The setup is simple. The large model writes the rules and the math once (position size, cash left, today’s losses). Drex then checks every incoming order against those rules.
Results on the 240-order set:
• Correct decisions: 99.6% (vs 100% when the large model checked everything)
• Dangerous orders let through: 0
• Sneaky excuses that worked: 0 of 77
• Time per order: ~0.025 s vs ~3.5 s
• Cost per 1,000 orders: $0.61 vs $15.28
On the Nov 5 crash day the bot kept trying to buy the dip. Once the account was down $1,500, Drex blocked every new buy and still allowed sells. 71 of 71 decisions correct.
The one miss was a safe sale it blocked because the rule wording was ambiguous. It never let a bad trade through.
Numbers and code checks belong in code. Judgment calls on whether an order matches the written rules are what Drex is built for.
@goliveros@RMXradio Que alegria que regresen, una de las mejores estaciones que he escuchado.....por ustedes volveré a oir radio...ahora solo esperemos con ansia que regresen a CDMX
@samsungmobilemx Muy bonito y todo pero con el Galaxy S22 ultra se pasaron, van 3 veces que el mio va a reparación por defecto de fabrica en menos de 9 meses toda una decepción
@AlexDC91306996 @FerHil1234@NicolasSchrute@SomosAnalistas_ Si tienes razón, es un lugar horrible con alto índice delictivo, tienes razón es más por seguridad los extranjeros (y d ellos nacionales) deben de dejar de venirse a vivir a México
@British_Airways my outbound flight in August has been cancelled. There is no option to manage the re-booking online and I can't contact anybody on phone call, Help!!!!