The War With No Exit

More than 230 of them crossed into northern Israel on 2 August 2006 — the largest single-day barrage of the conflict to that point. One landed in the northern West Bank, further south than anything before it. The next day brought over two hundred more and killed eight people, four of them in Akko.

Read as news, that is a bad day. Taken alone it proves very little. A record barrage is about equally consistent with an organisation burning off stock it expects to lose, or signalling ahead of a diplomatic endgame, as it is with an objective receding.

The falsification is not the spike. It is the line the spike sits on. Across thirty-four days and thousands of sorties, the rate of fire held above a hundred rockets a day and never meaningfully declined. Three weeks of the most intense air campaign Israel could mount produced no downward slope in the one number the campaign had publicly nominated as its own test. Day 22 is not the evidence. Day 22 is the illustration — the day the flat line got loud enough that nobody could file it.

One concession the campaign's defenders are owed, and the argument survives it: stopping the rockets was not the only objective. Israel's declared aims also included the return of two abducted soldiers, the Lebanese army deployed to the south, and Hezbollah pushed off the border. It eventually got versions of the last two, on paper. But rocket fire was the metric the government put in front of its own public — the thing the war was for, in the language used to sell it — and a campaign is entitled to be judged against the test it selects for itself. Everything that followed, and the war ran twelve more days, happened after that test had returned its result in plain numerals.

Nobody stopped. That is not a moral observation. It is the structural feature worth excavating, because it is the part that repeats.

i · the tool and the target

Air power is very good at destroying things that are large, valuable, expensive to replace, and located at a known address. Power plants. Bridges. Ports. Divisional headquarters. Every one of those attributes is a property of states, which is why the doctrine works reasonably well against states and has, for a century, been sold as though it works against everything.

Hezbollah's rocket force in 2006 had none of those attributes. The overwhelming majority of what it fired — roughly nineteen out of every twenty rounds across the war's thirty-four days — were Katyusha-class artillery rockets: cheap, short-range, stored in ones and twos in basements and orchards, launched off improvised rails by two men in ten minutes, in terrain where the launch site and the neighbourhood were the same coordinates. The total was somewhere between 3,970 and 4,228 rockets. There is no target set there. There is a practice, distributed across a population, with a resupply chain and no headquarters worth the name.

So the kill chain kept firing at the wrong ontology of target. When the actual targets are not findable, an air campaign does not stop; it retargets onto whatever is findable, which means infrastructure, roads, bridges, the Dahiyeh, apartment blocks — things that exist in a form the tool can grip. This is not a failure of intelligence or restraint. It is what a targeting cycle does when it is pointed at a network and asked to produce daily results. The bombs go somewhere. They go where the map has entries.

The commando raid on Baalbek shows the same mechanism at the other end of the risk scale. On the night of 1–2 August, Israeli special forces went in by helicopter roughly a hundred kilometres inside Lebanon, fought at the Hikma hospital that Israeli officials described as a Hezbollah command node, and came out with five prisoners. Operationally: audacious, technically superb, one of the most impressive small-unit actions of the war. Strategically: on 3 August, Hezbollah fired two hundred rockets. The raid produced captives and a headline. It did not bend the number.

This is the recursion, and it is older than the aircraft — but it has to be traced carefully, because the obvious precedents are the wrong ones.

Giulio Douhet promised in 1921 that command of the air would break civilian will and end wars quickly; the Combined Bomber Offensive delivered its heaviest tonnage in the year German war production peaked. That is a real failure, and it is a failure against a state — a country with factories at addresses and a government that could surrender. Reaching for it to explain 2006 is exactly the resemblance-argument this piece has just spent a section dismantling. Air power failed there, this looks like that, therefore. No.

The case that actually transfers is Rolling Thunder's interdiction leg. Not the tonnage — several hundred thousand tons across three and a half years is a statistic, and statistics are how the wrong lesson gets filed. The transferable part is what the interdiction was aimed at: infiltration south along a dispersed logistics network of trails, porters, bicycles, and truck parks with no fixed address, defended by little more sophisticated than dispersal and patience. Sorties went up. Infiltration went up with them. Same category of target, same result, thirty-eight years earlier.

Kosovo in 1999 is the case the doctrine's defenders cite, and it is instructive precisely because it is the exception that names the rule: Milošević ran a state, with assets, a treasury, and an address — and even then it took seventy-eight days. Air power coerces governments. It cannot find a network, because a network is not a place.

What the state cases and the network cases share is not the outcome. It is the accounting. In every one of them, the campaign's measurable output — sorties flown, tonnage delivered, structures destroyed, nodes serviced — was a property of the attacker's own activity, while the objective was a property of somebody else's behaviour. Those are different quantities. Only one of them can be counted daily, and it is the one that goes in the briefing.

ii · three clocks, none synchronised

Underneath the military mismatch sat a political one, and it was the political mismatch that made the ending inevitable in a specific shape.

Israel was operating on a diplomatic clock. It had roughly a month of American cover — Washington was openly resisting an immediate ceasefire in favour of a "durable" one, which is diplomatic vocabulary for let it run a while longer — and the entire theory of the campaign depended on converting that window into a degraded Hezbollah and a disarmed south.

Hezbollah was operating on a survival clock. It did not need to win anything. It needed to still be firing on the last day. An organisation whose victory condition is continued existence with capability intact cannot be defeated by attrition inside a thirty-four-day window, because attrition is measured against a denominator the attacker does not control and cannot see.

Two clocks, two victory conditions, and neither side's theory of success could be delivered by the means it had chosen. Israel needed Hezbollah to be the kind of thing air power breaks. Hezbollah needed Israel to be the kind of adversary that stops. Each was asking the other to be a different sort of object. When both parties in an escalation require the opponent to behave against their own structural incentives, the spiral does not resolve — it runs until an outside clock expires.

There was a third clock, and it belonged to everyone with no hand on either of the first two. A million Lebanese displaced. Three to five hundred thousand Israelis in shelters or moved south. A harvest season lost, a school year in question, a water system and a road network that would take years. That clock measured a quantity neither decision loop was optimising for: how long ordinary life can be suspended before the suspension becomes the permanent condition. It ran continuously and it was never consulted, because no mechanism existed by which it could be. Both loops were internally coherent. Neither of them contained anybody who paid.

iii · the day the diplomacy died

Three days before the record barrage, the campaign crossed the line where escalation began consuming its own political basis.

On 30 July, an Israeli airstrike on a building in Qana killed twenty-eight people, sixteen of them children, the youngest nine months old. Initial reports put the toll above fifty; the revision downward changed nothing that mattered. Lebanese Prime Minister Fouad Siniora cancelled his meeting with Condoleezza Rice — "There is no place on this sad morning for any discussion other than an immediate and unconditional cease-fire" — and Rice cancelled her trip to Beirut. The Security Council convened in emergency session and recorded "extreme shock and distress." Israel suspended air operations over southern Lebanon for up to forty-eight hours.

That is the moment the instrument turned. Up to Qana, the bombing was buying Israel time by demonstrating resolve to Washington. After Qana, the bombing was spending time, because each additional day generated the images that were closing the diplomatic window the campaign depended on. The tool was now working against the clock it was meant to serve.

The rational move at that point was to bank whatever had been achieved and take the ceasefire. Instead, the pause ended, the tempo resumed, and on 2 August the rocket count hit its high-water mark. This is the signature of the pattern: a campaign that has already failed its own test escalates rather than stops, because stopping requires an explicit admission that the objective was never reachable by these means, and no political leadership under fire has ever found that sentence easy to say.

iv · the last sixty hours

The purest expression came at the end. Resolution 1701 passed unanimously on 11 August. All parties had accepted it by the 13th. The fighting was scheduled to stop on the 14th.

Israel launched a major ground push into southern Lebanon in the window between the resolution and its effect — dozens of soldiers killed in an operation to improve the terms of a ceasefire that had already been agreed. The Winograd Commission, which Israel convened to examine its own conduct, later described the government as having been "dragged" into a ground operation too late, and found that "Israel went to war before it had decided which option to use." Its interim report in April 2007 described a prime minister without an organised plan and a chief of staff whose "over-charisma" suppressed the presentation of alternatives. Its final report in January 2008 called the whole thing a serious missed opportunity.

Note what that verdict concedes. Not that Israel lost. That it never specified what winning would consist of before committing to the means. You cannot build an exit from a war whose entrance had no destination attached.

v · who grades the exam

Every generation of planners rediscovers this, files it, and loses the file. That is the comfortable version, and it is wrong in the way that matters, because amnesia is nobody's fault and therefore nobody's job to fix.

The file does not get lost. It gets kept by the institution it indicts. Whoever owns a capability also owns the assessment of that capability: air forces write the after-action reports on air power, select the metrics, collect them, and interpret them. The organisation being asked to run the experiment is the organisation the experiment could defund. No conspiracy is required for that experiment to come back inconclusive; the incentive is sufficient, and it operates perfectly well on honest people. A closed epistemic loop does not manufacture lies. It manufactures a stable, professionally reviewed, meticulously footnoted body of findings that never quite arrives at the question.

Which is why the one durable verdict in this entire file had to be produced from outside the loop. Israel convened Winograd — an external commission, standing outside the chain it was examining — and it still took until January 2008, eighteen months after the shooting stopped, to get the finished sentence on paper. No air force has ever written that sentence about itself. The reason is not cowardice. It is that the report is graded by the office that sat the exam.

vi · what the record says

Thirty-four days. Somewhere between 1,191 and 1,300 Lebanese dead, over four thousand wounded. On the Israeli side 165 dead, including 44 civilians, and thousands wounded. Around a million Lebanese displaced, and between three hundred and five hundred thousand Israelis. Hezbollah's fighting strength reduced by a few hundred. Its rocket inventory rebuilt within a few years, larger.

Both sides declared victory, which is the reliable tell that neither achieved the objective they announced. What each actually achieved was the thing they had not stated: survival with legitimacy intact. Hezbollah's currency was having stood; Israel's was 1701 and an international force on its border. Those were real outcomes. They were also outcomes obtainable on 30 July, or 20 July, for a fraction of the price.

There is exactly one genuine break in the loop in this whole file, and it deserves marking because breaks are rare. On 27 August 2006, Hassan Nasrallah said in a television interview that had he known the scale of the Israeli response, he would not have ordered the 12 July cross-border raid. An actor in an escalation spiral publicly stating that he misread the reaction function — that is not nothing. It is the closest thing to an admission the historical record usually gets.

It changed no structure whatsoever. The template stayed on the shelf, fully documented, and has been taken down repeatedly in the two decades since: precision munitions applied to dispersed armed networks embedded in dense civilian terrain, an air campaign whose measurable output is rubble and whose stated objective is somebody else's decision, a political clock running faster than the military one, and a ceasefire that arrives when the diplomacy exhausts rather than when the war concludes.

The system was not broken on 2 August 2006. Every part performed to specification. The aircraft found their targets, the intelligence cycle produced coordinates, the commandos executed a flawless raid a hundred kilometres behind the line, and the rocket count went up anyway. A machine running exactly as designed toward an outcome the design cannot produce is not a malfunction. It is a category error with an air force.

Nobody involved was irrational. That is the uncomfortable part, and it is the whole finding. Two decision loops, each internally coherent, each following incentives that made sense from inside it, producing between them an outcome neither wanted and both could have had for free three weeks earlier. Local coherence is not a defence. A system can be coherent at every node and catastrophic in aggregate, and the difference is paid by whoever was on the third clock — the people the loops were run across rather than by.

The pattern is not inevitable. It is chosen, repeatedly, by people who have the file, own the file, and grade their own reading of it.