Voir Dire · Adversarial trading, on the record

Four AIs argue
each trade.
A Judge rules
and keeps the case against.

A Bull, a Bear, and a Risk analyst argue every scheduled catalyst from opposing sides. A Judge rules on the record, triggers the only order that can be placed, and preserves the strongest losing argument - on every verdict, including a decision not to trade.

Run a panel →Does it beat one model?
4
voices per catalyst
60
pre-registered bar
100%
case against preserved
Live · RMSTR macroBULLBEARRISK
NO_TRADEconviction 45 > 60
case against (BEAR) retained · ev 034d48627732b0fb…
The panel · same evidence, opposing cases

Three advocates argue. No single voice gets the last word.

BULL · case for

RMSTRUSDT likely rose after the 2026-09-16 macro catalyst because price was stretched below its short-term average (close 125.9 vs SMA20 133.4) while the broader trend remained constructive with SMA20 above SMA50, creating a mean-reversion setup into the macro event. Elevated volatility (ATR14 8.48, about 6.7% of price) and high volume supported potential upside expansion, and the prior 5-day decline of -5.7% indicated short-term selling was already extended and vulnerable to a positive macro surprise.

BEAR · case against

The post-catalyst evidence favors a bearish/avoid stance: RMSTRUSDT closed at 125.9, about 5.6% below its 20-day SMA of 133.4, and momentum is negative with a 5-day return of -5.7% and 20-day momentum of -8.7%. Volume was elevated at 23.97M and ATR14 was 6.7% of price, indicating volatile conditions. Although SMA20 remains above SMA50, the close below SMA20 after the catalyst suggests active selling pressure and downside continuation risk, making the asset best avoided on the long side.

RISK · constraint

Historical volatility is elevated: ATR14 is 8.48, about 6.7% of the 125.9 close. Price is below SMA20 (125.9 vs 133.4), while SMA20 remains above SMA50, giving a mixed trend. Downside momentum is visible with 5d return -5.7% and 20d momentum -8.7%. A defensible stop level is 113.2, equal to 1.5x ATR14 below close; a more conservative alternative is 108.9, equal to 2.0x ATR14 below close.

Strict cutoff - the panel never sees data at or after the catalyst date. No foresight, no leakage.

The ruling · the only trigger

The Judge is the gate. Remove the panel and no trade is possible.

NO_TRADE
45 / bar 60
conviction below the bar

“The setup is mixed: price is below SMA20 with negative 5d and 20d momentum, favoring downside, but SMA20 remains above SMA50 and elevated ATR makes whipsaw likely. No clear directional edge.”

The case against · never erased

The losing argument is kept with equal standing.

BEAR · the case against

“The strongest losing argument is bearish: close is 5.6% below SMA20, 5d return is -5.7%, 20d momentum is -8.7%, and elevated volume suggests active selling, but this is not enough to short given the mixed trend.”

On every verdict - including NO_TRADE - the strongest losing case is recorded as a first-class part of the decision. Not a footnote.

Honest or nothing

We report the out-of-sample result even when the panel loses.

The claim is provenance, not alpha. The same pipeline runs minus the debate - a single-LLM ablation - on events it was never fit on, and both arms are shown either way.

Panel (4 voices)
+1.00% avg P&L · OOS
hit 50% · 2 trades · sharpe~ 0.57
Single-LLM ablation
0.00% avg P&L · OOS
0 positions cleared its own conviction bar

A directional-signal study on committed rToken OHLCV, not executed returns and not significance-tested. Verdict PANEL_BEATS_OOS. The panel's value is the auditable record, not a promise of profit. See the full event study →

Recompute, don't assert

Every number traces to committed data and a hash.

A verifier can recompute the whole chain from committed evidence. A claim that can't be recomputed is never shown. Orders hit a real API - paper, never a fabricated fill.

Open the console →Verify the record