Three sheets per event. Locked before, answered on the day, graded after.
Each name is anchored to its own history. We take the last eight to twelve results-day reactions, split them into up moves and down moves, and work each side separately, because one statistic cannot size both a normal print and a tail.
The middle range runs from the median down move to the median up move, so it captures roughly half of that stock's prints by construction and leans whichever way the stock leans. The outer ranges open beyond it. The deepest range is set mechanically, never shallowed to flatter the record, and never set where the name could not actually trade.
The consequence is that the same words mean different numbers on different names. A contained reaction on Bolloré and a contained reaction on Vusion are not the same percentage, and they should not be.
Direction and magnitude are graded separately and never blended. Direction is whether the official close landed on the side the matched scenario called. Magnitude is whether it landed inside that scenario's range.
Four grades, and no others. Hit, the close is inside the range. Over-called, the move came but fell short of the range. Under-called, the move went further than the range. Miss, the close was on the wrong side or more than one range away. There is a half-point tolerance at a range edge, and no discretion beyond it.
We grade the scenario the release actually matched, not the one we leaned to. That scenario is fixed in the second sheet, before the open and before any close exists, which removes the possibility of choosing the flattering one afterwards. Whether our lean was right is reported as a fact in the verdict, and carries no grade.
There is no view on what any security is worth, no price target, and no recommendation. The scenarios describe how a print could trade and what to say in either case. They are written for a sales trader with ten minutes before the open, not for a portfolio construction process.
Everything is built from public disclosures: company releases, regulatory filings, exchange data and published analyst counts. No broker research is used and no issuer pays for coverage.
The published record begins on 1 September 2026, when the current method came into force. It is published in full, including calls that were wrong, and it is not filtered. It is here.
Earlier prints exist and are not published, because they were graded under a different convention and the two cannot be summed into one record.