This dashboard is a read-only view of the
xaitpe forward-validation experiment. It tracks one
pre-committed test (Primary, §12) and two
side experiments (Zoo, LLM), each running daily on its own
OANDA practice sub-account — no real money.
Window
2026-05-26 → ~2026-06-26. One month. The cron fires nightly at 14:30 UTC.
Primary (§12 validation)
The finalised TrendXSBlend strategy on the
OANDA-tradeable universe. The point of the run is
cost & execution fidelity, not P&L: does
live execution match what the backtest assumed?
Zoo (sidecar)
A combined portfolio of four control strategies —
BuyAndHold, TharpRandom, InverseBlend, ShortTermMomentum.
The account holds the sum of all four; per-constituent
virtual P&L is tracked alongside the real book.
LLM (sidecar)
Each day, Anthropic Claude is given a universe summary and
asked for target weights, subject to hard constraints.
Picks trade live, with a fallback to the previous day's
targets on parse / API failure.
Honest framing
One month of a daily strategy's P&L is noise.
Sharpe, win rate, alpha — none of these are computed or
displayed here, because the window is too short for any
of them to mean anything. What this dashboard does show:
NAV, daily / MTD P&L, max drawdown, modelled-vs-observed
cost fidelity, positions, trades. Paper, not a
profit play.