The wait for a load is set by where its bytes live, so the same loop climbs a four-step staircase as the table grows.
Question: Why does the same arithmetic run at so many speeds?
Published Published: the measured pointer-chase ladder of seed v1c10 (NUMBERS.md section B), on an Apple M1 Pro, two runs of the same protocol. Each rung is the wait per dependent load above that run's 4 KiB chase, so the loop's own cost of 15.2–16.2 ns per load is already removed. The bar spans the two runs; a rung is a range, not a digit. The frames move the gold highlight up the ladder from the chosen size; each frame names the rung, its band, the whole-walk total and the level it belongs to.
Frame 1 of 4: 2 MiB rung2 MiB: 20–22.6 ns per load, 20–22.6 µs per walk. The step from 512 KiB is 14× and 11.8×, and this rung opens the second level.
2 MiB rung. 2 MiB: 20–22.6 ns per load, 20–22.6 µs per walk. The step from 512 KiB is 14× and 11.8×, and this rung opens the second level.
Keyboard: Space plays or pauses, Left and Right step, Home goes to the first frame, End goes to the last frame.
Text description of this lab
A ladder of eight rungs, 4 KiB to 256 MiB, on a log scale of wait per load. Each rung is a band spanning two runs, so it is a range, not a digit. The wait is flat through 512 KiB, then jumps at 2 MiB and at 8 MiB; the top rung reaches 254 nanoseconds per load, 157 and 139 times the 512 KiB rung. A gold column climbs the rungs as the readout names size and wait.