Have none of that. They know their record, not whether their process has any edge at all.
01OKX.AI Genesis 2026 · A2MCP Grader
You can win the bet
You can win the bet
and still be wrong.
CLV Scout grades every pick against the closing line, not the scoreboard — a calibrated A+…F verdict for one cent, settled in USD₮0 on X Layer.
$0.01 / grade
$0.20 / audit
x402 exact · EIP-3009
USD₮0 zero-gas
Edy Cu · @edycutjong
02The Problem
Results lie.
Results lie.
The closing line doesn’t.
Short-run P&L is noise. The professional gold standard for whether a bet was actually good is Closing Line Value — did your price beat the market’s final price? Measuring it takes closing-odds capture, grading discipline, and a settled sample to calibrate against.
Retail bettors
Copy-trading agents
Follow wallets ranked by raw P&L — the exact metric CLV exists to correct.
The marketplace
Every ASP on the board sells new predictions. Nobody grades the ones already made.
A lucky coin-flipper and a sharp look identical over 5 bets — and totally different against the close.
03The Solution
A micro-priced grader.
A micro-priced grader.
Not an opinion — a number.
Paste a bet you placed (or up to 25 of someone else’s), pay a cent or two over OKX x402, and get a calibrated grade backed by a settled truth table you can independently recompute.
■Calibrated, not asserted
Every grade A+…F ships with its settled sample: n, win rate, ROI for that exact band — recomputable from source.
■Structurally honest
No recorded closing line? The answer is UNGRADED with a reason. There is no synthetic-close code path anywhere in the codebase.
■One capability, three prices
$0.01 single grade → $0.20 dossier of 25 → (roadmap) $0.005 streamed batch. Same engine, same discipline.
04Product Flow · Endpoint Surface
Five routes. Two paid.
Five routes. Two paid.
Zero accounts.
| Route | Price | Returns |
|---|---|---|
| POST /api/grade | $0.01 | clv_grade A+…F, clv_pct, beat_close, close odds + the settled truth table for that grade band |
| POST /api/audit | $0.20 | ≤25 bets → full CLV dossier: beat-close rate, grade distribution, weighted expectancy, Sharp Score 0–100 from disclosed sub-scores |
| POST /api/calibration | free | Live grade→outcome truth table + methodology, computed from the settled ledger — the proof, and the funnel |
| POST /api/me | free | Your grading history keyed off your X-PAYMENT signature — payment identity doubles as login |
| POST /api/receipts/verify | free | Re-check any settlement via the Facilitator’s GET /settle/status |
The paid wire: probe unpaid → 402 + x402Version:2 → sign EIP-3009 TransferWithAuthorization → replay with X-PAYMENT → 200 + settle on X Layer.
05Architecture
Boring on purpose.
Boring on purpose.
Verifiable by design.
Node 20 + TypeScript strict, Express 4, viem for real EIP-3009/EIP-712 cryptography, zod validation, and a flat-file settled ledger — no DB server, no build step, fully offline-testable.
BuyerAgent / bettorPOST + X-PAYMENT
→
PaywallokxPayGatex402 exact scheme, routes-map pricing
↔
SettlementOKX Facilitatorverify + settle → X Layer eip155:196
→
EngineCLV graderentry vs close → CLV% → A+…F
→
TruthSettled ledgerper-grade n · win% · ROI truth table
The payment rail hand-rolls the documented OKX wire shapes with only viem + express — signature verification is real, offline cryptography (recoverTypedDataAddress), exercised end-to-end against a live server.
06Live Demo · The Money Shot
The tout autopsy.
A tout posts his last 5 World Cup picks: 3–2 on flat stakes — looks like someone worth following. Feed the same 5 picks to /api/audit ($0.20) against the real settled closing-line ledger:
+12.0%
raw ROI · what the leaderboard shows
→
40.7/100
Sharp Score · what the close says
1 of 5
picks beat the closing line
{
"graded": 5,
"beat_close_rate": 0.2,
"grade_distribution": { "A+": 0, "A": 1, "B": 0, "C": 1, "D": 2, "F": 1 },
"weighted_expectancy_pct": -22.14,
"sharp_score": { "value": 40.7 },
"verdict_line": "Sharp Score 40.7/100 — the record may look fine on raw P&L,
but the numbers that predict future results say otherwise."
}
Real, unmodified output from this build against the 24-row settled seed ledger — plus the “WON but graded C” reveal and a live UNGRADED refusal, all driven by real x402 round-trips on the served proof page.
07Sponsor Integration
Why ONLY OKX.
- $0.01 with real settlementThe SDK turns a one-cent USD string into atomic USD₮0 and the Facilitator settles it on X Layer per call. On a general-purpose EVM rail, $0.01 dies the moment anyone pays gas.
- USD₮0 zero-gas promoThe buyer’s marginal cost is exactly one cent — the only thing that makes “audit a tout’s last 25 picks for $0.20” a rational agent action.
- Tiered pricing = a config block“$0.01” grade and “$0.20” audit are two entries in one routes map — not two integrations.
- Public sold countsThe 1¢-volume thesis is verifiable by judges on the marketplace listing — without trusting us.
- Same exact/EIP-3009 wirePlus X-PAYMENT-identity BuyerLens and Facilitator settle/status receipt re-verification.
Take OKX out and the product dies at the till. The grader is portable — the economics are OKX-specific.
08Market & Impact
Quality filters priced
Quality filters priced
below the mistake.
The thesis: when a grade costs one cent, checking becomes cheaper than being wrong — for humans and for agents. That price point only exists on a zero-gas micro-payment rail.
Bettors self-auditing
“Am I actually good, or early?” A number for every slip — and permission to stop betting the bad half of a slate.
Copy-trading agents
A $0.20 pre-follow audit vs one bad follow. The cheapest risk check an agent can buy, composable mid-flow via A2MCP.
Tipsters & other ASPs
Attach a neutral, recomputable grade to their picks and inherit the calibration — credibility as a service.
Honest framing: no invented TAM. The bet is that 1¢ x402 calls create a new volume category — and marketplace sold counts will prove or disprove it in public.
09Competitive Edge
Everyone sells picks.
Everyone sells picks.
We grade them.
■Different lane, same board
Prediction ASPs compete on being right next time. A grader compounds on every pick anyone else makes — including theirs.
■Refuses to hallucinate
UNGRADED is a feature: no recorded close, no grade. A grader that refuses to grade is nobody’s tutorial project.
■Every number decomposes
Sharp Score recomputes from its published sub-scores (tested invariant). Truth-table rows sum to the ledger count. Nothing opaque.
■Payment identity is the login
No accounts, no API keys for buyers — your X-PAYMENT signature is your history. Frictionless for agents by construction.
Deliberately narrow: World Cup 2026 knockout markets only, stated in every calibration response. Depth over breadth.
10Proof Today · Roadmap
Real signals only.
124
tests passing (Vitest)
100%
coverage — stmts / branch / funcs / lines
402·402·405·200
live-boot readiness self-check
Not deployed, no users yet — and we say so. Every number in this deck is reproducible from source: clone, npm test, run the money shot yourself.
30 days
Go live. Deploy the API to api.clvscout.edycu.dev, pass the okx.ai listing review, first on-chain paid-call receipts on X Layer.
60 days
Stream it. Tier-2 /api/grade-stream: $0.005/call via aggr_deferred — Session-Key signatures, TEE-batched into one on-chain tx.
90 days
Prove it harder. zk-Sharp attestation + staked calibration bond; SDK + MCP server; broaden line-history coverage beyond the World Cup.
11Team & The Ask
Solo-built. Fully auditable.
Builder
Edy Cu
Solo builder — design, engine, payment rail, tests, docs. GitHub @edycutjong. Serial hackathon shipper; CLV Scout is the second listing on one shared OKX x402 rail.
The ask
Run the 90-second demo: clone, npm run api, and watch a +12% tout collapse to Sharp Score 40.7/100 over a real x402 round-trip. Then check our math — the calibration table is free.
“Grade yourself before you copy anyone.”