A million food products,
and not a single recorded death Korea's product manufacturing registry, all 1,066,415 rows · shelf metabolism · public data only

01The whole registry, downloaded

Deduplicated by report number, the snapshot holds 1,066,415 products from 29,134 manufacturers. One row per product: when it was filed, who makes it, what type it is, and whether production has ended.

Start with the pace. The past twelve months alone account for 139,097 filings still alive today — roughly ten thousand a month, about 380 new products a day. "New products are flooding in" is not a figure of speech in this industry. It is arithmetic.

02A registry with zero deaths

But there is no exit door. Not one product carries the discontinued flag, which suggests that when production stops or a manufacturer folds, the report is simply removed from the public view (I can't confirm the mechanism — no public document describes it; see the methods section).

Surely the official statistics count it somewhere. I checked: the 875-page MFDS Statistical Yearbook, the national statistics portal KOSIS, the national indicators site, the MFDS industry trends report, and the agency's open data portal (2026-08-12). No yearly count of new product filings exists at the product level. Neither does any count of discontinuations. What exists is a count of manufacturers, and a "production performance" item count — 77,127 for 2023 — which, per its own footnote, collapses every product a company makes within one food type into a single item. Different unit entirely.

One contrast worth savoring: for health functional foods, the agency's data catalog lists a separate "production ended" dataset. For ordinary food, no such entry exists at all.

Why this matters: plenty of food companies call their launches "experiments." But half of any experiment is the termination call — what died, when, and why, recorded so the next launch learns something. In this industry's public data, that half is structurally empty. And as we'll see, most companies' internal ledgers look the same.

03So I measured the age of the survivors

If you can't count the dead, change the method — measure the age distribution of the living. Line up all million products in production today by filing date, and the median lands in 2021. Half of everything alive entered the registry within the past five years.

Read this honestly. The curve rises to the right for two entangled reasons — more products were born recently, and older ones have died out of the frame. This registry alone cannot split the two. Splitting them would take a denominator: how many were filed each year. As §02 established, that number is published nowhere. So this piece claims no survival rate. Instead it builds a metric that doesn't need the split.

04Every shelf runs on a different clock

The metric: shelf metabolism = products filed in the past 24 months ÷ all products alive today. Numerator and denominator come from the same registry on the same day, so no missing denominator has to be guessed. The meaning is plain — of what sits on this shelf right now, how much arrived within two years?

Across the whole registry: 25.0%. One in four living food products is less than two years old. Split by product type, though, and the gap between the fastest shelf and the slowest stretches to almost ninefold.

Ready-to-eat meals — lunch boxes, sandwiches, kimbap — run at nearly half the shelf under two years old. On that shelf, a position is rented, never owned. At the other end, seasoned laver sits at 5.0% — a shelf where years-old products still hold their places. Beverages split internally too: mixed drinks turn at 36.5% while infusion teas turn at 17.7%. "A new beverage launch" names two entirely different events depending on which shelf it lands on.

The ordering survives a change of window. Recomputed at 12 months and at 36, no type swaps ends — figures in the methods section.

05The rush belongs to a few

I also counted who generates those 380 filings a day. The median manufacturer holds 9 living products; the top 10% of manufacturers hold 67.0% of everything alive.

Filing activity is more concentrated still. Only 53.4% of manufacturers filed even one new product in the past 24 months — nearly half filed none at all. Among those who did, the median filed 5, and 54.1% filed five or fewer. Meanwhile, of the 266,862 filings in those 24 months, 53.4% came from the top 1,000 manufacturers.

06What to actually do

Compute your own metabolism first. Same formula — SKUs launched in the past 24 months ÷ all SKUs currently sold. With a clean SKU master this is a five-minute calculation, and if it takes longer than five minutes, that itself is the first finding. Then compare against two numbers: your product type's value in the table above, and the registry-wide 25.0%. That baseline is nobody's opinion — it is the whole market measured with the same ruler.

One rule holds across all three rows: record your discontinuations. What was killed, when, and against what criterion. In a ledger that erases deaths — like this registry — the next launch learns nothing from the last one. A discontinuation log is not a record of failure; it is the cheapest experimental data this industry can accumulate.

Three choices made in this calculation, stated with their counterfactuals:

• Why 24 months. At 12, the value wobbles with a single holiday-season cycle and with the lag between filing and reaching shelves; at 36, you're measuring a three-year average rather than the current speed. Twenty-four months is the longest window that still hugs "now" while containing two seasonal cycles.
• Why the denominator is living products. Using yearly filings as the denominator requires counting dead reports — and this data contains no deaths. Drawing numerator and denominator from the same day's registry dissolves the problem. The price: this metric measures the youth of the shelf, not survival-per-launch, and that limitation stays.
• What I deliberately left out. The 103,142 filings from food-additive manufacturers — flavorings, mixed preparations — are excluded from the table. They are factory-to-factory trade, not a consumer shelf. For reference, their metabolism is 18.7%, slower than the consumer shelves.

07What this registry cannot see

A filing is not a launch. Products that were filed but never reached a shelf, or shipped once and vanished, are all in here. Shelf metabolism therefore leans high relative to true shelf rotation — the direction is knowable, the magnitude is not.

No time or cause of death. So no survival rate and no half-life can be computed from this data. Those numbers exist only inside companies that kept their own discontinuation history — which is exactly the shared rule in §06.

The kill-criteria inputs aren't out here either. First-repurchase rate after launch, shelf turnover speed — the raw material for a termination call mostly lives in distributors' settlement data. Getting it into your own books is the subject of the third piece in this series.

08What this piece is really saying

The registry records births and erases deaths — so this industry calls its launches experiments while owning no data on the half of an experiment that matters: the termination call. If you can't count the dead, measure the youth of the shelf. Metabolism = products filed in the past 24 months ÷ everything alive, with the market-wide 25% as the baseline. Above it, change the launch KPI from launch count to survival count; below it, move budget from new products to keeping the ones you have. Either way, record your discontinuations — the cheapest experimental data this industry can accumulate.