# Deviations log — Supertrend against a moving average

Every departure from `preregistration-supertrend-vs-ma-2026-11-05.md`, dated, with the reason. The pre-registration itself is never edited.

## D1 — 2026-11-05 — "the public repository"

**Pre-registered:** "The commit that adds this file and the comparison's code is pushed to the public repository before the run."
**Fact:** the repository is private. The commit (`f71db9e1b23b293b80137fa42566008e7024d5fc`, 2026-11-05T22:25:32Z) was pushed to it before the run, and the run (2026-11-05T22:27:29Z) recorded that commit and a clean working tree in its output. Nobody outside can inspect the commit there.
**Deviation:** none in the analysis. The claim of public evidence is withdrawn. What can be checked from outside is the pre-registration as published on the site, byte for byte the registered file (`/study/supertrend-vs-moving-average-preregistration.md`), the full output with the commit it recorded, and the code if the repository is opened later.
**Reason:** an error of fact in the registration text, found before publication.

## D2 — 2026-11-05 — the outcome sentence, with the match on the edge of the grid

**Pre-registered:** "In past data Supertrend(10, 3) and a moving average that traded as often were not clearly different in how far they fell."
**Result:** ladder N was reached, and the turnover match fell on the edge of the grid: the closest length, 250 days, made 871 round trips against Rule v1's 682. No moving average on the grid traded as often as Rule v1.
**Deviation:** the phrase "a moving average that traded as often" is replaced by "the closest-matched moving average". The published sentence is **"In past data Supertrend(10, 3) and the closest-matched moving average were not clearly different in how far they fell."** The return clause ("Supertrend's return was higher.") is unchanged. The registered caveat stays beside the sentence. The ladder category, the four conditions and every number are unchanged. The run's own output files keep the registered string, as the record of what the run produced; the page and anything that quotes the outcome use the replacement and show the registered wording with it.
**Reason:** the registered phrase would claim a match the data does not show; the replacement claims less, not more. Proposed by the engineer, accepted by the owner on 2026-11-05.

## Notes (not deviations)

**N1 — the match fell on the edge of the grid.** The registration foresaw it: the closest length by trade count is 250 days, with 871 round trips against Rule v1's 682. It is reported as found, with the registered caveat. The registered sentence speaks of "a moving average that traded as often"; with the match at the edge that phrase overstates the match, so wherever the sentence is shown the caveat is shown with it. A replacement that claims less ("the closest-matched moving average") is the owner's to accept, as D1 of the crypto study was; until then the registered wording stands. **Accepted on 2026-11-05: see D2.**

**N2 — a check after the run, before publication.** The moving-average positions of Bitcoin and Ether were compared with known history (Bitcoin's close below its 200-day average on 2021-05-19; an average computed independently for 2022-06-15). They agreed. No code and no result changed.

**N3 — "time in the market".** As reported it is the harness's own measure: the share of days on which the basket held at least one position (95.2 % and 95.0 %). It is not the average share of the basket invested.
