# Hackathon #4 Hackathon #4 is designed to push **agentic applications** from “cool demos” into **usable, repeatable workflows** — where agents can delegate work, call tools, validate outcomes, and keep going. Hackathon Format (unchanged): Open-ended runs continue — variable inflow = high-value edge-case data. We’ll keep inviting outsiders to try agentic apps while hardening software, docs, and process. *** ### Objective Build proof-of-concept (PoC) **agent applications**, agent frameworks, bots, validators, and developer tools that leverage **Cortensor’s decentralized inference protocol** as the execution + trust backbone. #### Primary focus areas (Agent-first) * **Agentic Applications (core track)**\ Agents that *do real work*: monitor, act, coordinate, transact, and report — with clear tool interfaces and repeatable flows. * **Delegation + Execution Loops**\ Demonstrate a full loop:\ \&#xNAN;*Agent plans → delegates tasks → Cortensor executes (redundant compute) → results returned → agent continues.* * **Trust & Verification (PoI/PoUW utilization)**\ Use **PoI** (redundant inference) and **PoUW** (validator scoring) where it adds credibility: * cross-model or cross-run checks * rubric-based scoring * evidence bundles / audit logs * “why trust this output” surfaces * **Agent Tooling & Templates**\ SDK wrappers, CLI scaffolds, starter kits for: * agent + tool calling * session management * logging + replay * evaluation harnesses * **Operational Agents (DevOps / Community / Monitoring)**\ Agents that keep infra and communities healthy: * router health checks, latency alerts * miner/model usage analytics * automated incident summaries * support triage bots with safe guardrails * **Public Goods Agents**\ Open-access bots that provide ongoing community value (free inference tier, limited quotas, or public endpoints). #### Stretch goals (bonus points) *Note: ERC-8004, x402, MCP, and COR Prover are optional R\&D surfaces — not required to participate.* * **ERC-8004 artifacts**: emit agent identity/validation artifacts for discoverability. * **x402**: pay-per-call rails or UI flows for usage-based access. * **Router surfaces**: integrate Router v1 (REST) with a `/validate` endpoint; prototype COR Prover surfaces; explore MCP-compatible patterns. *** ### Prize Pool (in $COR equivalent) * 1st: $1000 * 2nd: $800 * 3rd: $500 * 4th–5th: $300 each * 6th–7th: $150 each * 8th–10th: $100 each * 11th–20th: $50 each **Ongoing Support:** Top projects may qualify for monthly $COR grants for continued maintenance and improvements. #### Important Notice on Prize Payouts (eligibility condition) To align incentives with Cortensor’s long-term growth: if a prize-winning participant is **not** (a) an existing node operator running >20 nodes, or (b) a holder of $500+ $COR prior to kick-off, or (c) a participant with active staked tokens, then their prize will be split: * 50% allocated directly to Staking Pool #1 under their address * 50% distributed upfront as liquid $COR This ensures rewards both support winners and strengthen network security and staking participation. #### Judging & Awarding Policy * **Quality over rank:** Prizes are **not** awarded strictly by leaderboard order. Cortensor will judge qualified entries on **overall quality, impact, technical rigor, usability, documentation, and alignment with Cortensor’s roadmap**. * **Discretionary placement:** Judges may award a prize to any qualified entry at a given place (e.g., a lower-ranked but higher-quality project may receive a higher prize tier). * **No guarantee to fill all places:** Cortensor may **withhold or reallocate** any prize tier if submissions do not meet the quality bar. * **Ties / partial awards:** Cortensor may declare ties, split awards, or issue **honorable-mention grants** at its discretion. * **Compliance:** Entries must comply with all rules; violation may result in **disqualification** and forfeiture of prizes. * **KYC requirement:** Cortensor may require **KYC** verification for any qualified or prize-winning entry. If requested, successful completion of KYC is **mandatory** to receive prizes. *** ### Suggested Project Ideas #### New and emphasized for #4 (Agent Apps) **1) Real Operators: agents that run workflows** * **Release Manager Agent**\ Watches repos, builds changelogs, drafts release notes, posts weekly dev summaries. * **Incident Commander Agent**\ Monitors router/validator health → detects anomalies → opens tickets → posts a short incident report. * **QA / Regression Agent**\ Runs daily test scripts against endpoints, compares outputs, flags drift, and files issues with evidence. * **Community Support Agent**\ Triage questions in Discord/Telegram, routes to docs, escalates edge cases to humans. **2) On-chain & crypto-native agents (safe + verifiable)** * **Receipt Verifier Agent**\ Takes tx hashes/receipts → explains what happened → validates expected state changes → emits an attestation artifact. * **Treasury Watch Agent**\ Monitors balances/flows, posts alerts, generates periodic reports (with policy-based thresholds). * **Governance Analyst Agent**\ Summarizes proposals, finds risks/tradeoffs, and provides structured “for/against” reasoning (with sources). **3) Multi-agent systems and coordination** * **Coordinator Agent**\ Breaks a goal into subtasks, delegates to specialized agents/tools, merges results, validates final output. * **Disagreement Resolver**\ When two agents disagree, run multi-run validation + rubric scoring and return a structured arbitration bundle. **4) Memory & personalization (done responsibly)** * **Project Memory Agent**\ Maintains a structured memory store (decisions, constraints, TODOs), produces weekly deltas, supports “why did we decide X?” * **Knowledge Base Builder**\ Turns docs/issues into a searchable map, then answers questions with citations + confidence + validation. #### Tooling / DX ideas (still highly valued) * **Agent Starter Templates** (TS/Python): tool registry, session helpers, structured logging, replay runner * **Evaluation Harness**: local scripts to run N trials across models/runs; generate validation reports * **Observability dashboards**: agent task success rates, latency, failure reasons, validator scores over time *** ### Ongoing Support * Outstanding projects may receive monthly $COR grants for ongoing improvements. * Exceptional teams may be offered dedicated roles in the Cortensor developer community. *** ### How to Participate * Join Discord: discord.gg/cortensor * Build openly in the community repo: * Submit your final app with a clear README showing functionality and Cortensor integration: * Create PR as submission at * Demo App/URL * Documentation / Video * Cortensor Session IDs & Router Endpoint (if you used web2 integration) * Post all of the above into Discord **#build-ground** #### Deliverables Checklist (Agent-focused) * Public repo with permissive license (MIT/Apache-2.0) * README with: * quickstart + runbook * architecture diagram * tool list (what the agent can do) * safety/constraints (what it refuses to do) * Demo link (live URL or recorded video) + reproduction steps * **Agent runtime proof**: * sample transcripts / logs * structured outputs (JSON where relevant) * replay script or test command * **Verification** (recommended): * rubric prompt(s) / scoring policy * cross-run checks or validator usage * evidence bundle format (JSON + optional IPFS) *** ### Timeline * Kick-off: **TBD (Q1 2026)** * Submission deadline: **TBD (≈6 weeks after kick-off)** > If dates shift, we’ll announce updates in Discord. *** ### Evaluation Criteria Mandatory: Working public demonstration during the hackathon (Discord/Telegram or designated channel). Projects not demonstrated will not qualify. Scoring rubric: * **Agent capability & workflow completeness** — 30% * **Integration with Cortensor** (sessions, routing, validators, staking/payment flows; PoI/PoUW where applicable) — 25% * **Reliability & safety guardrails** — 20% * **Usability & demo quality** — 15% * **Public good impact** (free access, docs, community value) — 10% Bonus considerations: ERC-8004 artifacts, x402 flows, MCP patterns, `/validate` usage, on-chain records, tests/benchmarks, observability dashboards. No Auto-Generated Submissions: Purely AI-generated entries without a working implementation do not qualify. *** ### Terms & Notes * Optional surfaces (FYI): COR Prover, ERC-8004, x402, and MCP are internal/WIP and not required. * Safety/Compliance: Projects must follow applicable laws and reasonable safety guidelines. * Transparency: Clearly mark any centralized components or paid services used. * Data: Use public or properly licensed datasets/APIs. * Contact: Questions and office hours will be posted in Discord. *** Hackathon #4 is about **agents that actually ship work** — with **execution you can rely on** and **results you can verify**. Let’s build agent apps that hold up in the real world.