The Bayesian Absurdity of the Gazan Genocide Charge

They don’t hate Jews because they think it’s a genocide. They think it’s a genocide because they hate Jews.

I turned that into a meme after October 7. I wasn’t joking.

I have never treated the Gaza genocide charge as a close call. Hamas invades Israel, massacres civilians, kidnaps families, retreats into the dense civilian territory it rules, and Israel goes to war with Hamas. Almost immediately Israel is answering not charges of particular crimes, but the claim that the war itself is a project to destroy the Palestinian people.

Quite a trick. Hamas becomes scenery. Israel becomes the defendant.

I don’t accept that framing. For this essay I’ll humor it. Promote GENOCIDE from accusation to hypothesis and make it compete against WAR — not because they deserve equal standing, but to see how badly GENOCIDE does under ordinary inference.

WAR (W): After October 7, Israel goes to war against Hamas to destroy its military and governing capabilities, recover hostages, and prevent another massacre. Palestinians as a people are not the object.

GENOCIDE (G): After October 7, Israel uses the campaign to physically destroy Palestinians in Gaza, in whole or substantial part, as a people.

W can contain crimes, recklessness, cruelty, unlawful deprivation, intentional killings by individual soldiers, and some horrific rhetoric without becoming genocide. G is the bigger claim: physical destruction of the group is itself an objective of the campaign.

Bayes gives us two separate questions. First: which hypothesis better predicts the evidence that followed? That’s likelihood. Second: before that evidence arrived, which hypothesis was more plausible? That’s the prior.

So let’s do them in that order.

1. WEIGHING THE EVIDENCE

A fact is not evidence for G merely because G can produce it. It favors G only if it is more expected under G than under W.

The numbers below are toys — rough powers of ten for scale, not measured probabilities. If a fact looks an order of magnitude or more likely under W than G, I’ll just call it 10:1 for W. Hate one of my 10s? Fine. Replace it. The virtue of putting the arithmetic on the page is that you can see exactly what has to change.

And let’s not pick easy evidence. Let’s take six of the strongest facts commonly offered for G: tens of thousands of civilian deaths, apparently indiscriminate killing, destruction of homes and hospitals, harm to children, mass displacement, and deliberate deprivation.

Which hypothesis actually predicts each one better?

1. Tens of Thousands of Civilians Have Been Killed

Grant, for argument, 50,000 civilian deaths — fewer than 70 per day across roughly two years. That is not a settled civilian count; I’m deliberately granting G a large number.

And grant something more. Some Palestinian civilians may have been deliberately murdered by Israeli soldiers. Some Israeli officials have said horrific things. Individual intentional killings, if established, would obviously be crimes.

But individual murders are not genocide. W never predicted that every Israeli soldier would behave lawfully throughout a long and brutal war. Revenge killings, criminal soldiers, reckless shootings and intentional murders can occur during W without the purpose of W becoming the physical destruction of Palestinians.

G predicts something much larger: a systematic program. If Israel is killing civilians because they are Palestinians, Palestinian identity itself should function as a reason to kill. We should expect that purpose to show up systematically in orders, targeting rules, rules of engagement and operational behavior.

And then there is sheer rate. Israel could plainly have killed vastly more. Rwanda murdered on the order of 8,000 people per day for about 100 days. At Babyn Yar, the Nazis murdered 33,771 Jews in two days. Operation Harvest Festival killed roughly 42,000 Jews in two days. And in Iran in January 2026, two senior Health Ministry officials told TIME that as many as 30,000 people were killed on January 8 and 9 alone during the regime’s crackdown — a figure that remains unverified.

None of these establishes a minimum killing rate for genocide. There isn’t one. Nor does genocide legally require killing to be its mechanism; deliberately inflicting conditions of life calculated to bring about physical destruction can also qualify.

We’re comparing predictions. Under W, significant numbers of civilian deaths spread across a long urban war — including some possible murders — are grimly unsurprising. Under G, Israel supposedly intends to physically destroy Gaza’s Palestinian population, possesses vastly greater killing capacity than it uses, has the population largely trapped within a tiny territory, and yet kills civilians at a tiny fraction of the rate it plainly could.

If extermination by killing is the objective, why is Israel so extraordinarily bad at doing something well within its power?

10:1 for W.

2. Civilian Deaths Are Widespread and Seemingly Indiscriminate

The dead are not confined to some neat battlefield. Civilians have been killed across Gaza — in homes, streets, shelters and places to which they had fled. Up close, the deaths can look chaotic and indiscriminate. Zoom out, however, and the larger spatial pattern is anything but random: civilian danger rises and falls with the geography and intensity of military operations.

There are millions of Palestinians in the West Bank. Israel has not launched anything like the Gaza war against that population. The IDF conducts operations against armed groups and suspects there, sometimes substantial ones, but there is no general Israeli military campaign against the Palestinian population of the West Bank.

Then there is Israel itself. Roughly a fifth of its population is Arab, much of it Palestinian in identity and ancestry. There is obviously no military campaign against that population either.

And even inside Gaza, civilian danger varies enormously with the geography and intensity of the fighting. Some places have been devastated; others far less so. The danger moves as air campaigns and ground operations move.

That is much closer to what W predicts. Hamas is geographically localized. Battles move. Air campaigns move. Ground operations move. Civilian deaths follow them.

G has a stranger spatial prediction. If Palestinian identity itself is the target, why should killing track Hamas and the battlefield so much better than it tracks Palestinians themselves? Why Gaza rather than the West Bank? Why some parts of Gaza much more than others? Why not Palestinian citizens of Israel? Why should proximity to Hamas and the battlefield predict danger so much better than simply being Palestinian?

Under W, the answer is obvious: the killing is geographically widespread because the war is geographically widespread. Under G, that geography needs explaining.

10:1 for W.

3. Israel Destroys Homes, Hospitals and Other Civilian Infrastructure

Apartment buildings, schools, hospitals, mosques and entire neighborhoods have been damaged or destroyed. By October 2025, UN satellite analysis assessed approximately 81% of Gaza’s structures as damaged.

But Gaza is not a battlefield where Hamas sits on one side of town and civilians on the other. Hamas built an enormous tunnel network beneath the urban environment and has fought from within that environment.

That also makes sense of something otherwise peculiar: Israel warns civilians to leave places it intends to attack. Evacuation notices, neighborhood warnings, phone calls, texts, evacuation maps, corridors, pauses and other civilian-casualty-reduction measures are difficult to square with the proposition that those civilians are themselves the target.

Under W there is no contradiction. Israel can regard a building or neighborhood as militarily relevant while regarding the civilians inside it as people it wants out of the way.

Nor is the destruction spatially uniform. The same UN satellite assessment found the largest increases in newly recorded damage since July 2025 in Gaza and North Gaza.

None of that establishes that every hospital strike was justified, every destroyed apartment concealed Hamas, every demolition was lawful, or every Israeli targeting decision was sound.

G can explain the rubble. W better explains why the rubble follows the fighting — and, crucially, why Israel simultaneously uses warnings, evacuations, corridors, pauses and other civilian-casualty-reduction mechanisms. G can explain destroying civilian infrastructure. It has a much harder time explaining why a state supposedly trying to destroy the civilian population repeatedly takes measures designed to get civilians out of the way first.

10:1 for W.

4. Israel Has Put Hundreds of Thousands of Palestinian Children at Risk

Children have been killed, injured, displaced, deprived of food and exposed to disease. That is among the most emotionally powerful facts offered for G.

Then something rather strange happened.

A polio outbreak threatened Gaza’s children. Area-specific humanitarian pauses were agreed so they could be vaccinated. During the first round, 558,963 children received the vaccine, with nine-hour daily humanitarian pauses agreed for each phase. When the northern phase of the second round could not safely proceed amid intense bombardment, mass displacement and lack of assured humanitarian pauses, it was postponed. Once another pause was secured, the campaign resumed.

Under W, none of this is surprising. Palestinian children are being endangered by a war being fought around them, but they are not the object of the war. Pausing military operations to protect more than half a million of them from disease fits perfectly well.

Under G, come on. Israel supposedly intends to physically destroy this population while repeatedly interrupting military operations so more than half a million Palestinian children can be protected from disease.

International pressure? Appearances? Constraints? Sure. Those are possible explanations. The question is how much extra explaining G repeatedly needs before the simpler hypothesis starts looking rather obvious.

10:1 for W.

5. Israel Has Ethnically Cleansed Large Parts of Gaza

Strictly speaking, ethnic cleansing is not genocide. Removing a population and physically destroying it are different things. But mass forced displacement is certainly evidence one might reasonably put into the G column.

So grant the ugly version. Israel repeatedly ordered enormous numbers of Palestinians from their homes. Nearly the entire population of Gaza has at some point been displaced, many repeatedly, under dire conditions.

Fine. Keep watching.

Under the January 2025 ceasefire, Israel withdrew from the Netzarim corridor and hundreds of thousands of Palestinians returned to northern Gaza.

Under W, easy. Palestinians returning home does not frustrate the objective because Palestinians are not the object of the war.

Under G, considerably stranger. Israel had already accomplished the supposedly telling fact: massive removal of Palestinians from northern Gaza. Then it withdrew and allowed hundreds of thousands of those same Palestinians to go back. If physically destroying that population is the objective, why facilitate the reversal?

And it is awkward for the ethnic-cleansing interpretation itself. If the objective is permanent removal, letting the removed population return is a peculiar way to accomplish it.

Again G can survive: international pressure, negotiations, tactical concession, constraints. But those are additional pieces the explanation now has to carry.

10:1 for W.

6. Israel Deliberately Deprives Gazans of the Necessities of Life

Here is probably the hardest case for W, so let’s not sanitize it.

In 2025, more than 60 days passed with humanitarian aid and commercial supplies blocked from entering Gaza. By May, the IPC said the entire population faced high levels of acute food insecurity and about half a million people faced starvation. In August, the IPC confirmed famine in Gaza Governorate and reported more than half a million people across Gaza facing catastrophic conditions characterized by starvation, destitution and death.

If G is going to have a good piece of evidence, this is one of them.

The G case is not merely that war made Gazans hungry. It is that Israel deliberately restricted food, water, fuel, electricity and humanitarian assistance while destroying infrastructure necessary for civilian survival — precisely the sort of conditions-of-life mechanism through which genocide can in principle be committed.

So grant deliberate deprivation rather than explaining everything away as accidental wartime chaos. Now look at the entire time series rather than one interval. Aid enters, stops, resumes and expands. Access changes with military conditions and Israeli policy. Vaccination campaigns are facilitated. Ceasefires produce large increases in humanitarian activity. Food again enters. The mechanism supposedly being used to physically destroy the population is repeatedly relaxed or reversed.

Under W, deliberate restrictions can have military, siege or coercive purposes without physical destruction of Palestinians being the object. They might still be cruel, disproportionate, unlawful or criminal. W does not require Israeli policy to be good.

G requires something more specific: that these conditions are being deliberately imposed for the physical destruction of Palestinians. The severe deprivation therefore gives G something real to explain. But the reversals give G something real to explain too. If starvation is the mechanism of destruction, why repeatedly permit resupply? Why facilitate vaccination? Why does deprivation change so dramatically with fighting, negotiations and access rather than proceeding as a sustained program of destruction?

Pressure? International constraints? Deniability? Those are possible. But they are additional explanatory machinery. Even after granting G its strongest version of this evidence, the complete pattern is substantially more expected under W.

10:1 for W.

Six Arrows

Six pieces of evidence, deliberately chosen because they look like evidence for G. And yet, one by one, each looks more expected under W.

At my deliberately crude 10:1 apiece:

10⁶:1 for W.

They may not be perfectly independent. Some plainly overlap. But they are also substantially different facts: the number and character of civilian deaths, their geography, destruction of civilian infrastructure, treatment of children, displacement and return, deprivation and recovery.

There is, of course, a way to make G explain all six just as well as W — perhaps even better. Build a new G specifically to do it.

Call it G′. Israel is committing genocide — but it’s 5D-chess genocide, exquisitely designed so that everything that looks inconsistent with genocide is actually part of the ingenious genocidal plan. The strangely low killing rate? Deliberate. Kill slowly enough to fool everyone. Civilian deaths follow the battlefield rather than Palestinian identity? Brilliant. Make the genocide look exactly like a war against Hamas. Warn civilians before bombing? Of course. Save the people you’re trying to destroy so nobody realizes you’re trying to destroy them.

Pause fighting to vaccinate 558,000 Palestinian children? Genius. Protect them from polio as part of the plan to eliminate them. Drive Palestinians out, then let hundreds of thousands return? Obviously. Reverse the ethnic cleansing to conceal the ethnic cleansing. Starve them, then let food flood back in? Naturally. Interrupt the mechanism of destruction before it destroys too many people. No systematic extermination order ever appears? Ah, but that’s the cleverest part: the absence of the extermination program is evidence of how brilliantly the extermination program has been hidden.

The warnings. The vaccines. The returning civilians. The aid. The suspicious absence of an extermination program.

All part of my ingenious plan!!

Bwooohahahaha!

Once G works like this, it is almost impossible to falsify. Evidence for genocide supports it; evidence against genocide proves how cleverly disguised the genocide is. At that point we’re one corkboard away from full conspiracy theory.

Sure, any one of those auxiliary explanations could be true. International pressure exists. Military and diplomatic constraints exist. Governments compromise. The problem comes when the hypothesis is successively rigged to absorb every inconvenient observation after we have already seen it.

And if G′ has been constructed specifically to erase roughly six orders of magnitude of inconvenient likelihood evidence against ordinary G, all that extra machinery has to be paid for somewhere.

Which brings us to the other half of Bayes.

2. COMPARING THEIR PLAUSIBILITIES BEFORE THE EVIDENCE

Now rewind to October 7.

Erase everything that happened afterward. No Gaza casualty figures. No destroyed neighborhoods. No aid controversy. No polio campaign. No displacement and return.

Before any of that evidence existed, which hypothesis was more plausible?

1. Democracies Almost Never Do This

Since World War II there have been hundreds of major armed conflicts worldwide. One major conflict catalog counts 334 episodes from 1946 through 2019. Established democracies have participated in wars and major armed conflicts on the order of a hundred times, depending on exactly where one draws the boundaries around “democracy,” “war” and “participation.”

And yet I cannot identify a clean case in which an established democracy, while remaining an established democracy, used one of those wars to deliberately physically destroy an enemy civilian people as such.

That doesn’t make it impossible. Someone can dispute the reference class — colonial wars, occupations or other state-population conflicts may be the comparison they prefer. Fine. Reference classes are arguments, not commandments.

But for my deliberately crude bookkeeping, the historical rarity of an established democracy doing what G alleges is substantial prior evidence. I’ll call it 100:1 for W.

2. Israel Had Already Had Decades to Do It

Now narrow from democracies generally to Israel specifically. Israel had overwhelming military superiority over Palestinians for generations. It fought wars, uprisings and terrorist organizations, occupied Palestinian territories and repeatedly possessed the physical capacity to kill Palestinians on a vastly greater scale than it did.

Yet it did not physically destroy Palestinians as a people. That history does not prove Israel could never begin a genocide. But a hypothesis saying that after October 7 Israel suddenly acquired precisely that objective starts behind one saying it again went to war against an armed enemy.

Give W another deliberately modest 10:1, bringing my running prior to 10³:1 for W.

3. Israel Actually Left Gaza

Then there is 2005. Israel forcibly removed roughly 8,500 of its own settlers, dismantled its settlements in Gaza and withdrew its permanent military presence from inside the Strip. Israel continued to exercise important forms of control from outside Gaza, and Gaza’s legal status remained disputed.

But the direction of the act itself is hard to miss. If the underlying project was physical destruction of Gaza’s Palestinian population, physically removing Israel’s own settlements and permanent forces from Gaza is a peculiar chapter in the story.

Give W another modest 10:1, bringing my running prior to 10⁴:1 for W.

4. Millions of Palestinians Aren’t Being Targeted for Being Palestinian

There are millions of Palestinians in the West Bank, and there is no general Israeli war against that Palestinian population. Roughly a fifth of Israel’s own population is Arab, much of it Palestinian in identity and ancestry. They vote, elect representatives to the Knesset, and an Arab party even joined an Israeli governing coalition in 2021.

This is related to the geographic evidence considered earlier, but it is doing different work here. There we asked where deaths during this war occur; here we are asking what Israel’s conduct toward Palestinians generally gave us reason to expect before the war began.

Still, the two are cousins. If you think that makes them partly redundant, discount one. I’m not going to pretend Bayesian bookkeeping becomes exact merely because I put exponents on it.

None of that makes a Gaza-specific genocide logically impossible. But if the underlying claim is that Palestinians are being physically destroyed because they are Palestinians, the fact that millions of other Palestinians live under Israeli power or inside Israel without being subjected to such a campaign is awkward background evidence.

Give W another 10:1, bringing my toy prior to 10⁵:1 for W.

5. And Then Hamas Just Massacred Israelis

Finally, October 7 itself. Hamas attacks Israel, massacres people, takes hostages and retreats into Gaza.

What does W predict next? Israel goes to war with Hamas. No auxiliary story required. It is almost embarrassingly straightforward.

G requires something additional: the massacre somehow produces a new Israeli objective — not merely destroying Hamas, but physically destroying Palestinians in Gaza as a people. Could that happen? Sure. But it is the extra proposition.

Give W one final 10:1, bringing my toy prior to 10⁶:1 for W.

Put the Five Together

My deliberately crude prior now looks like this:

100 × 10 × 10 × 10 × 10 = 10⁶.

One million to one for W before seeing a single piece of the subsequent evidence.

I have not measured the true prior to six decimal places. I have made five explicit judgments and put numbers on them so you can attack the judgments rather than hiding the disagreement behind the word “plausible.”

Our six likelihood ratios from the earlier section then give another 10⁶:1 for W. Under these assumptions:

10⁶ × 10⁶ = 10¹².

One trillion to one for W.

That is the output of the toy model, not a scientifically measured probability. Think my numbers are ridiculous? Good. Change them. Make every one of the six likelihood ratios 2:1 instead of 10:1. Reduce the prior by three entire orders of magnitude. Discount the geographic overlap. Pick a less favorable reference class. Do the arithmetic again.

The exact answer will change enormously. The question does not disappear: after replacing my assumptions with yours, how much prior weight on G is required to overcome the evidence you yourself judge more probable under W?

And G′? That ridiculous 5D-chess genocide we mentioned earlier — somehow pre-equipped to explain the low killing rate, battlefield geography, civilian warnings, vaccinations, Palestinian returns and humanitarian reversals? Its prior has to pay for all of that cleverness. If the hypothesis needs a stack of auxiliary assumptions specifically tailored to erase six orders of magnitude of inconvenient likelihood evidence, those assumptions do not come free.

G′ solves nothing. It just moves the improbability from the likelihood column to the prior column.

3. HOW BIG WAS THEIR GENOCIDE BIAS?

Then I noticed the dates.

On October 15, 2023 — just eight days after the Hamas massacre — more than 800 scholars and practitioners of international law, conflict studies and genocide studies signed a public statement warning of the possibility of genocide in Gaza. The statement was careful about what it claimed: it sounded an alarm about the possibility of genocide; it was not a judicial finding that genocide had occurred.

Eight days.

And even that wasn’t the beginning. The statement was joining a chorus already underway. Within the first several days after October 7, journalists, activists, legal scholars and genocide scholars were already publicly using the word. By October 12, legal scholar Jeff Handmaker was writing of a “prelude to genocide.” On October 13, human-rights lawyer and Rutgers professor Noura Erakat was calling Israel’s response a “genocidal campaign.” That same day, Holocaust and genocide scholar Raz Segal went considerably further: “A Textbook Case of Genocide.”

And obviously an 800-plus-signatory statement doesn’t materialize instantaneously. It had to be conceived, written, circulated and signed. October 15 is the date on the statement, not the birthday of the idea. The statement itself even reached backward to discussions of genocide concerning Gaza from before October 7.

There had already been evidence by October 15. Eight days of warfare had passed. Israel had announced a siege. Israeli officials had made statements critics considered evidence of intent. Fine. I don’t need these people to have believed G was more probable than W, and I don’t need them to have concluded G. I don’t even need to know what probability they assigned G. The point is much simpler: G was already plausible enough to them to warrant public accusations and, within eight days, an extraordinary warning signed by more than 800 scholars and practitioners.

Now consider the comparison class. Established democracies have fought scores of wars and major armed conflicts in the postwar era. Those wars have killed civilians, destroyed cities, displaced populations, produced humanitarian disasters, generated allegations of war crimes and included ugly statements by political and military figures. Yet I cannot identify a comparable episode in which hundreds of genocide and international-law scholars mobilized within days to warn that an established democratic belligerent might be committing genocide. Maybe there is one. Show me.

Because the argument here is not that the October 15 signatories had already concluded G. They didn’t need to. The more basic question is why G had already cleared the bar for serious consideration at all. And practically all the evidence we have just spent six subsections examining cannot answer that question, because it hadn’t happened yet.

Nobody knew the eventual civilian death toll, what the geographic pattern of two years of fighting would look like, or the eventual scale and distribution of structural destruction. Nobody knew that Israel would later interrupt military operations for a polio campaign that vaccinated 558,963 children during agreed humanitarian pauses. Nobody knew about the later ceasefires, withdrawals, humanitarian reversals or mass Palestinian returns. The genocide accusation did not emerge at the end of that evidentiary trail. It was already sitting near the beginning.

The arrow of time is rather inconvenient that way.

Now run Bayes backward. Earlier, we took six later pieces of evidence commonly offered for G and, using my deliberately crude assignments, obtained a likelihood ratio of 10⁶:1 for W. How strongly would you have to favor G before that later evidence arrived merely to cancel that 10⁶:1 push toward W? Exactly the reverse: 10⁶:1 for G. Start at 10⁶:1 for G, let the subsequent evidence push 10⁶:1 toward W, and you finish at 1:1 — undecided.

Now compare that with the prior produced by my assumptions in Section 2:

Our starting point: 10⁶:1 for W

Required starting point just to get to 50:50 given the evidence: 10⁶:1 for G

Those starting points are twelve orders of magnitude apart. Put less politely, to get merely to 50:50, you have to inflate the reasonable prior odds of G by a factor of one trillion. A trillion-fold. And remember: that only gets you to undecided. To actually conclude G, you have to inflate the prior still further.

That trillion is not an independently measured fact about anybody’s mind. It is the output of the assumptions above. So hate the assumptions. Change them. Turn every 10:1 likelihood into 2:1. Gut my prior. Throw out an overlapping factor. Give G every reasonable break. Maybe you knock twelve orders of magnitude down to six, four, or two. Maybe, if you radically disagree with my likelihood judgments, you reverse the calculation altogether.

But then we have finally located the real argument. It isn’t that dead civilians exist, that buildings have been destroyed, or that people have been displaced. Both hypotheses can produce those facts. The argument is over which hypothesis predicted those facts better, and what we had reason to believe before they occurred.

And those first eight days tell us something different. They don’t tell us exactly how much probability any particular person assigned G. They tell us that G had already cleared the threshold for serious consideration almost immediately — indeed, for some, for outright accusation — before practically all the subsequent evidence now marshaled on its behalf existed.

Antisemitism? Anti-Zionism? Israel as uniquely evil? Oppressor-versus-oppressed ideology? Academic conformity? Signaling? Some mixture?

Whatever the source, the Bayesian point is hard to escape: the evidence didn’t create the extraordinary predisposition toward G. It was already there — a trillion times over.

Which brings me back to the meme:

They don’t hate Jews because they think it’s a genocide. They think it’s a genocide because they hate Jews.

The belief in genocide had to come first.