HUNT-PERF gains the trap #2223 was filed on: the last cell of a sweep has no neighbour past it, so the outlier check that caught a 1,091.9 ms reading mid-axis had one side to hold the 226.2 ms reading at 128 KB against, and #2242 found that figure reproduces in neither arrangement it took. the entry says what is known and no more, since #2247 withdrew the claim that the figure rested on one reading; it names the arrangement of each ratio, 4.8x in the call and 3.4x as a process, and records the refutation leaning on the same end cell until #2242 carried the axis to 1 MB. the remedy is stated as a property rather than a number of sizes: a cell a finding is filed on needs a neighbour on both sides or a repetition spread reported with it. prose only with no behavioural seam, so there is no mutation proof; the generated work-counter cell is untouched and work_counters reads 12 passed before and after. the workspace suite is green (4254 passed over 135 binaries, 10 ignored) (#2245) 1e60a43b · dbf3dbe6…
Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.