The cobra story is unverified. Hanoi's rat bounty is documented.

A reward can make a problem worse when keeping the problem alive pays better than ending it. That is the cobra effect. The distinction that does the work is between the measure you pay for and the outcome you want. Rat tails handed in can be counted. Fewer live rats is the goal.

In the familiar story, British officials in colonial Delhi paid for dead cobras, people bred snakes to collect the bounty, and the snakes were released when the payments stopped. The sources reviewed here do not document those farms. Hanoi’s 1902 rat-tail bounty has a stronger paper trail, including accounts of rat farming. Both stories illustrate the same incentive problem. They do not deserve the same confidence as history.

What does “cobra effect” mean?

A perverse incentive rewards behavior that works against the goal it was meant to serve. The cobra effect is the case in which that behavior makes the original problem worse.

A bounty pays a fixed amount per unit of evidence, such as a dead snake or a rat tail. The tail stands in for the outcome the payer wants, which makes it a proxy. A KPI, or key performance indicator, is a proxy too, and it carries the same risk once pay depends on it. The risk is that someone can improve the rewarded number while the underlying problem stays the same or gets worse.

Where does the cobra story come from?

The name is commonly credited to the German economist Horst Siebert and his 2001 book Der Kobra-Effekt. A catalog record confirms the book; its text was not consulted for this article. A 2012 episode of Freakonomics Radio retells the Delhi version, with people farming cobras to collect a bounty.

A retelling shows that a story travels. It does not show that the story happened. For that, you need records closer to the alleged events.

What do the records actually show?

Snake bounties were real. In 1887 the Government of Bombay asked the Bombay Natural History Society about the rewards it paid for killing venomous snakes in Satara and Ratnagiri, and about rumors that people bred snakes to turn them in. The Society’s honorary secretary, H. M. Phipson, replied on 8 June 1887 that “such a thing is highly improbable”. To his knowledge, cobras had never been known to breed in captivity. That was his judgment at the time.

Phipson offered a duller explanation for the rumors. People collected wild snake eggs and kept them until they hatched, then claimed the reward if the young were venomous. He thought that practice “should be encouraged”. Hatching collected eggs is not the same as breeding snakes for a bounty, and his reply kept the two apart.

A report by the Friends of Snakes Society, an Indian snake conservation group, traces the breeding rumor to an 1873 newspaper allegation about a snake bounty in the Madras Presidency. It also discusses an 1877 Bombay government resolution that predicted breeding without documenting a case, and an 1891 note in which Sir Joseph Fayrer dismissed the objection that bounties would encourage breeding. The report says its search found no contemporary documentation of cobra-breeding operations and no prosecutions. The breeding rumors it traces come from Madras, Bengal, and Bombay. None comes from Delhi.

Read the report as an advocacy source: it argues for retiring the story. Of the historical documents it cites, only the 1887 journal page was checked independently for this article. And missing records do not prove that nobody ever bred a snake for a reward. The claim the evidence supports is narrower than the legend: bounties existed, breeding rumors existed, the naturalists consulted in 1887 doubted the rumors, and the familiar Delhi version remains unverified.

What happened in Hanoi in 1902?

Here the paper trail is stronger. Historian Michael G. Vann describes finding forms in the French archives that recorded rats reported killed in Hanoi from April to July 1902. He published the episode in the journal French Colonial History in 2003, and later as The Great Hanoi Rat Hunt, a graphic history with illustrator Liz Clarke. This article relies on his published essay and interview, not on the archive itself.

Vann’s accounts describe this sequence:

1. The sewers the French had built gave rats shelter. As plague spread, rats were coming up through toilets in the French quarter.
2. When workers refused to hunt them, officials offered a bounty. They soon accepted tails alone, to avoid handling possibly infected carcasses.
3. The tail counts looked like success. But they recorded what was handed in, not how many live rats remained.
4. In July an official reported a tailless rat in the suburbs. Investigations turned up rat farms and a network bringing rats into Hanoi. Vann describes people cutting off tails and releasing the rats to breed.
5. Officials abandoned the bounty. Vann also describes the quarantines that followed and the burning of sick people’s homes and belongings, measures that angered the Vietnamese community.

The evidence supports an account of people gaming the bounty. On its own, it does not tell us the scheme’s net effect on the rat population. A tail on the counter was weak proof of a dead rat, and weaker proof still of fewer rats in the city.

How do you grade a famous story before you repeat it?

Stories need the same separation. An explanation of how an incentive could backfire does not establish that it did. LEVEL is a five-step check for deciding how hard to lean on a claim. It comes from The Evidence-Aware Life.

• Locate: name the claim in one sentence, including what supposedly caused what. If you cannot, you are grading packaging.
• Estimate: set the bar by what being wrong would cost, not by how badly you want the claim to be true.
• Value: name the rung the support actually sits on, then check its quality, independence, and relevance. The book’s Evidence Ladder has seven rungs: assertion, anecdote, authority, mechanism, study, convergence, and your own test. They are kinds of support, not a quality scale.
• Expose: state the gap between the support you hold and what the decision requires.
• Lean: choose a proportionate next step: a bounded action, the right help, or a justified deferral. For a story, that includes how firmly you word it.

Here are both stories through the LEVEL check:

Step	Delhi cobra story	Hanoi rat bounty, 1902
Locate	A Delhi cobra bounty led people to breed cobras; ending it led them to release the snakes.	A tail bounty led people to keep, breed, and bring in rats.
Estimate	Low stakes as an aside. Higher if you teach it as history or use it to justify a policy.	The same.
Value	Mostly retellings, plus a plausible mechanism. Contemporary records show bounties and rumors, and naturalists doubting the breeding claim.	Vann’s published accounts of archival forms and official investigations.
Expose	No direct support for the Delhi farms in the sources reviewed here. Missing records cannot rule them out.	One historian’s published accounts, not the archive itself. They do not establish the net effect on the rat population.
Lean	Present it as an unverified parable.	Cite Vann for the documented gaming of the bounty.

If your team uses either story in a deck, record the source and its limitation beside the claim. An evidence ledger keeps those qualifications attached when the story gets reused.

What do both stories teach about paying for a proxy?

Hanoi’s bounty paid for tails delivered to officials, not for fewer rats in the sewers. Once a tail earned money, a rat released without its tail could keep breeding rats that had tails. The count could rise while the source of future payments stayed in business. The cobra story runs on the same mechanism, which is why it feels true. A plausible mechanism still cannot supply missing records.

Consider a hypothetical software company that pays its support team a bonus per ticket closed. The proxy is a closed ticket; the goal is a customer whose problem is solved. If a reopened ticket earns another bonus when it is closed again, closing tickets early could pay repeatedly without solving anything. Whether that can happen depends on the payment rules. It is a loophole to test, not a claim about how support staff behave.

The cost can also land outside the rewarded number. In this example, customers could spend more time chasing help while the team records more closures. That would be a downstream effect: a cost that arrives later, lands outside the measure you optimized, and had no owner when the bonus was designed. Checking the outcome means looking past the count that sets the bonus.

The four-line check before you set a bounty, bonus, or KPI

Line	Question	Hanoi, 1902
1	What do I reward, and what outcome do I want?	Tails handed in; fewer live rats.
2	How could someone earn the reward without delivering that outcome?	Hand in a tail and release the rat, as Vann describes.
3	Could keeping the problem alive earn future payments?	Breeding rats could supply more tails. Vann reports rat farms.
4	What evidence could show that the rewarded count is misleading?	The reported tailless rat: a tail handed in need not mean a dead rat.

If line 2 has an easy answer, or line 3 is a yes, revise the rules before launch. Decide in advance how you will check the real outcome and what happens when payments stop. Ending a reward does not necessarily remove what people built to earn it.

Try this today (10 minutes)

Pick one bonus, commission, referral reward, or KPI that is tied to pay. Write the actual reward rule beside the outcome you want, then fill in the four lines.

Run one hypothetical case through the rule. For the support example, take a ticket that is closed, reopened, and closed again. Would it earn a second bonus while the customer’s problem stays unsolved? Check the written rule or the calculation instead of guessing.

Then pull one recent rewarded case and look for evidence of the outcome itself. Was the customer’s problem actually solved? Write down what you can establish, what remains unknown, and the rule or check you would change. A loophole on paper does not prove anyone has used it, and one clean case does not prove the system is sound.

For decisions that are not about pay, use the second-order thinking audit. The question to keep beside any reward is simple: could we pay more while the problem stays the same or gets worse?

Cite this:The cobra story is unverified. Hanoi's rat bounty is documented.Len P. van der Hof. https://lenvanderhof.com/en/blog/cobra-effect/ · Published 14 October 2026.