Korea's exchange prices the same gold twice. The retail bar is not the dearer one.

The Korea Exchange lists gold in two sizes. One contract is a 1kg bar of 99.99 percent gold; the other is a 100g mini bar of the same metal at the same purity. Two listings, one substance.

We have counted this pair once before, on a different axis: in August we reported that turnover in the retail-sized 100g contract rose through the price decline, running 72.3 percent above the 2025 daily average. That piece measured how much of each contract traded. This one measures what the two of them cost.

The obvious assumption is that the small bar costs more — smaller units usually do. We counted 49 sessions to see whether that is true on this exchange. It is not, or at least not reliably.

The gap is small, and it changes sign

A median of 0.015 percent is, for practical purposes, no premium at all: on a 190,000-won gram price that is about 29 won. And the mini closed below the 1kg bar on nearly half the sessions, which is the part that rules out a standing retail markup inside the exchange.

This is worth saying plainly because a single session would have told the opposite story. On 9 September the mini closed at 193,000 against 191,000 — a 1.047 percent premium, and the widest gap in the whole window. Read that day alone and you would report that Korean retail-size gold trades a percent dearer. It does not; that day was the outlier we happened to look at first.

Five days when the same metal moved in opposite directions

The more interesting count is not the level but the direction. Each listing reports its own daily change, so we counted how often the two agreed.

Gold did not do two things on those days. The same metal, on the same exchange, in the same session, closed up in one size and down in the other. Four of the five fall in a two-week stretch at the start of September, which is also where the widest gaps sit.

The likely reason is depth, and we can show the depth without claiming the mechanism: the 1kg listing turned over a median of 221,569 units per session against 10,983 for the mini — about twenty to one. A thin book moves on fewer orders. But this dataset holds closes and reported changes, not order books, so that is where our evidence stops and we are leaving it there.

Why this had never been counted

The data was already open and approved, and the collector for it was written on 4 August 2026. When we looked today, archive/raw/commodities held zero files. The collector had never been run.

That is a worse failure than not having the data, because the header of that same collector records the previous version of the mistake: “the data was open and we had not attached a collector, so we believed there was no source.” This time the collector existed and nobody pointed it at anything.

We have now pulled 1 July to 9 September — 49 sessions, 245 rows across gold and oil — and put the collection under the same daily watch as the bond and broker archives. There is no way to recover the sessions before 1 July that the exchange no longer serves, and we are not going to pretend the ten weeks we have is a year.