How these systems actually behave.

One long read and two shorter pieces, each following a single failure or constraint far enough down that the design decisions around it stop looking arbitrary.

• LONG READ
  6 MIN

  A timeout is not a fill

  Every OMS has to decide what to write when the socket goes quiet. Most get this wrong once, lose money, and then build the thing I'm describing.
• NOTE
  3 MIN

  A fill cannot use the signal bar

  Two things kill backtests before the idea is even wrong: peeking at future data and pretending Indian costs are one flat rate.
• NOTE
  3 MIN

  Not every hypothesis graduates

  Ideas die in backtest. That's fine. The gates are what keep bad ones away from client money.