Skin

skin in the game

An AI analyst that has to bet its own money on every call it makes.

Its track record is not a claim it writes about itself. It is settled positions on Binance prediction markets, scored with a rule it cannot game.

Live dashboard →

The problem

Every AI agent will tell you what it thinks the market will do. None of them pay for being wrong.

An agent can say "BTC looks bullish, roughly 80% confidence" a hundred times. It is never scored, so the confidence number means nothing, and it costs nothing to inflate. This is the single largest unsolved problem with AI financial agents: there is no price on being wrong, so there is no information in anything they say.

The fix

Binance Agent OS shipped prediction markets to the Agentic Wallet. A prediction market quotes belief directly: an outcome token at $0.62 is a 62% probability. That makes it the only venue where an agent's forecast and its money can be the same object.

So this agent is not allowed to have an opinion for free.

1. It reads a market and computes a probability from realized volatility: no LLM guess, a closed-form model anyone can recheck.

2. If it disagrees with the market by more than 4 points, it sizes a stake by the Kelly criterion and buys with real USDT.

3. Before the outcome is known, the conviction goes into an append-only journal.

4. When the market resolves, the result is scored by Brier score, a strictly proper rule, meaning the agent's best possible strategy is to state what it actually believes.

5. It sweeps its own winnings, because prediction markets do not pay out automatically and an agent that never claims goes broke while being right.

Every number on the dashboard traces back to a settled on-chain position. The agent does not get a vote in its own performance review.

Table of contents

• What it looks like

• Quick start

• Binance Agent OS surfaces used

  • Using the MCP Server

• Why it refuses

• How the model works

• Architecture

• Repo layout

• Commands

• Safety

• Testing

• For judges and reviewers

• Honest limitations

What it looks like

The dashboard is not a report you scroll. It is the agent's loop with a button on it. Run a scan prices five markets one at a time and says out loud why it refuses most of them; Stake makes you type the word stake before any money moves, exactly as the CLI does; Claim sweeps the winnings that prediction markets do not pay out on their own. Every number in it comes from the scan block of record.json, which the CLI produced by calling the same formOpinion and sizeStake a live run calls, so the demo cannot quietly disagree with the tool it is demonstrating.

The record

  SKIN  skin in the game   ● DEMO  synthetic fixtures · no wallet, no money moved

THE RECORD
computed from settled positions · the agent does not get a vote
──────────────────────────────────────────────────────────────────────────
  realized PnL                +$0.07  on $13.20 staked
  return on stake               0.5%

  Brier score                  0.261  roughly as useful as always saying 50%
                                      0 perfect · 0.25 = always saying "50%" · lower is better

  settled calls                   12
  won / lost                   7 / 5
  hit rate                     58.3%
──────────────────────────────────────────────────────────────────────────

That is the bundled demo record, and it is deliberately unflattering: up seven cents on $13.20, a Brier score barely better than a coin flip, and the biggest single loss is the call it was 88% sure about. A demo that showed a winning agent would be showing you the marketing, not the mechanism.

The point of the project is that this number is not editable. So it is not edited.

A betting slip

Every stake prints a receipt before any money moves: the full derivation, and which of the four ceilings actually bound the size.

┌┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┐
┆ BTC above $109,800 · 1 h                                      PENDING  ┆
┆ side NO                                                                ┆
┆┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┆
┆ agent says            73.8%                                            ┆
┆ market says           48.0%                                            ┆
┆ edge                  25.8%                                            ┆
┆ payout odds           1.08:1                                           ┆
┆ full Kelly            49.6%                                            ┆
┆ applied (¼ Kelly)     12.4%                                            ┆
┆ bound by              kelly                                            ┆
┆┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┆
┆ STAKE                 $1.22                                            ┆
┆┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┆
┆ BTC spot $109,420, realized vol 52% (200 × 1m candles). Zero-drift     ┆
┆ lognormal over 58 min puts P(BTC above $109,800) at 26.2%; market      ┆
┆ prices the NO token at 48.0%.                                          ┆
└┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┘

A scan

Three of these five markets are refused. That is the system working.

SCAN
5 market(s) · bankroll $9.83 · run cap $6.00 · ¼ Kelly
──────────────────────────────────────────────────────────────────────────
  pass      BTC Up or Down · 5 min                    50% vs 50% · no-edge
  BET       BTC above $109,800 · 1 h                  74% vs 48% · $1.22
  BET       ETH above $4,140 · 15 min                 95% vs 70% · $1.50
  pass      SOL above $204 · 1 h                      65% vs 58% · below-minimum
  pass      BNB above $872 · 4 h                      54% vs 52% · no-edge
──────────────────────────────────────────────────────────────────────────

The dashboard

skin-in-the-game-five.vercel.app

A React page rendering the same record.json the CLI exports: the interactive console, then the equity curve split at zero, a reliability diagram, the slips, and the ledger. No backend: the page reads one file, so what it shows is exactly what skin export produced. The mode badge in the header is permanent, so a demo number can never be mistaken for a live one.

The equity area is clipped at zero, green above and red below, because the one question anyone asks of that chart is "is it above or below the line?", and that should be answerable from across a room.

The reliability diagram plots stated confidence against what actually happened; the amber diagonal is perfect honesty, and a bubble's area is the number of calls in that bucket, so a single outlier cannot be mistaken for a trend. The bars beside it say the same thing in words for anyone who does not read scatter plots: what it claimed, what it delivered, and by how much it missed.

Every call is a betting slip carrying the reasoning that produced it, stamped with what happened. Both charts are hand-drawn SVG, with no charting library, because each has a reference line a generic library fights you to draw, and the whole page gzips to 68 kB.

npm run web:dev

The dashboard lives in web/, and this repo has two package.json files: the CLI's at the root and the web app's in web/. Set the Vercel project's Root Directory to web, or the build runs the root package.json's build script (tsc -p tsconfig.json, which compiles the CLI) and deploys nothing servable. The symptom is a successful build followed by a 404 on every route. web/vercel.json supplies the rest.

Quick start

No wallet, no money, no signup. The demo path is fully offline and is what a reviewer should run first:

git clone https://github.com/martinvibes/skin-in-the-game
cd skin-in-the-game
npm install

npm run skin -- record --demo     # the track record
npm run skin -- scan   --demo     # opinions, no money
npm run skin -- stake  --demo     # receipts + typed confirmation
npm run skin -- claim  --demo     # sweep unclaimed winnings
npm test                          # 49 unit tests

The dashboard:

npm --prefix web install
npm run web:dev

Going live

Live mode needs a Binance MPC Wallet (created in the Binance mobile app; an agent cannot create one for you) and a small USDT balance on BSC. Stakes are about $1; $10–15 is enough to run this for real.

# 1. Install the Agentic Wallet skill
npx skills add binance/binance-skills-hub/skills/binance-web3/binance-agentic-wallet

# 2. In your agent: "Sign in to Binance Agentic Wallet", approve on your phone

# 3. Optional: connect the MCP Server for market data
claude mcp add binance-mcp-server --transport http https://agent.binance.com/mcp/agentic

# 4. Run for real
npm run skin -- record --live
npm run skin -- stake  --live --budget 6 --per-call 1.5

skin auto-detects: if baw is on your PATH it goes live, otherwise it falls back to demo and says so in the banner. --live forces the real path and fails loudly rather than silently pretending.

Binance Agent OS surfaces used

The wallet's daily limit as a sizing input is the part worth pausing on. Most agent projects treat wallet limits as an error to handle. Here it is one of the four terms in the position-sizing minimum, on equal footing with Kelly.

Using the MCP Server

Connect it once:

claude mcp add binance-mcp-server --transport http https://agent.binance.com/mcp/agentic

Then the agent fetches candles under its own authenticated session, writes them to a file, and hands the path over:

{
  "BTCUSDT": { "interval": "1m", "closes": [109380.2, 109412.5, "…200 closes"] },
  "ETHUSDT": { "interval": "1m", "closes": ["…"] }
}

skin scan --mcp-data ./klines.json

A committed fixture lets you exercise that path right now, with no MCP session and no network:

npm run skin -- scan --demo --mcp-data fixtures/klines.example.json

  pass      BTC Up or Down · 5 min                    50% vs 50% · no-edge
  BET       BTC above $109,800 · 1 h                  73% vs 48% · $1.18
  BET       ETH above $4,140 · 15 min                 96% vs 70% · $1.50
  pass      SOL above $204 · 1 h                      50% vs 42% · below-minimum
  pass      BNB above $872 · 4 h                      50% vs 48% · no-edge

Two of those markets have a strike sitting exactly on spot, and the model prices both at exactly 50%, the cheapest available check that the lognormal is wired up correctly. At the money, over any horizon, a zero-drift walk is a coin flip.

Without --mcp-data the CLI falls back to Binance's public market-data REST endpoint, which needs no credentials. Same arithmetic either way; the MCP path is the one that runs inside the user's Agent OS session. It tries four hosts in order, including data-api.binance.vision, because api.binance.com does not resolve from every country this was developed in.

Why it refuses

Money moves only if a call survives all five checks. Each one has a single job, and each produces a named verdict rather than a silent skip.

A scan where nothing clears is a successful scan. The demo fixtures are tuned so all five verdicts are reachable, because a gate you cannot see fire is a gate you cannot trust.

How the model works

No language model touches this path. Everything is computed, so it can be rechecked.

Pricing. Threshold markets are barrier questions with a standard answer. Assume zero-drift geometric Brownian motion:

P(S_T > K) = Φ(d₂)        d₂ = [ln(S/K) − σ²T/2] / (σ√T)

Zero drift is the honest assumption. Over a 5-minute horizon any drift you could estimate is swamped by noise, and assuming one is exactly how forecasters talk themselves into positions. μ = 0 makes the price a martingale and leaves volatility as the only real input.

Volatility. Realized, from log returns of 1-minute klines, annualised. Under 10 returns it returns null and the market is declined, because a model that always produces an answer is a model that sometimes lies.

Sizing. Kelly, for a binary token bought at cost c:

b  = (1 − c) / c
f* = p − (1 − p)·c/(1 − c)

f* is zero exactly at p = c, so agreeing with the market produces a zero bet with no special-case rule. Quarter-Kelly by default: full Kelly is only optimal if p is exactly right, and p is a model output. Kelly is brutal under overestimated edge, and the edge estimate is the least reliable term in the equation.

Scoring. Brier score, (1/N)·Σ(conviction − actual)². It is strictly proper: minimised only by reporting true beliefs. The agent cannot improve it by sounding confident and cannot improve it by hedging to 50%. It returns null, never 0, when there is nothing to score, since zero would read as perfect.

Full derivations in skill/references/model.md.

Architecture

                      ┌──────────────────────────────┐
   Binance MCP  ─────▶│  marketdata.ts               │
   (klines)           │  realized volatility         │
                      └──────────────┬───────────────┘
                                     ▼
                      ┌──────────────────────────────┐
                      │  analyst.ts    P(S>K) = Φ(d₂)│──▶ conviction
                      └──────────────┬───────────────┘
                                     ▼
   baw prediction ───▶┌──────────────────────────────┐
   market list        │  sizing.ts   Kelly + 4 caps  │──▶ stake or decline
                      └──────────────┬───────────────┘
                                     ▼
                      ┌──────────────────────────────┐
                      │  journal.jsonl (append-only) │  ◀── written BEFORE
                      └──────────────┬───────────────┘       the outcome
                                     ▼
   baw trade      ◀───┌──────────────────────────────┐
   place-order        │  cli/index.ts                │
   redeem             └──────────────┬───────────────┘
                                     ▼
   baw position   ───▶┌──────────────────────────────┐
   settled-history    │  record.ts  Brier + calib.   │──▶ record.json ──▶ web
                      └──────────────────────────────┘

Three layers, and the boundary between them is enforced, not aspirational:

• adapters/: everything that touches the outside world. Processes, HTTP, the filesystem. All failure lives here.

• engine/: pure and total. No I/O, no clock, no randomness, no network. This is why 49 unit tests are worth something.

• cli/: rendering and confirmation. Formats; never computes.

The PredictionClient interface is the seam. LiveClient shells out to baw; DemoClient returns fixtures. Nothing above the seam knows which it has, which is what makes the demo path a real exercise of the code rather than a mock of it.

Repo layout

src/
  domain/types.ts        the vocabulary: Call → StakeVerdict → Position → Record
  adapters/
    exec.ts              execFile-based process runner (never a shell string)
    baw.ts               Agentic Wallet client + defensive JSON pickers
    marketdata.ts        REST / MCP / static kline sources, realized vol
    demo.ts              12 settled fixtures + 5 markets, all five verdicts
  engine/
    analyst.ts           erf, Φ, P(S>K), market-title parsing
    sizing.ts            Kelly, edge, the four ceilings
    record.ts            Brier, calibration bins, equity curve
    journal.ts           append-only JSONL, first-write-wins
  cli/
    index.ts             record | scan | stake | claim | positions | export
    render.ts            receipts, ledgers, sparklines, ANSI-aware padding

skill/
  SKILL.md               the Agent OS skill · the policy the agent follows
  references/            command surface + model derivations

web/                     Vite + React dashboard ("Ledger Noir")
fixtures/                a klines payload for exercising the --mcp-data path
docs/                    dashboard screenshots used by this README
test/math.test.ts        49 tests over the pure engine

~3,300 lines of TypeScript, strict with noUncheckedIndexedAccess.

Commands

Flags: --demo --live --yes --json --budget --per-call --kelly --min-order --limit --mcp-data --out. Full reference: skill/references/commands.md.

Safety

This spends real money from a real wallet, so:

• Typed confirmation. stake and claim require the whole word typed. y is rejected. On a non-TTY they refuse outright unless --yes was passed, and --yes announces itself in the output.

• No shell strings. exec.ts uses execFile with an argv array. A market title cannot become a command.

• Market titles are never sent to a model. They are attacker-controlled text from a public venue. They are parsed by regex and rendered as data. This removes the prompt-injection surface rather than trying to filter it.

• The wallet's daily limit is a sizing input, not an error to handle. The agent cannot raise it; that lives in the Binance app.

• Errors are reported verbatim. BawError carries the CLI's own words. A wallet error is never paraphrased into a guess.

• Nulls are never defaulted. A missing price is null and produces a decline. It never silently becomes 0, which would read as a 100% edge.

• The journal is append-only, first-write-wins. Editing it is the one way to make the Brier score meaningless, which is why re-recording a token is refused rather than overwritten.

Testing

npm test          # 49 tests
npm run typecheck # tsc --noEmit, strict

The tests cover the mathematics, not the plumbing: erf against known values, Φ symmetry, Kelly against hand-computed cases, the zero-at-p=c property, which ceiling binds in each regime, Brier against textbook examples, the null-not-0 rule, calibration bucketing, and equity accumulation.

CI runs all of that on every push, plus a smoke test that executes every command in demo mode on a clean machine, with no wallet, no network, no journal in $HOME. It also re-exports web/public/record.json and fails if it differs from the committed copy, so the dashboard payload can never quietly drift from the fixtures that are supposed to produce it.

One of them caught a real bug: parseClaim('BTC Up or Down · 5 min') returned 'below', because the below pattern was tested first and matched the word "Down". That silently inverted every two-sided short-duration market, most of the tradable universe. It is fixed, and the test that found it is test/math.test.ts.

For judges and reviewers

Five minutes, in order:

1. npm install && npm run skin -- record --demo: the whole thesis in one screen. Note that the demo agent is barely profitable and has a mediocre Brier score. That is on purpose.

2. npm run skin -- scan --demo: three of five markets refused, each with a named reason.

3. npm run skin -- stake --demo: the receipt, and the typed confirmation. Try typing y; it will not accept it.

4. src/engine/sizing.ts: ~120 lines, the heart of the project. Four ceilings, the tightest binds, the binding one is named in the output.

5. skill/SKILL.md: the Agent OS skill, including the rule the agent is bound by: it does not get to state an opinion about a market without backing it.

What is genuinely different here, stated plainly: this is the only entry we are aware of where the agent's track record is adversarially verifiable. You do not have to trust the dashboard, the README, or the agent. The positions are on-chain and the scoring rule is strictly proper. If the agent were lying about its confidence, the Brier score would get worse, and the Brier score is the number it is judged on.

Honest limitations

Written before anyone asks.

• The model is simple by choice. Zero-drift lognormal on realized vol. It has no view on order flow, funding, news, or microstructure. It will be systematically wrong around scheduled events. A better model would improve the PnL; it would not change what the project demonstrates.

• Track record length. Brier over a dozen calls is indicative, not conclusive. The CLI says so itself: under five scored calls it prints "too few to judge" instead of a verdict.

• Fees and slippage are inside the realized PnL because it is computed from settled positions, but they are not modelled ahead of a trade. On $1 stakes the effect is small; at size it would need to enter the edge threshold.

• Orders are submitted, not filled. place-order returning success means accepted. The CLI says this and points at baw prediction order history --status FILLED.

• Journal integrity is local. Append-only and first-write-wins, but it is a file on the operator's machine. Anchoring a hash on-chain would make it externally verifiable; the settled positions already are.

License

MIT. See LICENSE.

Built for the Binance Agent OS Mini Hackathon, Track A.