fourteen days

how long a fourteen-day average remembers a mistake

the four mistakes

what the rows say

What this is. Four prints in SPY's daily history were wrong, and a person corrected them by hand; the store with the corrections is layer 1.0.2, the store without them is set 1.0. Each wrong print changes one day's range. atr14 is Wilder's average of the range, and every day it keeps thirteen fourteenths of yesterday. So a mistake does not leave; it shrinks. This page compares the two stores' atr14 on every trading day and draws the difference on a log scale, where shrinking by the same fraction every day is a straight line. The empty stretches are the days the two stores agree to the last bit. Every number here is computed from the rows the page carries; none is typed in.

Made from the two baked SPY stores in the io repo (provenance/rooms/SPY/derived/historical.json and layers/1.0.2/derived/historical.json, both built 2026-09-19). fourteen-days-INTENT.md is what I expected before I looked. I had the slope and the first mistake's first two lines exactly, and the rest wrong in useful ways. The two tails that finished end with atr14 between 1 and 2, where one bit is 2.2e-16; that was checked outside the page.

···