Epictetus had an iron lamp stolen and decided the fear of losing it had been the real cost. The FDA has kept a fifty‑year ledger of the same lesson: every time a company opens a device category, a clock starts on somebody else clearing the same thing.
Primary source: Epictetus, Discourses I.18.15 (Oldfather trans.) — the stolen iron lamp
Grounding data: MIMI Labs · FDA 510(k) Premarket Notification database, cleared submissions 1980–2026
Epictetus kept an iron lamp beside his household shrine. Someone came through the window and took it. He did not get angry — he asked himself why he had owned a thing he was afraid to lose, and said that tomorrow he would find one of earthenware.
That is a hard argument to test on a philosopher. It is easy to test on a regulator. The FDA 510(k) database is a fifty-year public record of people clearing a device by pointing at a device that already exists. Every device gets a product code — a category. Somebody is always first into a category. And then the file tells you exactly what happened next.
Every category opened since 1980, whole file
Categories opened
2,772
Clearances inside them
38,822
Share held by the company that got there first
10.7%
Median days to a second, different company
750
Two years and change. That is the median gap between the first clearance in a device category and the first clearance by anybody else. A third of categories — 935 of 2,772 — never got a second company at all.
1,558 categories, one dot each
Each circle is an FDA device product code with at least three cleared submissions, plotted at the year it opened. Vertical position is how long the first company had the category to itself before a different company cleared into it — log scale, so one screen holds two weeks and forty years. Circle area is the total number of clearances the category has attracted since. The strip along the top is every category where nobody else has ever shown up.
Read the readouts against the panel above, not instead of it. Requiring three clearances throws out 1,214 near-empty categories, and almost all of the never-copied ones go with them: 66% of all 2,772 categories ever saw a second company, but 97% of the 1,558 plotted here did. The first mover’s share moves the same way — 10.7% across the whole file, 7.7% here — because a category with a single clearance is a category the first mover owns outright. The chart is a picture of categories where something actually happened. That is a choice, and it is the one doing most of the work.
Highlight a review panel
3
1980
Categories shown
—
Ever got a second company
—
Median days alone
—
First mover’s share of clearances
—
highlighted paneleverything elsecircle area = total clearances in the category
—
How long you actually get
The same filtered categories, read as a clock: the share that had picked up a second company by t months after opening. It answers the only question that matters if you are sitting on something in a folder — how much time does being first actually buy? Because it inherits the three-clearance floor, this curve runs well above the whole-file version, which reaches 22% at a year and 50% at five.
—
The 80/20 lens
The first company into a device category ends up holding 10.7% of everything ever cleared inside it. Not 80%. Not 50%. Roughly one clearance in ten. Being first is worth about a year of quiet and a tenth of the category — and it is worth that whether or not you guarded it, because a 510(k) summary is a public document by design.
Which means the protection was never doing the work. What the first mover actually had — the reason to believe the category should exist before it existed — is not in this file. It is not in any file. Nobody has worked out how to clear a 510(k) for judgment.
Where this dataset is thin
Right-censoring is real and it cuts toward the story. A category opened in 2019 has had six years to attract a second company; one opened in 1985 has had forty. That inflates the “never” band for recent years and deflates the median for old ones. Drag the “opened after” slider past 2010 and watch the median gap fall — most of that fall is the calendar, not a real acceleration. This is the single most seductive artifact in the file.
A product code is not a product. FDA categories are administrative buckets that get split, merged and renamed. Two devices in one code can be barely related; one device can straddle three codes. “Second company in the category” is a decent proxy for “somebody else made your thing,” not a measurement of it.
The small-n trap, and the filter that hides it. The three-clearance floor is already a survivorship filter — it is why the “ever got a second company” readout starts at 97% instead of the file’s true 66%. Push min-clearances to 15 and roughly two-thirds of the remaining dots vanish; if a pattern you liked disappears with them, it was never a pattern. Applicant names are matched on raw text, so a company that reincorporated or was acquired can read as a second entrant when it is the same shop.
This measures clearance, not building. The 510(k) clock starts when a company files, not when someone opens an editor. The essay’s claim — that the cost of a build has collapsed — is not what this data shows. What it shows is the older, slower version of the same fact: even when a build took a year and a purchase order, the thing itself was never the scarce part.
So take it out of the folder
The lamp story has a footnote. After Epictetus died a collector paid a small fortune for the earthenware lamp, convinced the wisdom would come to him by its light. He bought the crockery of a man whose entire teaching was that the crockery is worthless.
Somebody will always want the lamp. Nobody has ever worked out how to steal the reading.
The median category in this file gave its founder 750 days and one clearance in ten. Whatever is in your folder is holding less than that, and it is holding it against nobody. Send it ugly this week.