Corporate life · Murphy's laws
The Lindy effect: why nothing is more permanent than a temporary fix
The short version: for things that don't wear out, such as software, procedures and workarounds, the longer something has survived, the longer it is likely to keep surviving. Under a simple model where lifetimes follow a power law, a thing that has lasted a year has a 50% chance of lasting another year and a 10% chance of lasting ten more. In our simulation, 77% of temporary fixes disappear within a month, but the survivors are hard to kill. If a workaround has made it past its first year, plan for it to stay.
Kevin finds the file while looking for something else. It's called "TEMP_payroll_bridge_DO_NOT_USE.xlsx". It was created in 2016, it runs a step of the monthly payroll, and the comment in cell A1 says "Delete after migration (Q3)". It doesn't say which Q3.
His boss confirms the migration is "on the roadmap". Darren points out that the file is older than the roadmap.
The law
The Lindy effect is named after Lindy's, a New York deli where comedians gathered. In a 1964 essay, the writer Albert Goldman noted that a comedian's future career seemed to depend on how much material he had already used. The mathematician Benoit Mandelbrot and later the writer Nassim Taleb turned it into a general idea: for things that don't age the way bodies do, life expectancy grows with age.
Milk, cars and people wear out, so each extra day brings the end closer. A book, a procedure or a spreadsheet doesn't wear out. Each extra year it survives is evidence that it's useful, embedded or simply forgotten, and all three help it survive more years.
How long will it last?
A simple model that produces the Lindy effect gives lifetimes a power law: the chance of surviving to some age falls in proportion to that age. In the most common version, the chance that something reaches k times its current age is 1 in k:
| If it has already lasted | Chance it lasts at least as long again | Chance it lasts 10 times as long in total |
|---|---|---|
| 1 week | 50%: another week | 10%: 10 weeks |
| 1 year | 50%: another year | 10%: 10 years |
| 10 years | 50%: another 10 years | 10%: 100 years |
So the typical remaining life equals the age so far. The astrophysicist J. Richard Gott used the same logic in a 1993 paper in Nature: with nothing else to go on, there's a 50% chance that the future of something lies between a third and three times its past.
Most die young, the rest live for ever
This doesn't mean every temporary fix is permanent. In our simulation of fixes with power-law lifetimes, 77% are gone within the first month: someone does the proper job, or the problem goes away. But that early cull is exactly what makes the survivors so durable. By the time a workaround has reached its first birthday, it has passed every test that kills workarounds, and among the fixes still running after ten years, a third are more than a year old.
Statistics for noobs: survival depends on what you condition on
The survival rate of something that has already survived is different from the survival rate of something new. For things that wear out, it falls with age. For things that don't, it rises, because the weak ones have already been removed. Before guessing how long something will last, ask which kind it is.
Where it costs you money
Investors are told that a company's age says nothing about its future. For a business, it's not a guarantee either, but firms and products that have survived several cycles have shown something that a start-up hasn't yet. Treat "new and exciting" and "old and boring" as different bets, not the same bet with a different label.
The moral
Kevin raised the file at the weekly meeting. The decision was to replace it properly, starting next quarter. In the meantime, to reduce the risk, the team agreed to build a small temporary script that checks the temporary spreadsheet. The script was written that afternoon. It's called "TEMP_check_v1".
What this can't tell you
- The power law is a model. Real lifetimes may fall off faster or slower; the figures above use the simplest version.
- Things that wear out don't follow it. Hardware, people and contracts with an end date age; Lindy applies to ideas, code and habits.
- Age isn't the only clue. If you know a fix will be switched off on a date, use the date.
The spreadsheet, Kevin's boss and Darren are an illustration. The Lindy effect is named after Albert Goldman's 1964 essay in The New Republic; Gott's estimate is from Nature (1993). The table uses a power-law model in which the chance of reaching k times the current age is 1 in k; the 77% figure is from our own simulation of 100,000 fixes with a minimum life of a week. Educational content, not business or financial advice.