Sportwarren - Rec Football Gets Its Intelligence Layer

#devchallenge #githubchallenge

GitHub “Finish-Up-A-Thon” Challenge Submission

This is a submission for the GitHub Finish-Up-A-Thon Challenge

What I Built

SportWarren is a tactical command center for recreational football — the 5-a-side, 6-a-side, Sunday league kind that hundreds of millions of people play every week and that has, until now, left absolutely no trace.

You play every week. You score goals. You build chemistry with the same five people over years. And then the final whistle goes and none of it happened. No stats, no record, no reputation. Rec players are invisible.

SportWarren fixes that. Set your formation, put your squad on the pitch, share a challenge link to the group chat — and every matchday becomes verified stats, XP, and permanent squad history. Six attributes per player (pace, shooting, passing, dribbling, defending, physical) that actually evolve with every game. AI coaches that give you tactical banter and real analysis. Group verification via Telegram or WhatsApp so results are confirmed by the people who were there.

But the piece that makes this more than a stats tracker — and the piece that was entirely missing from the hackathon prototype — is the agentic commerce layer. There are two payable units baked into SportWarren now, both settled via x402 and USDC on GOAT Network's Bitcoin-secured infrastructure:

AI scouting reports (0.005 USDC) — an agent analyses your squad's attributes, form, and tactical tendencies and returns a structured breakdown
Match verification attestations (0.10 USDC) — once a result clears group confirmation via Telegram or WhatsApp, an agent produces a signed on-chain attestation of the verified outcome

The social layer is the integrity layer in SportWarren — group verification is what separates real stats from self-reported ones — and turning that verification into a purchasable on-chain attestation makes the commerce primitive inseparable from the core product mechanic. You're not paying for a feature. You're paying to make your match permanently true.

That's the Kite AI Agentic Commerce track in a sentence: rec football as an entry point into agent-native payments. Unglamorous use case, real economic primitives.

Built in collaboration with @udirobert · Find me on Farcaster · Lens

Demo

🔗 Live app: sportwarren.com
📦 Repo: github.com/udirobert/sportwarren
📱 Telegram bot: t.me/sportwarrenbot

The Comeback Story

The honest state of SportWarren before the Kite AI hackathon: everything was scattered. The tactics board existed but the challenge flow was broken. The player cards rendered but XP didn't update correctly after matches. The group verification system — the mechanic that makes stats trustworthy, not self-reported — was stubbed out. And the AI layer was a collection of disconnected pieces that didn't form a coherent loop.

Most critically: there was no agentic commerce layer at all. The verification system existed in fragments but wasn't connected to any on-chain settlement. The AI analysis ran and returned results into a void — no identity, no payment, no record.

The Kite AI Global Hackathon — and the support from GOAT Network's builder grants program — gave this a forcing function. Kite's agentic commerce track is specifically designed for agents that discover, pay, and manage via x402 and programmable constraints settled on-chain. The natural fit turned out to be verification: a squad logs a result, the group confirms it, and an agent produces a signed attestation someone pays for. That loop didn't exist before. Here's what actually got built and fixed:

The core loop stabilised. Formation setup → challenge link → opponent counter-pick → simulation → result logging → group verification. All five steps now work end-to-end without broken states or dead ends.

XP and player progression wired up correctly. Six attributes now actually update after every logged match. The progression curve — slower at higher levels, volatile early — is tuned to feel like a real career arc rather than a grind.

Group verification shipped. Squadmates can now confirm or dispute a logged result via Telegram or WhatsApp, with a 6-hour TTL, configurable confirmation thresholds, and silence-as-consent handling. This is what separates SportWarren from self-reported stat apps: the social layer is the integrity layer.

The simulation engine built out. The tactical AI runs opponent formations against each other using Monte Carlo probabilities, and delivers a readable breakdown — not just a scoreline but an explanation of why a 4-1 lost to a 1-2-1.

x402 payment integration on GOAT Network. Two payable endpoints now live: AI scouting reports at 0.005 USDC and match verification attestations at 0.10 USDC, both settled via x402 protocol through GOAT Network's facilitator. The economics are deliberately micro — this is about proving the primitive works at the right layer of the product, not extracting revenue.

Kite AI agent passport + ERC-8004 identity registered. The agent has its own verifiable on-chain identity, making it a first-class participant in the agentic economy rather than a backend function with a webhook.

The project went from a scattered prototype with broken flows to a working app with a genuine on-chain commerce layer. Rec football as a trojan horse for agent-native payments.

My Experience with GitHub Copilot

I used Copilot heavily on SportWarren — both inline autocomplete and Copilot Chat, often switching between them mid-session depending on what the problem needed. Alongside other AI tools in the repo (the codebase reflects a multi-tool workflow), Copilot was the constant thread.

Inline autocomplete earned its keep most on the repetitive but precision-sensitive work: building out the player attribute system, writing the XP calculation logic for different match outcomes, scaffolding the Telegram bot handlers. These are tasks where the pattern is clear but the details matter — getting the attribute deltas wrong would make the progression feel fake — and Copilot's suggestions were consistently close enough that reviewing them was faster than writing from scratch.

Copilot Chat was where the more interesting collaboration happened. The x402 integration on GOAT Network was genuinely new territory — x402 is a young protocol and there's limited prior art to draw from. Walking through the payment flow architecture in Chat, getting hypotheses about where the 402-challenge/response cycle should sit relative to the verification workflow, and debugging EIP-3009 permit handling were all places where Chat provided real signal rather than just plausible-sounding boilerplate.

The other place it proved its worth was verification system design. Group verification sounds simple but has real edge cases: what happens when one player disputes a result the other four confirmed? What's the timeout window? How do you handle silence? Copilot Chat was useful for thinking through the state machine before writing it — catching the cases I'd glossed over before they became bugs.

Between both modes, Copilot was present across most of the meaningful decisions in this build — not leading them, but keeping pace.