an analyst that cannot bluff

Every figure it quotes is checked against its own context.

The analyst receives tonight's full board, each name's historical accuracy, the shrinkage weights in force and the walk-forward error table — and nothing else. No browser, no memory, no prices of its own. Then every number it cites is matched back against that context before it renders. A figure the model invented is a figure you never see.

Research0 names · — of context
start here
waiting for the board before it can answer anything…
How an answer is built
  1. 01
    The board is assembled

    Live quotes from Bitget, the session close for each name, and the fair value our frozen calibration produces at this hour.

  2. 02
    History is attached

    Each name carries its own track record — how wrong we have been on it, how wrong the venue has been, and how much its quote typically overshoots.

  3. 03
    The model is given that and nothing else

    No browser, no memory, no prices of its own, and an instruction to say so when the context does not answer the question.

  4. 04
    Its figures are checked

    Every number it cites is matched back against the context. Anything it could not have read there is stripped before you see it.

The complete prompt context is published, not summarised — — characters of it. Open it and you can check any figure in any answer above against the line it came from.