What do grades A, B and C mean?

How many of the five guards a study cleared. A cleared them all with room to spare, B cleared them all at the minimums, C failed exactly one. Two failures gets no grade at all.

The guards are fixed thresholds, the same for every study, so a grade means the same thing everywhere it appears.

To be graded at all, a study must clear all five:

• At least 30 signals. Below that the sample cannot support any of the other numbers.
• An edge of at least 10 percentage points over the all-days baseline.
• The same direction in at least 70% of qualifying years, where a year qualifies by carrying at least 10 signals of its own. A flat year counts as disagreement, not as support.
• At least 3 qualifying years to measure that consistency over.
• At least 2 signals inside the last trading year, so a condition that stopped firing in 2022 is not still being described in the present tense.

Then:

• Grade ACleared every guard with room to spare: 100 or more signals, an edge of 15 percentage points or better, and every qualifying year agreeing. Chance almost never produces an A.
• Grade BCleared every guard, at the minimums. Solid, and worth reading — but shuffled data passes Bs too, which is why a B should be read next to the chance figure rather than on its own.
• Grade CFailed exactly one guard. Kept as a lead, never badged as an edge, and never shown on the desk row.
• No gradeFailed two or more. It goes in the drawer and is not offered as a finding.

Studies that measure a size rather than a rate — how big the move was, not how often it happened — are graded on the same skeleton with one threshold swapped: instead of a 10-point edge it wants a median shift of at least a quarter of the baseline's usual spread, and an A wants four tenths.