Overview
Risk guardrails
Every cap is a fixed % of the live bankroll, enforced in code. Here's each limit and where Vic stands against it right now.
P&L over time
Realized only. Bars per-day; line is cumulative.
Positions
Open positions ●
—
| Market | Family | Side | Qty | Entry ¢ | Mark ¢ | Unrealized |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Recent settled trades
Last 20, newest first.
| Market | Family | Side | Qty | Entry ¢ | Exit ¢ | Realized |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Biggest winners
—
Top 20 settled trades by realized P&L — tap any row for the full story (entry, edge, hold time, CLV, and how the thesis played out). These are the biggest single-trade wins, not a profit claim; the all-time P&L up top is the honest bottom line.
Analysis
Realized P&L by hypothesis
All-time, top 10 by absolute P&L.
Calibration by hypothesis family
Predicted vs. realized hit rate. Lower Brier is better.
| Family | n | Predicted | Realized | Δ | Brier |
|---|---|---|---|---|---|
| Loading… | |||||
Trade activity & extremes
Buys per hour of day (ET) + best/worst settled trade, over the selected window.
Costs
Cost to run Vic vs P&L
—
—
API & LLM Usage & Costs
—
Per-call LLM cost from the OpenClaw gateway's session logs (xAI Grok, Gemini, Anthropic, OpenAI) — Victoria's spend only, excludes Claude Code dev. Combined adds the prorated fixed cost of ownership (next card).
| Item | / mo | / day |
|---|---|---|
| Loading… | ||
Steady-state renewal rates; operating view (Pi electricity only — board capex sunk).
| Provider | Today | 14-day | Calls | Source |
|---|---|---|---|---|
| Loading… | ||||
Metered = the gateway's own per-call price (OpenAI). Token-priced = priced from the call's token counts at the vendor's list rates — used where the gateway reports $0 (Anthropic, Gemini) or a wrong table (xAI Grok). Chat runs on grok-4.6 ($2 in / $6 out per 1M tokens); the daily heartbeat on grok-4.3.
| Date | LLM cost | LLM calls | Trades |
|---|---|---|---|
| Loading… | |||
LLM costs are metered from the OpenClaw gateway's own per-call accounting (vic-meter-llm-cost.py) — Victoria's autonomous spend only, excluding Claude Code dev sessions (David's subscription, a separate pool). Vendors the gateway doesn't price are token-priced from official list rates and labeled token-priced/mixed in the breakdown (trading itself is zero-LLM since Phase 40 — the residual spend is the daily heartbeat). Every priced call is now persisted by responseId, so days survive the gateway's ~1-week log rotation instead of undercounting once their logs age out. Fixed cost of ownership (hosting, domain, Pi electricity) uses steady-state renewal rates, operating view; the ~$120 board capex is sunk and excluded from totals.
Method
Principles & method
Vic runs on a self-hosted Raspberry Pi 5 and trades Kalshi with real money — disciplined, percentage-of-bankroll stakes with hard risk caps. Every position carries a written rationale, and once it settles she scores the outcome against that rationale and folds the lesson back into the next scan. She researches, she sizes with discipline, and she learns. The full ledger — wins, losses, fees, and running P&L — is published here, warts and all.
Vic trades the Kalshi book on a disciplined daily cycle — research first, a position sized only when the edge justifies the risk, and an honest record of how each one turns out. The specifics of how she finds and sizes those edges stay in-house.
Honesty is the floor. Every price, edge, P&L, and settlement value on this page traces to a real source and is shown exactly as it stands — including when it's negative. Each settled position is judged against the rationale it was opened on, and the lesson carries into the next.