
Cozy idle build/tend	Grow a Garden, Townscaper, Animal Crossing	Solo-fun, daily habit via offline progress, calm, screenshot-worthy, no lobby needed	Leans directly on AI building + asset deposits; works at 50 users; persistent by nature; a real growing market	Core
Co-op settle-the-frontier	Valheim, Romestead, Deep Rock, 99 Nights	Shared goal manufactures collision; building together is the stick; solo-viable	Matches the Founders' Quest in the Alpha Playbook; building is the verb we are best at	Core (layer on top)
Social hangout	Brookhaven	Highest session length on Roblox; density is the product	Needs a crowd to feel alive; dangerous at 50; but the shared ground is exactly our meeting-place layer	Secondary (emerges, don't lead)
Tycoon + light heist	Steal a Brainrot, Adopt Me!	Self-contained sessions, viral, player-vs-player content without raids	Tone risk; PvP/steal undercuts "founding citizens build a place"; pay-to-win baggage; revisit post-PMF	Park (borrow the virality, not the mechanic)
.io instant arena	Agar.io, Krunker, Redcoats.io	Best acquisition channel that exists; ten-second fun	Shallow commitment; ignores building and assets entirely	Borrow the onboarding, not the genre
AI world model	Project Genie, Oasis	Stunning prompt-to-world demo	Ephemeral, single-player, no persistence; the dead end we define ourselves against	Anti-pattern (position against it)
Pure creative tool	Dreams, Project Spark	Deep ceiling for the few	The exact thing the graveyard killed; tool-first has no daily habit	Never lead with this

Recommendation: a cozy-frontier build loop (the Founders' Quest, specified)

Ship the Core archetypes fused: a cozy, persistent build-and-tend loop (Grow a Garden's offline-progress habit engine, Townscaper's share-ability) wrapped in a settle-the-frontier objective chain (Valheim/Romestead's "build the place together"). This is the most trend-supported shape available and the only one that clears all five constraints. Concretely, the first session, with the cold-start tactics from Trend 6 baked in:

1. Arrive in the shared world at a visibly inhabited frontier (the ~30% AI seed plus team showcase plots, labeled honestly). Not an empty plot, not a tech demo. This is the "seed the empty state" best practice.
2. Claim a plot near the frontier edge in seconds, with no sign-in gate yet (time-to-fun under 60s; defer the account).
3. Assemble your first structure in under 60 seconds from snap-kits and one-click build (terrain + props + scatter; AI is an optional helper, not the path, since generation takes minutes). It stays. It is now everyone else's discovery (the deposit layer), and ideally it is screenshot-worthy enough to film (the meme-chaos virality lesson, minus the chaos).
4. Follow a personalized objective chain toward the next landmark (the path layer): a solo-viable reason to be here on a Tuesday when no one else is online.
5. A persistent/idle hook gives a reason to return tomorrow (Grow a Garden): something grows, completes, or is waiting when you come back.
6. When 2+ are online, route them to the same place via a lobby/session list, not strict matchmaking (Trend 6), filling gaps with clearly labeled NPCs/bots. Density emerges instead of being required.
7. Offer the friend invite only after the first build/landmark, never on arrival, and lean on instant-play surfaces (a shareable link, eventually a Discord Activity) for trial.

This keeps us a game first (a loop with a goal, solo-fun, retention-driven) while every action quietly feeds the platform second (persistent creator deposits accreting into a shared place). It is the asset-plus-persistent-engine camp of Trend 3, deliberately built as the opposite of the world-model dead end.

For how this loop maps onto what we have actually shipped (and the gap that Bet 1 is still 100% unbuilt in code), see the Build Plan doc. Short version: the engine already does shared multiplayer building and persistence, and we commit to one shared world with earned, adjacency-gated territory claims (the world_plots scaffold already exists). The missing pieces are an objective tracker, a minimal claim system, and packaging the assembly tools into a sub-60-second first build.

Copy precisely / avoid

Copy: Grow a Garden's offline-progress daily habit and calm tone. Townscaper's one-click auto-build (driven WFC over a marching-cubes tileset — the literal "no skill, instant, beautiful" assembly we should adopt) and its "beautiful enough to screenshot unprompted." Valheim's snap-build conventions and "the base is why they stay." Fortnite/UEFN/Sims/LEGO blueprints, prefabs, and galleries (save a build, re-place it; ship starter prefabs). The .io one-click, no-account entry. Steal a Brainrot's "make the session produce something worth filming," without its steal mechanic. The 2026 cold-start consensus: time-to-fun <60s, seed the empty state, lobby-list over matchmaking, labeled bots, late friend invites. Live where instant play already is (browser link first, Discord Activity later).

Avoid: Putting AI generation on the critical path of the first build — it takes minutes and the quality is inconsistent, so lead with hand assembly (terrain, snap-kits, scatter, one-click build) and keep generation async/curated. Leading with PvP/steal mechanics (tone clash with founding-citizen framing; pay-to-win baggage). Leading with a creative tool and no game (the graveyard). Optimizing everyone into creation (rebuilds Dreams; honor the dual funnel). Any loop that needs a full lobby to be fun (we launch at 50). Marketing ourselves as a "world model" (that is the ephemeral dead end; we are the persistent-shared-world opposite). Over-claiming AI given the negative pro sentiment; frame it as "you assemble it, AI optionally helps."

Open questions to settle in alpha

1. Does the cozy/idle hook actually pull unprompted week-2 returns at 50 users, or do we need a stronger co-op objective to manufacture the collision?
2. Which assembly tool clears the 60-second creator aha most reliably at the hands of a newcomer — one-click auto-build, snap-kits, or a starter blueprint they tweak?
3. Does AI-seeded frontier content read as worth-exploring (Townscaper-beautiful) or as filler (Decentraland-empty)? Track visit/interaction on seeded vs human plots, per the synthetic-supply discipline in Metrics.
4. Does the persistent build produce shareable artifacts on its own (the Steal-a-Brainrot virality test, minus the chaos), or do we have to manufacture the moment?

None of this is proven. It is the best trend-supported starting bet, and it is falsifiable against the same retention-curve north-star as everything else. If the cohort does not flatten, the loop is wrong, and we change it.

Sources

• Macro / UGC: Newzoo PC & Console Report 2026 and Deconstructor of Fun's "Inflection Point" walkthrough; Niko Partners 2026 predictions ("UGC as a revenue generator," evergreen squeeze); Naavik State of UGC 2026 (modding payouts, CurseForge/Mod.io/Overwolf); Taylor Wessing 2026 UGC analysis.
• Generative AI in games: Google DeepMind Project Genie / Genie 3 blog; TechCrunch and WeTheFlywheel coverage (10-60s worlds, stock drops, "not a game engine"); Roblox Cube / agentic Studio writeups.
• Cozy + meme loops: QY Research / IndexBox cozy-market sizing; Quantic Foundry demographics; Gameflip / GameDaily / The Click on Grow a Garden; Wikipedia / Lootbar / GamesBeat / PCMag on Steal a Brainrot and the brainrot IP dispute.
• Distribution: Discord "Build Where the World Plays" and Activities docs; Sacra Discord data (6x/8x friend playtime, 4x trial lift); CrazyGames / Viverse on .io scale; Redcoats.io.
• Onboarding / cold-start / creator economy: 2026 onboarding-retention playbooks (time-to-fun, labeled bots, seed-the-empty-state, late invites); lobby-list vs matchmaking analysis; Epic "Engagement Payout" and "In-Island Transactions" docs; Deconstructor of Fun on Roblox discovery metrics.