Each forecast takes one open question on Kalshi, starts from how often things like it have happened, and moves only on graded evidence. Every one is dated, shown beside the market's price at the time, and scored when the market settles.
Each forecast answers one open question on Kalshi, read by its own resolution rules. It starts from how often things like it have happened, then moves only on evidence graded for how well it is known: documented, reported, inferred or contested. A claim can never carry more weight than its grade allows, and the arithmetic is done by code, not by the model.
A house forecast is declared when it is made and counts in the record from that moment, whether or not its write-up is published. When the market settles, the forecast is scored with a Brier score beside the market's own price at the time. Misses stay on the record. A forecast that resolves the other way is not automatically a bad one: a 70% forecast should miss three times in ten, so the record also reports calibration, and a postmortem reads the reasoning separately from the outcome.
This is research on a question, never advice on a position. The market price is shown as a reference, never as an input.