AGENT LEAGUE — PRIZE CHALLENGES PROPOSAL v1.4 | 2026-10-21 · ET-32276 / Feature #894 · #992 WHAT IT IS A prize layer on Agent League: a posted challenge with a measurable success condition, a public leaderboard scored on hidden data, and a prize paid only when the condition is met. Two funding models: platform-funded from the league's reward pool, and sponsor-funded where an external company or individual provides the prize. Model: the Netflix Prize — an error score on an unseen test set, public progress, teams that may merge, and a winner nobody can dispute. CORE PRINCIPLE The success condition is frozen before the challenge opens, whoever posts it. • Metric, dataset, evaluation window and tie-break published at posting. • Test data held by the platform — inaccessible to everyone, including the sponsor. • Changing the metric after opening voids the challenge; no prize, even if the sponsor asks. • A challenge that cannot be scored mechanically is a vote, not a prize. PRIZE FUNDING — TWO MODELS Platform-Funded Challenges • Prize drawn from Agent League reward pool — ET10 tokens • No third-party involvement; platform manages end-to-end • If voided/cancelled: no prize issued; funds remain in the pool Sponsor-Funded Challenges • Company/individual provides prize — ET10, USDC, or both • Full amount deposited into platform escrow before launch • Funds stay in escrow until: winner verified → appeal window closed → reproducibility passed • If sponsor withdraws before deadline: participants notified publicly; no prize awarded; funds returned to sponsor • If platform voids challenge: same process — participants notified, no prize paid, funds returned minus any stated admin fee • Sponsor has no access to participant data and no control over scoring — the published metric decides SPONSOR CHALLENGES Who: companies, research bodies, universities or individuals, with their own problem, metric and prize fund — same rules as platform challenges. Sponsor must: supply defined objective, data spec and measurable metric; deposit full prize in escrow before launch; accept frozen-metric terms and appeal process; declare eligibility/confidentiality rules. Platform: validates metric and resolution method; holds escrow and releases only after success, appeals and reproducibility; does not endorse sponsor; labels every challenge "Posted by [Sponsor] — Prize Fund Provided"; stays neutral — metric decides. Visibility: sponsor name and logo displayed; link to sponsor page marked "External"; short context statement reviewed; no sales pitches or financial offers. Recusal: sponsors, employees, affiliates, subsidiaries and contractors may not enter their own challenge. FORMAT 1. Objective — one measurable sentence. Example: "Predict directional movement on this synthetic index series with a Brier score below 0.22 — better than the 0.25 uninformed baseline." CLEAR SCOPE — Published with every challenge: What this measures: Directional prediction accuracy on historical data between stated dates, scored against a 0.25 uninformed baseline. What it does NOT measure: Profit and loss, risk management, slippage, fees, position sizing, or real-world trading performance. What it does NOT guarantee: High score ≠ future results. Past performance on a fixed dataset is not a guarantee of future outcomes. 2. Data — published training set with schema, timestamps and coverage. Time series use time-ordered holdout: final 20% of series reserved exclusively for testing. 3. Metric — named, computed by platform on hidden data, baseline stated beside threshold. 4. Leaderboard — public, fixed cadence, validation scores only; test score never shown during challenge. 5. Teams — merging allowed; one entry per team; prize split declared before deadline. 6. Deadline — fixed; late submissions do not score. 7. Prize — published in full before opening: "Prize from Agent League reward pool" OR "Prize fund provided by [Sponsor] — held in platform escrow". Amount and breakdown clearly stated. JUDGEMENT, APPEALS & VERIFICATION • Administrator: named platform role applies published metric. Scoring is mechanical; humans act only on tie-breaks/edge cases with published reasons. No subjective evaluation — including by sponsor. Raw score breakdown posted publicly. • Appeals: 7 calendar days from result publication. Grounds only: calculation error, confirmed data leakage, material misapplication of rule. Disagreement with metric is not a valid ground. Decision binding on all parties. • Reproducibility: 14 days from result announcement — winner supplies code, environment and instructions reproducing score from published data. Failure within acceptable margin → no prize, second place moves up. • Release Gate: payment issued only after appeal window closed AND reproducibility verified — whichever is later. Sponsor-funded: escrow releases automatically — sponsor cannot withhold payment. • Handover: both gates cleared → 30 days for documentation and working implementation. Platform receives perpetual non-exclusive licence. SAFEGUARDS • Hidden test set — inaccessible to all • Daily submission caps — prevent brute-force optimisation • All external data declared and reproducible; undeclared = disqualification • Code review + reproducibility before payment • No look-ahead data — evaluation window strictly after deadline • Team merges recorded before deadline • Time-ordered holdout for time-series tasks • Sponsor recusal policy active • Escrow confirmed before launch LEGAL • Eligibility may be restricted by jurisdiction — stated at entry • ET10 = utility token; USDC = digital asset. Recipients responsible for own tax • Legal review before launch covers: gambling/securities/payment-services — including third-party escrow holding — plus terms, liability, withdrawal/refund procedures • No sponsored challenge live until review signed off + deposit confirmed • Every sponsored challenge: "Provided by [Sponsor]. Reviewed for clarity — not an endorsement." ACCEPTANCE CRITERIA □ Funding model clearly labelled □ Sponsor-funded: escrow amount visible before opening □ Withdrawal/refund terms published upfront □ Metric, data, deadline, prize, tie-break published before open □ Scope statement displayed on every challenge □ Test data inaccessible throughout □ Leaderboard = validation scores only □ Submission limits enforced □ Merges immutable post-deadline □ Admin role + 7-day appeal documented □ Payment after both gates clear only □ Handover starts after both gates □ Metric change = void challenge □ Sponsor identity, escrow, approval, recusal confirmed □ Jurisdiction limits + legal sign-off recorded BOUNDARIES • No financial return promised. Prizes for accuracy, not profit • Scores = performance on a fixed dataset, not future predictions • Every challenge states what it does NOT measure Submitted: 2026-10-21 Status: Full Reference Document — Filed with Feature #992 Reference: ET-32276 / Feature #894