Marketing Brief
The one page anyone writing about Trade Clash should read first — a post, a clip caption, a press reply, a deck. It supersedes the auction-era social memo, now in legacy/internal.
Last updated 30 July 2026.
1. The product, in one sentence
Trade Clash is a real-time AI agent battle arena. You build an agent, it fights other players' agents on a real RTS engine, and the crowd watches and bets the outcome.
Say it in that order. "AI agent battle arena" is the category; "build, own, watch" is what a person does; everything else is detail.
2. Why anyone should care right now
People already want to watch AI systems compete — the "Claude vs ChatGPT, who'd win" instinct is everywhere and nothing satisfies it. Trade Clash makes it literal and watchable: two agents that strangers designed, fighting a real-time strategy match with commentary and a live probability, on a loop that never stops.
The interesting inversion is what the skill is. You never touch the controls. You are not competing on execution — you are competing on whether your model of a fight was correct, and you find out in public. That is a different feeling from playing an RTS, and it is the thing to sell.
3. Audience
Primary
Builders. People who would rather design a system than execute one — Factorio, Zachtronics, fantasy sports, bot-jam and AI-tinkering crowds.
"No code. Five dials and plain English, and it fights without you."
Secondary
Watchers. Twitch/esports viewers who like a board to read and a number to argue with.
"Two AIs, a live win probability, and a crowd pricing it."
Tertiary
Collectors. Crypto-native, but the pitch is record, not speculation.
"Your agent's configuration hash and its record live on-chain."
Primary is the one that compounds — a builder who ships one agent gives the watchers something new to watch. Optimise the funnel for build, not for views.
4. Message hierarchy
Use this order. Cut from the bottom, never the top.
Hook (0–3s): two machines fighting. The arena never sleeps.
What it is: a real-time AI agent battle arena.
How you play: no code — five dials and plain-English standing orders.
Why it's fair: the server decides the winner, matches are seeded and deterministic, results anchor on Base.
Money: points only.
5. The copy system — three strings, one job each
Mirrored across the product, the site and these docs. Don't invent a fourth.
Descriptor
Real-time AI agent battle arena
titles, meta, first line of any post
Verbs
BUILD · OWN · WATCH
kickers, bullets, thumbnails
Tagline
THE ARENA NEVER SLEEPS
end cards, mood beats, the closer
Everything a player builds is an AGENT. Not a bot, not a fighter, not a war-agent.
6. Proof points — all true today, all checkable
Reach for these instead of adjectives.
A real RTS engine in a browser — an OpenRA fork compiled to WebAssembly, not a toy simulation.
Five dials and twelve order types compile to ~80 real engine parameters. The Forge shows you the compiled file before you commit.
Orders the compiler doesn't understand are reported back as ignored — the product tells you when it didn't understand you instead of pretending.
The win probability is measured, not vibes. Out-of-fold, split by match: log loss 0.513, Brier 0.175, accuracy 72.3%, AUC 0.831, ECE 0.099.
We publish where the model was wrong. Before calibration it scored worse than answering 50% every time while being 72% accurate. Fixing that is a better story than hiding it.
Seeded and deterministic — same seed plus same configurations replays the same match.
The server derives the winner. A client-reported result is never accepted, which is the whole reason the book can be trusted.
Six blocs, four archetypes, and a 30-second trailer that exists and is cut.
7. Voice
Trader-sharp but clear. Confident, short, present tense. Honesty beats hype wherever trust is at stake — this product's edge is that it shows its work.
Never:
Name-call the audience. "Degen" is banned from all copy.
Gambling language: wager, luck, house edge.
Imply real money, cash-out, returns, or yield.
Promise dates, or describe a gated capability as though it ships.
Quote the model as more accurate than the measured numbers.
Present the earlier products' figures as the arena's own numbers. See "Track record" below for the framing that is approved.
Cryptic stat runs. "169 kills · 26 structures", never "169k 26b".
Money copy is never playful, and it is stated plainly exactly once per surface: points only — one gas-free signature, never a transaction.
8. Track record — how to use it
The studio has shipped two earlier Trade Clash titles, and their numbers are real and verifiable. They are usable, and they should be used — a first-time reader has no other reason to believe this team ships. What is not allowed is letting them read as the arena's traction.
Approved sentence, use verbatim:
Seroto Labs' earlier Trade Clash titles connected 100,000 wallets across four tournaments and took 1st place for New Consumer App at Base's On-Chain Summer Awards.
Also true and usable: $2,000 USDC distributed to top players across those four tournaments.
The rules around it:
Always attribute to "earlier Trade Clash titles" or "Seroto Labs" — never to the arena, and never in a sentence whose subject is the current product.
Never imply the arena has users yet. It has no public build.
Keep it in the credibility slot — a footer, an about line, a deck's team page. It is never the hook.
The award links to its source. Keep the link; an unlinked award reads as a claim.
Full history: Tournament History and Previous Game Concepts.
9. Constraints that shape the plan
State these plainly to anyone building a campaign.
There is no public build yet. Every CTA points at the community, the docs, or the trailer. Never "play now".
The engine's current art is EA-derived, so paid media and a press push wait for the original-asset remake. The decision on record is to build attention with content first and swap the art in when there is traffic to justify it — that is a deliberate trade, not an oversight, and it means organic/social is the only channel until the remake lands.
Real value is gated. Audit → determinism receipts → legal → KYC → custody. None are cleared, so "when real money?" is answered not planned. See Roadmap.
Footage quality is bounded by bot behaviour. How dense a clash looks depends on the agents' squad sizes, and arena pairing is random. Don't write a line the picture can't honour — if the copy says "forty units in the blender", the footage has to show forty units.
10. Assets on hand
30-second trailer (cut, with real music and narration) · poster frame showing the live broadcast UI · six bloc leader portraits · the shared globe · broadcast stills with the plans overlay and the market card · the Forge spec sheet.
Missing and worth making: short vertical cutdowns of single clashes, a Forge screen-capture that shows an order compiling, and one clip of a win probability swinging hard mid-match — the most legible "this is a real system" moment we have.
11. Channels
X
Clash clips and probability swings. Short, no preamble, the picture does the work.
Telegram
The community. Where builders show agents and argue about doctrine.
Vertical video
Trailer cutdowns and single-fight clips. Design for muted viewing — legible text, no reliance on audio.
These docs
Credibility. When someone asks "is this real", the architecture and win-probability pages are the answer.
tradeclash.xyz
The one-screen version of this brief.
12. What success looks like
Ranked. The first one is the only one that matters early.
Build-to-build retention — someone who builds one agent builds a second.
Watch-through on the live rotation.
Share of standing orders that compile rather than coming back ignored — a direct read on whether the build experience is teachable.
Agents forked from another player's design — the sign that designs, not just matches, have become the content.
Vanity metrics (impressions, follower count) are diagnostics, not goals.
Last updated