How the desk learns
Journal
The model is not retrained. Every close produces a graded review (orders, how each veto turned out, calls versus the index), and patterns go into a lesson book: a hypothesis becomes confirmed after three supporting cases and is retired when the evidence turns against it. Every analyst run reads the lesson book first. Lessons sharpen judgement; they never loosen risk limits.
Daily reviews1
Analyst reports1
Confirmed lessons03+ supporting cases
Hypotheses0being tested
Lesson book
No lessons yet. They need evidence first: the first candidates come once vetoes have outcomes.
Latest reflection
Full review →First session, a partial day: entries at about 14:15 New York time, close at 16:00.
- Difference from the control: the LLM review vetoed LLY and MRK (results Thu 29 Oct) and AAPL (results Mon 2 Nov),
shrank MET (results 4 Nov) and bought KO (beat and raised guidance on 27 Oct), which the control did not hold.
The control held LLY and AAPL instead of KO.
- Result: both accounts were flat into the close (main +0.01%, control -0.01%) while the S&P 500 gained about 0.19%
from the entry time. That is too small a difference to mean anything.
- Vetoes: no outcomes yet. LLY and MRK report before tomorrow's open, so the 1-day outcome of the earnings vetoes
is known at tomorrow's close. This is the first real test of the "no entry before earnings" rule.
- No stop-outs, and no calls have elapsed outcome windows yet.
- Process note: in earnings season, 3 of the 8 top S&P 100 momentum names had results within 3 trading days.
The earnings check is the most important part of the US review right now.
- No lessons added: no evidence yet. Candidate to watch: whether vetoing names into earnings beats holding them.