⚠ PAPER candidate signal — unproven, not financial advice
7-Day Trial DAY —/7
—
Pipeline — latest cycle
History
Paper equity — simulation, net of fees
Coinbase balance VIEW ONLY · REAL
Read-only view of your actual Coinbase account. Ghost cannot trade, transfer, or withdraw. Separate from the simulated paper equity above — do not confuse the two.
Paper equity vs buy-&-hold benchmark
Paper equityBuy & hold
Market & sentiment context CONTEXT, NOT ADVICE
Edge validation — skill vs luck
—
EV calibration — does the model's edge claim hold up?
—
The LLM over-claims edge — so we discount its EV (haircut) before acting and track claimed vs realized. A big positive gap = still overconfident.
Social arm — LunarCrush + Edge Lab RESEARCH · NOT trading
Loading…
Edge Lab — does social predict price?
Listening only — logging signals against forward returns to learn what leads price. Most will be noise; |corr|>~0.1 over a real sample is the first hint of signal.
Sentiment A/B — does it improve the validated result?
Budget — month-to-date Anthropic spend
Run feed
Betfair Research
PAPER — no real bets · read-only · not advice
Loading…
Paper ledger — AU racing arbitrage
Riskless arbitrage DUTCHING
Validation — did the paper arbs lock profit?
—
Recent paper bets
Run / refresh
Run = settle finished races → scan AU racing books → paper-place any overround arb. Refresh = re-pull ledger only (no scan). Paper/read-only; Ghost places nothing.
🎡 The Wheel
RESEARCH — capturing odds · NO bets · building private data
Loading…
Private dataset — Challenger / ITF odds + CLV
CLV verdict — do we have an edge?
—
How to read:Avg CLV = did the market price move toward our pick after we flagged it — the honest early signal. Positive CLV over ~100+ graded bets = a real edge. “Beat close” = % of flags that got a better price than the closing line. Paper ROI is noisier — trust CLV first. Until ~100 graded, it says “too early.”
Value signals — model vs market
Upcoming markets being captured
Capture / refresh
Capture = snapshot every lower-tier tennis market's odds → grows the private dataset (auto-runs hourly). Refresh = re-read stats only. Read-only, no bets. The Elo edge-model + CLV grading switch on once enough data accrues.