Nineteenth-century Delhi, under British rule, and there are too many venomous cobras in the streets.
The authorities have an obvious solution: offer a reward for every dead cobra residents bring in.
Soon enterprising locals began to breed cobras in order to make a living from the bounty.
The government worked this out and cancelled the programme. And then: the breeders, resentful of the rulers and angered by their actions, decided to release their cobras back on the streets, thereby tripling the population from before the government programme.1
Three moves. Each individually sensible. Ending with three times the original problem.
And Greene's assessment of what would have prevented it is deliberately deflating: a simple move partway up the mountain would have made clear the possible negative consequences so obvious to us in hindsight — for example, offering a reward for dead cobras would naturally cause impoverished residents to breed them.2
Partway. Not brilliance. One question, asked once, about what a poor person does with a standing offer to buy something they can produce.
Greene identifies the shape and it is the same in every case:
Alarmed by something in the present, we grab for a solution without thinking deeply about the context, the roots of the problem, the possible unintended consequences that might ensue.
Because we mostly react instead of think, our actions are based on insufficient information.3
And the specific missing information, per case: Caesar was not planning to start a monarchy; the poor people of Delhi despised their colonial rulers and would not take kindly to suddenly losing money; Americans would be willing to go to war if attacked.3
Note that none of those is obscure. Each was available and none was consulted, because the reaction was triggered by something alarming and the alarm supplies its own apparent sufficiency.
Then the cognitive form Greene names, which is the page's core:
Invariably in these cases people's thinking is remarkably simple and lazy: kill Caesar and the Republic returns, action A leads to result B.4
A leads to B. A single arrow, one hop long, terminating at the intended outcome — and stopping there because the intended outcome is where the thinking was aimed.
Greene's four historical examples are worth having because each fails at a different point in the chain.
Caesar, 44 BC. Men loyal to the Republic feared Caesar was making his dictatorship permanent, so they assassinated him thereby restoring the Republic. In the ensuing chaos and power vacuum, Octavius quickly rose to the top, assumed power, and permanently ended the Republic by establishing a de facto monarchy.
And the detail that makes it unbearable: after Caesar's death it came out that he had never intended to create a monarchical system.
The conspirators brought about precisely what they had tried to stop.5
Two failures at once — a wrong premise about the target's intentions, and no model of the vacuum the act would create.
The cobras, above: correct premise, no model of the incentive created.
Prohibition, 1920. The Eighteenth Amendment was designed to stop the spread of alcoholism and only ended up increasing alcohol consumption by a substantial amount.6
Pearl Harbor, 1941. Designed to decimate the U.S. naval force in one blow and bring America to its knees. Instead it shook the American public out of its deep isolationism, ensuring the total mobilization of the country's superior manpower and resources — to defeat Japan and obliterate its military for good.
And Greene's summary line, which is the most useful formulation on the page: the very success of the attack guaranteed the opposite of the intended result.6
The success guaranteed the reversal. That is a category worth naming separately, because it is the least intuitive: a plan that fails produces a bad outcome, and a plan that fails to consider its own success produces the reverse of its objective.
The same structure at ordinary scale, and Greene's examples are chosen so that the reader is in them.7
We try to control a rebellious teenager by putting some restrictions on his behaviour, only to make him even more rebellious and uncontrollable.
We try to cheer up a depressed person by making her realize that her life is not that bad and that the sun is shining, only to find out we have made her even more depressed. And the mechanism, which is exact: she now feels guilty about her feelings, worthless, and more alone in her unhappiness.
A wife tries to get her partner to open up more to her. With the hope of establishing more intimacy, she asks what he is thinking, what happened during the day. He interprets this as intrusiveness and closes up further, which makes the wife more suspicious and more prying, which closes him up even further.
That last one is the important one, because it is not a single reversal. It is a loop — each party's response is the other party's input, and the loop tightens with every iteration.
Which identifies the general form the historical cases share with the domestic ones. The intervention does not merely fail. It supplies the other party with a new reason to do more of the thing you were trying to stop — and the more energetically you intervene, the more reason you supply.
Greene makes one claim in this section that is genuinely counterintuitive and it is the most valuable thing here.
A variation on this, one that is quite common in the modern world, is to believe that if people have good intentions, good things should be the result. If a politician is honest and means well, he or she will bring about the desired results.
In fact, good intentions often lead to what are known as cobra effects, because people with the noblest intentions are often blinded by feelings of self-righteousness and do not consider the complex and often malevolent motivations of others.4
Good intentions as a risk factor, not as a mitigating one. The mechanism is specific and worth separating from the moral point.
Self-righteousness suppresses the and then what question, because asking it feels like doubting the goodness of the aim — and the aim's goodness is not in question, which is exactly why the question does not get asked.
And it produces a specific blindness about other people's motives. Someone acting from a good motive tends to model others as also acting from good motives, or as passive. The Delhi authorities did not consider that residents might respond to a bounty as an economic opportunity, or that cancelling it might produce resentment rather than compliance. Neither is a subtle inference. Both require modelling the other party as a self-interested agent, which is the thing self-righteousness quietly removes.
Which means the correlation runs the wrong way for comfort: the more confident you are in the rightness of an intervention, the less likely you are to have gamed out its consequences — and the sincerity of the actor is not evidence about the outcome. It is evidence about how little the outcome was examined.
Greene's diagnosis of the present is short and connects to the rest of the law:
Nonconsequential thinking is a veritable plague in the world today that is only growing worse with the speed and ease of access to information, which gives people the illusion that they are informed and have thought deeply about things.8
The illusion that they are informed. That is a real distinction and it is the same one this law makes elsewhere: information at ground level is not altitude, and rapid access to more of it produces confidence without producing perspective.
His examples: self-destructive wars such as the 2003 invasion of Iraq, the attempts to shut down the American government for short-term political gain, the increasing number of financial bubbles from tech stocks to real estate. And a further contributor: a gradual disconnect from history itself, as people tend to view present events as if they were isolated in time.8
The historical disconnect matters most for this page. Unintended consequences are only learnable from cases, because the mechanism is invisible in any single instance — you cannot see the counterfactual. A population that does not read history has no access to the one form of evidence this failure mode leaves behind.
Greene's prescription is unusually explicit about what is and is not achievable.
Any phenomenon in the world is by nature complex. The people you deal with are equally complex. Any action sets off a limitless chain of reactions. It is never so simple as A leads to B. B will lead to C, to D, and beyond.
Other actors will be pulled into the drama and it is hard to predict their motivations and responses. You cannot possibly map out these chains or get a complete handle on consequences.9
You cannot possibly map out these chains. That concession is what makes the rest credible.
But by making your thinking more consequential you can at least become aware of the more obvious negative consequences that could ensue, and this often spells the difference between success and disaster.
You want depth of thinking, to go to several degrees in imagining the permutations, as far as your mind can go.9
The more obvious ones. Which is the correct target: every case on this page was foreseeable at the first or second degree by someone who asked, and none required prophecy.
And two further instructions worth keeping.
Waiting is an output. Often, going through this process will convince you of the wisdom of doing nothing, of waiting. Greene's counterfactual: who knows what would have resulted in history if the conspirators had thought this out and chosen to wait until Caesar died naturally or in battle?10
And the organizational version, which is his best practical suggestion in the whole law:
In any group or team, put at least one person in charge of gaming out all of the possible consequences of a strategy or line of action, preferably someone with a skeptical and prudent frame of mind. You can never go too far in this process, and the time and money spent will be well rewarded.11
A named role, held by a specific person, with a temperament requirement attached. Which works where a shared intention fails, because a shared intention has no owner and evaporates under deadline.
You are about to intervene in something — a policy, a rule, a conversation, a fix.
Write the intended chain out to three degrees, not one. I do A, which produces B. Then: given B, what does each affected party do? Then: given that, what happens next?
Most plans do not survive degree two, and almost nobody performs it because degree one arrives with the plan attached.
At each degree, name the parties, individually. Not people will comply. Which people, with what incentives, and what does this arrangement now pay them to do? The cobra bounty fails immediately under that question and only under that question — what does a poor person do with a standing offer to buy a thing they can breed?2
Then ask the reversal question explicitly: what if this works? Pearl Harbor's designers modelled failure and did not model success.6 So: if this succeeds completely, what does the other side then have to do? A successful attack on someone who can retaliate produces a retaliating enemy, and that is a consequence of the success rather than of the failure.
Then check for the tightening loop. If your intervention increases pressure on someone, ask what their most likely response is — and whether that response is the thing you were trying to stop. The prying-and-withdrawing marriage is the model,7 and the signature is that the correct-seeming response to their reaction is more of what you just did. If your plan has that property, it will run indefinitely and get worse.
Then apply the intentions check to yourself, honestly. How certain are you that this is the right thing to do? If the answer is completely, downgrade your estimate of how well you have thought it through, because self-righteousness is the specific state in which the and then what question feels like disloyalty to a good aim.4
Then price waiting. What happens if I do nothing for three months?10 The option is real and the present-tense view never generates it, because the present-tense view is constituted by the alarm that prompted the intervention.
And if this is an organization, name the person.11 One individual, sceptical by temperament, whose stated job on this decision is to produce the failure modes — and who is not the person who proposed it.
The four historical cases are all real and all standard. Two need flags. The Delhi cobra story is the origin of the term "cobra effect" and is not well documented — it circulates widely, its primary source is thin, and it may be apocryphal. 🚩 SINGLE SOURCE; [CONTESTED]. The Prohibition claim that consumption increased substantially is contested: the best estimates suggest consumption fell sharply at first and then recovered to below pre-Prohibition levels, with the durable effects being on organized crime and enforcement rather than on total consumption. 🚩 CONSENSUS MISREPRESENTATION on the "increased by a substantial amount" formulation.
That is a real problem for the page, and worth stating plainly: the two cleanest illustrations of the mechanism are the two weakest historically. The mechanism is nonetheless well attested elsewhere; the vault should hold the principle and discount these particular exhibits.
The load-bearing tension is survivorship. History records the interventions that produced spectacular reversals and does not record the ones that worked. Greene's cases are selected on the outcome, which means they cannot establish anything about base rates — and without a base rate, "actions have unintended consequences" is compatible with intervening being usually right or usually wrong.
This matters because the page's implicit recommendation is caution and delay, and caution has unintended consequences too, which by construction are not in the sample. The Caesar counterfactual is the clearest case: Greene asks what would have happened if the conspirators had waited,10 and the honest answer is that nobody knows, and that a Caesar who died naturally at sixty-five might have produced a hereditary monarchy sooner.
Second, the intentions claim needs a boundary it does not have. Good intentions are a risk factor is a real mechanism and is also available, verbatim, as a rhetorical instrument for discrediting any reform. The mechanism concerns self-righteousness suppressing consequence-analysis — which is checkable by asking whether the analysis was done — and does not license the general inference that well-meant interventions fail. The check is procedural, not motivational.
Open question: Greene's remedy is to think several degrees deeper, and he concedes the chains cannot be mapped.9 But each additional degree multiplies the branches and the reliability of any single path collapses. At what depth does consequence-gaming stop adding accuracy and start generating plausible fictions — and is the honest answer that two degrees is the useful limit, with everything beyond it being an elaborate way of confirming what you already suspected?
The South Sea Bubble contains this law's own best example and Greene does not label it as one. The Bubble Act was a defensive measure that destroyed the thing it was designed to protect — by outlawing the rival ventures, it left thousands of people with only one asset they could sell, and the forced liquidation came straight into South Sea stock.12
That is a first-degree failure by an author who had every incentive to think it through, and it establishes something the cobra case does not: the person with the most at stake is not thereby better at this. Blunt's whole fortune depended on getting it right and he still stopped at degree one, because his time frame had shrunk to days.
The Mountain and Manufacturing Time holds the remedy, and this page supplies its most concrete test. Greene's claim that a simple move partway up the mountain2 would have caught the cobra bounty is falsifiable in a useful way: the required inference used no information the authorities lacked. Which means this failure mode is genuinely one of inattention rather than ignorance — and it is therefore the class of error the manufactured-hindsight method can actually address.
The relationship is condition and symptom, and stating it precisely gives the page a diagnostic it otherwise lacks.
Greene's account of unintended consequences is that we react instead of think and act on insufficient information.3 His account of the shrinking frame is that the horizon collapses under whatever is most immediately vivid, until a person can only consider the next few days.13
These are the same failure. A one-degree chain — A leads to B — is precisely what a short horizon can hold. Degrees C and D live outside the window, so they are not overlooked; they are not representable.
Which supplies the useful inversion: unintended-consequence failures are a measurement of horizon. You can read someone's time frame off the depth of the chains they model, without asking them anything. A plan whose reasoning terminates at the intended outcome was made by someone thinking in days, whatever they say about strategy.
And it explains why intelligence does not protect. Greene's examples include a Roman senatorial conspiracy, a colonial administration, an American constitutional amendment and a Japanese naval staff — all of them competent, none of them stupid. The failing faculty is temporal, not analytic, which is why smart institutions produce these reliably and why more expertise does not help.
That hub's subject is how systems degrade, and the cobra structure belongs in it as a named mechanism: the intervention that recruits its own opposition.
What this case adds to the hub's material is the three-move pattern, which is more specific than a single backfire. The bounty created breeders. The cancellation converted breeders into enemies. And the enemies had, in their possession, the exact thing the programme existed to remove.
That is a distinctive decline shape: an intervention creates a constituency, and withdrawing the intervention weaponizes it. The constituency did not exist before the policy — it was manufactured by the policy — and it ends up better armed against the original objective than anything that preceded it.
Which generalizes well past cobras, and it predicts a specific institutional trap: any programme that pays people to produce a measurable output creates a population dependent on that payment, and the population's response to termination is the risk that was never modelled. Subsidies, bounties, targets, and metric-linked funding all have this shape.
And it explains why such programmes are so hard to end — which is usually attributed to political capture and is at least partly this: the exit is more dangerous than the continuation, and everyone involved can see it, so the programme persists on its own failure.
Sharpest implication: the failure is not that consequences are unforeseeable. In every case on this page the decisive consequence was available at the first or second degree to anyone who asked what the other party would do next — and nobody asked, because the alarm that prompted the action also supplied the feeling of having thought enough. Which means this is not a limit on human foresight. It is a limit on human pausing.
Generative questions: