The answer depends on your book.

Every AI says “buy NVDA”. Nobody asks what you already hold — or that rTokens print all weekend while New York sleeps. Weekend Copilot grades the trade you are about to make against the book you actually have, and hands you a receipt.

60-second answermeasured in-token weekendsbreaker at 2× costsread-only · no keys

1 · Your book

Presets always resolve. Unknown tickers are rejected, never guessed.

60% NVDA into the weekend — the concentration trap.

2 · Propose the trade

Type it like you would say it, or use the controls. Policy decides in code — never in a prompt.

Side

Symbol

Quantity

Variants tried

Read-only · no keys · no orders. Costs: 5bp fee/side + 80bp weekend-spread assumption. FCN leg illustrative, not a quote.

Hypothesis graveyard

Negative results are the most undervalued asset in quant research. Ours compound.

Method & limits

What we claim, what we do not.

Native history: Yahoo 2y daily (keyless). In-token weekends: Bitget public daily prints (keyless). No login, no keys, no orders — read-only by design.
Weekend spreads assumed 80bp / overnight 45bp / regular 8bp one-way; fee 5bp/side (promo). Post-Sept fee schedule TBA — flagged, not hidden.
Close→close proxies labeled per event; in-token drift shown only when ≥4 weekends measured, with confidence.
Beta/correlation need ≥20 return obs or fall back to 1.0/0 — never silently precise.
DSR withheld unless you declare variants tried. A passing audit is evidence of non-obvious-overfit, not future returns.
Qwen writes prose around deterministic cards (number-locked); it never emits a number first. Kill switch lives in code.