# Is gold sold down into the London AM fix?

**Verdict: supported** — F = 0.6410 against a placebo centre C = 0.4840, one-sided p = 0.0000, on 234 London days.

Bars: 73902 one-minute rows, 2025-01-02 07:00:00+00:00 to 2025-12-31 11:59:00+00:00 (UTC as stored; every window below is on the Europe/London clock). Price is the mid of the bid and ask closes. 257 London days have at least one bar, 2025-01-02 to 2025-12-31, and 14 Monday-to-Friday London dates in that span have no 10:15 or no 10:30 bar at all (holidays are counted as holes: this is the fetch's coverage, not a trading calendar).

## The test

| | |
|---|---:|
| days kept, n | 234 (of 257 London days seen) |
| down days (run_in < 0) | 150 |
| F, fraction negative | 0.6410 |
| C, placebo centre | 0.4840 |
| null sd of F | 0.0324 |
| one-sided p | 0.0000 |
| critical count k01 (p < 0.01 at C) | 132 |
| certified floor q\* | 0.619 |
| null's own size at k01, exact binomial | 0.0085 |
| null's own size at k01, from the draws | 0.0077 |

run_in is ln(mid(10:30) / mid(10:15)) x 10,000 bp. The null is the pre-registered placebo-window null: 10000 draws, each picking one start minute per day uniformly from the 226 minutes 08:00-11:45 London that have a bar exactly 15 minutes later (10:15 is one of them and is not excluded). A day survives if its 10:15 and 10:30 bars exist and at least 180 of those starts are valid. Seed 20260912.

## Days dropped by the floor

| reason | days |
|---|---:|
| placebo pool under 180 of 226 | 12 |
| no 10:15/10:30 bar and placebo pool under 180 of 226 | 11 |

## What this sample could have concluded

The sealed branch table (`rule.branch_table`) at the observed n = 234 and C = 0.4840: k01 = 132, k05 = 127, q\* = 0.619. Each row is the chance this many days would have landed in each branch had the true down-rate been that value.

| true down-rate | supported | killed | inconclusive |
|---:|---:|---:|---:|
| 0.50 | 0.0289 | 0.8929 | 0.0782 |
| 0.55 | 0.3572 | 0.3855 | 0.2573 |
| 0.60 | 0.8822 | 0.0325 | 0.0853 |
| 0.63 | 0.9837 | 0.0026 | 0.0137 |
| 0.67 | 0.9997 | 0.0000 | 0.0003 |
| 0.70 | 1.0000 | 0.0000 | 0.0000 |

## Descriptive only — none of this can move the verdict

**The mean.** MU = -2.4418 bp, per-day sd 9.5368 bp, placebo p = 0.0002 (the fraction of draws whose mean is at or below MU; the null's own mean centre is +0.1589 bp).

**The auction itself, 10:30 to 10:40.** Mean +0.0872 bp, negative on 47.9 % of 234 days (0 kept days lacked one of its two bars).

**After it, 10:40 to 10:55.** Mean -0.0180 bp, negative on 47.0 % of 234 days (0 kept days lacked one of its two bars).

**Busy, not just lower.** Median |run_in| = 5.2907 bp against a median |placebo return| of 5.3027 bp, pooled over all 52760 valid placebo windows.

**By weekday** (descriptive; the rule fixes no weekday claim).

| weekday | days | F | mean run-in |
|---|---:|---:|---:|
| Monday | 45 | 0.7333 | -4.668 bp |
| Tuesday | 48 | 0.6250 | -1.280 bp |
| Wednesday | 49 | 0.5510 | -1.889 bp |
| Thursday | 46 | 0.6304 | -2.126 bp |
| Friday | 46 | 0.6739 | -2.381 bp |

**By clock regime** (the London UTC offset of that day's 10:15; descriptive, and the split is small enough to be noise).

| regime | days | F | mean run-in |
|---|---:|---:|---:|
| GMT | 101 | 0.6436 | -2.440 bp |
| BST | 133 | 0.6391 | -2.444 bp |

Method, windows and kill rule: `rule.py`, sealed before a single tick of this year's data was on the server. This file is written by `analyse.py` and never by hand.

This is a test record, not advice.