The Historical Band

The Shelter

On the morning of February 13, 1991, two F-117 Nighthawks dropped two two-thousand-pound laser-guided bombs through the roof of a reinforced concrete bunker in the Amiriyah district of western Baghdad. The first bomb cut through the upper level. The second exploded in the lower-level shelter where families had been sleeping. Approximately 408 Iraqi civilians, most of them women and children, died.¹

US intelligence had identified the bunker as a military command and control facility based on signals traffic, antenna placement, and reinforced construction. The Iraqi government had been using the lower level as a public air-raid shelter without informing the coalition or marking it externally as a civilian site. The two facts were both true. Neither was known to the other side at the time of the strike.

Amiriyah remains the single largest US-attributable civilian casualty event in any one strike since the Korean War. It is also the event that anchors the modern comparator class for evaluating stand-off precision air campaigns. Every air campaign that follows has been measured, implicitly or explicitly, against the question Amiriyah raised: when a country’s military attacks targets inside another country it does not occupy, using precision-guided munitions against an enemy embedded with civilians, what is the realistic floor for civilian deaths and what is the realistic ceiling?

This article walks through the answer the historical record gives. It is not a defense of any particular campaign. It is the comparator class against which the Iran 2025-2026 ratio (Hengaw’s Day-40 figure of 6.4 military deaths per civilian death) has to be evaluated.

The Comparator Class

Not every air campaign is a useful comparison for every other. The variable that matters most is target type. A campaign that targets dense urban districts where civilians live is going to produce a different civilian-share rate than a campaign that targets military bases in semi-rural settings. A campaign against a non-state actor with no air defense or industrial base is not comparable to a campaign against a national military with both.

The comparator class for the 2025-2026 Iran campaign is stand-off precision air war, conducted by an attacker that does not occupy the target country, against a target catalogue dominated by national air defenses, command and control nodes, missile-production facilities, and leadership compounds. The campaigns that fit this class are NATO Allied Force in Kosovo (1999), the Gulf War 1991 air phase, NATO Unified Protector in Libya (2011), the US 2003 Iraq invasion air phase, and the Coalition anti-ISIS air operations in Iraq and Syria (2014-2019).

The campaigns that do not fit this class but are sometimes invoked as comparisons are Israel-Hezbollah 2006, Israel-Hamas 2014, the 2023-2024 Gaza campaign, and the Russian campaigns in Syria and Ukraine. They do not fit because their target environment is fundamentally different. Hezbollah, Hamas, and ISIS in dense urban settings are non-state actors with no national air defense or industrial-base catalogue. The Russian campaigns deliberately targeted civilian infrastructure as a political objective rather than as collateral to military targeting. Russia is not in the comparator class because the comparison is to attackers that operate within precision-air doctrine, not those that reject it.²

The two camps produce different civilian-share rates not because one attacker is more careful than the other, but because the target environment determines the floor. A precision strike on an air-defense radar in open desert produces almost no civilian harm. A precision strike on a fighter sheltered next to an apartment building produces civilian harm by definition.

Iran 2025-2026 sits in the first camp. The Iran Prism analysis of the comparator class proceeds accordingly.

Lebanon 2006: Why It Is Not the Test

The Israel-Hezbollah war of July to August 2006 lasted thirty-four days. Israel conducted approximately twelve thousand sorties against Hezbollah positions in southern Lebanon and across the country. Human Rights Watch documented at least 1,109 Lebanese deaths, the vast majority of them civilians. Hezbollah claimed approximately 250 fighters killed; Israel claimed approximately 530 identified Hezbollah dead. The best-estimate ratio is roughly one military death per two civilian deaths, the worst civilian-share record of any modern stand-off air campaign.³

Lebanon 2006 is the example most often cited by critics of any subsequent precision-air campaign. The reason it is not a useful comparison for Iran 2025-2026 is that the target environment was fundamentally different. Hezbollah operated in dense Shi’ite villages of southern Lebanon, embedded with local civilian populations who in many cases supported the organization politically. The Alma Research Center later assessed that “every third house in Shi’ite villages of south Lebanon is used in some way by Hezbollah for military purposes.” The targeting environment was urban-embedded by definition; precision weapons cannot extract a fighter from a residential building without producing harm to the building’s other occupants.

Iran 2025-2026’s target catalogue is not Lebanon 2006’s. Roughly two-thirds of Iran 2026 strikes hit IRGC bases, missile-production complexes, air defenses, and naval assets in semi-rural or industrial settings: Khojir, Shahroud, Parchin, Hakimiyeh, Bandar Abbas, the Bushehr region. Lebanon 2006 has no analog for the Khojir missile complex strike. The target environments are not in the same comparator class; the civilian-share rates that result from targeting them are not comparable.

This is not a defense of Israel’s conduct in Lebanon 2006 (which was the subject of detailed HRW critique). It is an analytical point about which comparisons are forensic and which are rhetorical.

Kosovo 1999: The 78-Day Test

NATO Operation Allied Force lasted seventy-eight days and conducted approximately thirty-eight thousand sorties, of which approximately ten thousand four hundred were strike sorties. The targets included Yugoslav air defenses, command and control nodes, leadership compounds, and infrastructure including bridges, power generation, and the state broadcaster RTS. Human Rights Watch documented 489 to 528 civilian deaths across approximately ninety incidents. Yugoslav military deaths attributable to NATO action are estimated at approximately one thousand to fifteen hundred (the Cohen estimate is widely cited but contested).⁴

The resulting civilian-share rate for Kosovo 1999 is approximately thirty to fifty percent of strike-attributable deaths. The military-to-civilian ratio runs from approximately one to one to approximately three to one. This is the comparator class’s worst stand-off case in the late twentieth and early twenty-first centuries.

Kosovo’s civilian-share rate was driven by three factors. First, the campaign’s tempo was constrained by political consensus among NATO members, which required clearance procedures that often slowed targeting cycles to the point where weather windows closed before strikes could be executed. Second, the Yugoslav military dispersed effectively, mixing military and civilian assets in ways that produced a higher rate of erroneous targeting. Third, several high-profile civilian-harm events, including the April 14 Djakovica convoy strike (approximately seventy-three Albanian refugees killed by NATO cluster munitions in two strikes intended for Yugoslav military vehicles), the Grdelica train bridge (April 12, civilian passenger train killed), and the May 7 Chinese embassy bombing in Belgrade, generated sustained scrutiny.

The Iran 2026 ratio (approximately 4.8:1 to 6.4:1) is substantially more civilian-favorable than Kosovo 1999. This is not because the Iran campaign was conducted with more discipline than NATO showed in Kosovo (the Hegseth-era reduction of the Pentagon CHMR apparatus in 2025 makes that comparison unfavorable to the Iran campaign). It is because the target environment in 2026 was less urban, the target catalogue was less infrastructure-heavy, and the precision-weapon mix was more advanced. Same camp; different inputs.

Gulf War 1991: The Air-Phase Anchor

The forty-three-day air phase of the 1991 Gulf War conducted more than one hundred thousand sorties. Iraqi government claims of civilian deaths during the air phase ran to approximately three thousand. Human Rights Watch’s Needless Deaths report documented similar magnitude (approximately twenty-two hundred to thirty-five hundred). Iraqi air-phase military casualties are estimated by USAF at approximately ten thousand to twenty-five thousand (the wide range reflects the difficulty of distinguishing air-phase from ground-phase deaths). The resulting military-to-civilian ratio runs from approximately three to one to approximately eight to one.⁵

Gulf 1991 is the closest historical anchor for evaluating the Iran 2026 campaign’s civilian-share rate. The campaign duration is comparable (forty-three days versus Iran 2026’s approximately forty-day Phase 2 duration). The target catalogue is comparable (national air defenses, command and control, leadership compounds, infrastructure). The comparator-attacker is the same (US-led coalition operating with the precision-weapon doctrine of the era).

Iran 2026 (approximately 4.8:1 to 6.4:1) is at the favorable end of the Gulf 1991 band. The reasons match: precision-weapon technology has improved across thirty-five years, the target catalogue’s leadership-decapitation tilt was higher in Iran 2026 (more people who clearly meet combatant criteria), and Iran’s air defense network was less effective at producing collateral debris than Iraq’s was in 1991.

The Amiriyah shelter strike on February 13, 1991, is the defining single civilian-harm event of the Gulf 1991 air phase. It accounts for approximately twelve to eighteen percent of HRW-documented air-phase civilian deaths in a single event. Minab plays a structurally similar role for Iran 2026, accounting for six to seventeen percent of all Phase-2 civilian deaths depending on which casualty count is used. Both campaigns have an apex case that drives a disproportionate share of the campaign-wide civilian-share rate. Both campaigns also have a campaign-wide rate that sits inside the historical band even when the apex case is included.

Iraq 2003: The Invasion Phase

The forty-two-day air phase of the 2003 Iraq invasion produced approximately three thousand seven hundred fifty to seven thousand four hundred civilian deaths according to the Project on Defense Alternatives and Iraq Body Count cumulative tracking. Iraqi combatants killed by coalition forces during the invasion phase are estimated by PDA at approximately nine thousand two hundred. The resulting military-to-civilian ratio runs from approximately one point two to one to approximately two point five to one.⁶

The 2003 invasion’s lower military-to-civilian ratio relative to Gulf 1991 and Iran 2026 reflects the campaign’s different objectives. The 2003 air phase was a precursor to and integrated with a major ground-invasion campaign that aimed at regime overthrow rather than at degrading specific military capabilities. Targeting included urban environments where civilian density was high, including downtown Baghdad. The Shock and Awe doctrine emphasized speed and psychological effect over civilian-protection optimization.

Iran 2026’s civilian-share rate is substantially more favorable than the 2003 Iraq invasion air phase. The reasons are structural: Iran 2026 was not coupled to a ground invasion, so urban targeting was selective rather than comprehensive; the target catalogue was leadership and IRGC infrastructure rather than the Baathist regime at large; and the campaign tempo, while high, did not include a Shock and Awe equivalent doctrine that explicitly accepted higher civilian-harm rates as a feature.

Coalition anti-ISIS, 2014-2019: The Five-Year Reference

The Coalition air campaign against ISIS in Iraq and Syria conducted approximately thirty thousand strikes over roughly five years, releasing tens of thousands of munitions. Airwars cumulative tracking documented at least eight thousand to thirteen thousand civilian deaths attributable to coalition action (confirmed plus likely). Tens of thousands of ISIS combatants were killed across the campaign. The resulting military-to-civilian ratio is approximately three to one to five to one campaign-wide.⁷

The dense-urban operations within this campaign produced sharply higher civilian-share rates. Mosul (October 2016 to July 2017) produced approximately nine thousand to eleven thousand civilian deaths from all parties, with the coalition attributable share running into the low thousands. Raqqa (June to October 2017) produced at least sixteen hundred Airwars-and-Amnesty-documented civilian deaths from approximately twenty-one thousand coalition munitions over five months. Civilian share frequently exceeded twenty to thirty percent in dense-urban operations.

Iran 2026 is more favorable than the OIR campaign-wide aggregate (approximately 4.8:1 to 6.4:1 versus approximately 3:1 to 5:1) and dramatically more favorable than the OIR dense-urban operations like Mosul and Raqqa. The reasons match again: Iran 2026 had no Mosul or Raqqa equivalent because the IRGC concentrated assets in identifiable bases, missile-production complexes, and command compounds rather than dispersing into urban occupation patterns the way ISIS did. The OIR comparison is useful not as a target-similarity anchor but as a precision-air capability anchor: the same kind of weapons, the same kind of attacker discipline, the same kind of independent monitoring infrastructure.

NATO Libya 2011: The Modern Best Case

NATO Operation Unified Protector lasted two hundred twenty-two days and conducted approximately ninety-seven hundred strike sorties releasing approximately seventy-seven hundred precision-guided munitions. The UN Commission of Inquiry documented sixty civilian deaths in twenty investigated strikes. Human Rights Watch retrospectively confirmed seventy-two civilian deaths. The Airwars 2021 retrospective put the range at two hundred twenty-three to four hundred three likely civilian deaths. Estimated military deaths attributable to NATO action are in the low thousands. The resulting military-to-civilian ratio runs from approximately five to one to approximately ten to one. Libya 2011 is the modern best case for stand-off precision air war.⁸

The Libya 2011 best case rests on three structural advantages that Iran 2026 partially shared and partially did not. First, Libya’s geography concentrated Qaddafi-loyalist military assets along a narrow Mediterranean coast and in a small number of identifiable garrisons, allowing NATO to target without dense urban exposure. Second, the campaign benefited from extensive coordination with anti-Qaddafi rebel forces, which provided real-time targeting intelligence and reduced the need for area-effect munitions. Third, the campaign’s political constraints (UNSCR 1973) explicitly limited strike permissions to civilian-protection objectives rather than open regime-change targeting, which constrained the target catalogue.

Iran 2026 shared the precision-weapon mix and the standoff-attacker posture but had less geographic concentration, no equivalent ground-coordinator force providing real-time targeting intelligence, and a much denser urban target catalogue centered on Tehran (population approximately nine million). The difference between Iran 2026’s approximately 4.8:1 to 6.4:1 and Libya 2011’s approximately 5:1 to 10:1 is therefore structurally explainable rather than evidence of indiscriminacy.

The most-cited Libya 2011 incident, the August 8 strike on Majer (thirty-four civilian deaths in HRW’s documentation), is structurally analogous to Minab. Both are the campaign’s apex single-event civilian-harm case. Both account for a disproportionate share of campaign-wide civilian deaths. Both campaigns retain a campaign-wide ratio inside the favorable end of the historical band even with the apex case included.

Where Iran 2026 Lands

The historical band for stand-off precision air campaigns against military-saturated target catalogues, summarized:

Campaign	Duration	Civilian deaths (independent)	Military:civilian ratio
Israel-Hezbollah, Lebanon (2006)	34 days	1,000+ (HRW)	~1:2
NATO Allied Force, Kosovo (1999)	78 days	489-528 (HRW)	1:1 to 3:1
US-led invasion, Iraq (2003 air phase)	~42 days	3,750-7,400 (PDA, IBC)	1.2:1 to 2.5:1
Coalition anti-ISIS, Iraq + Syria (2014-19)	~5 years	8,000-13,000 (Airwars)	~3:1 to 5:1
US-led, Gulf War (1991 air phase)	43 days	2,278-3,500 (HRW, IBC)	3:1 to 8:1
Iran 2025-2026 (Hengaw-anchored)	52 days	~1,466 named	~4.8:1 to 6.4:1
NATO Unified Protector, Libya (2011)	222 days	72-403 (HRW, Airwars)	5:1 to 10:1

Iran 2026 sits at par with Gulf 1991, modestly less favorable than Libya 2011, and substantially more favorable than every other modern comparator. The civilian-share rate of approximately 13.5 percent (Hengaw Day-40) sits inside the comparator class median for stand-off precision campaigns of this duration and target type. It is the favorable end of the modern band, not an outlier. The 4.8:1 to 6.4:1 range in the table reflects two scopes: 4.8:1 covers both phases combined (with Phase 1’s larger unclassified pool defaulting toward civilian classification under the IHL doubt rule); 6.4:1 covers Phase 2 alone, where named-victim documentation is densest. The Phase 2 figure is the cleaner methodological anchor; the combined figure is the conservative floor.⁹

The reading the band supports is not that Iran 2026 was a model campaign. It is that the campaign-wide ratio it produced is what the comparator class predicts when a precision-air attacker engages a military-saturated target catalogue in a country with a substantial urban target component (Tehran) and a defender that systematically co-locates military assets with civilian infrastructure.

The factors that pushed Iran 2026 to the favorable end of the band (precision-weapon technology, leadership-decapitation tilt, IRGC concentration in identifiable bases) and the factors that prevented it from reaching the Libya 2011 best case (CHMR gutting in 2025, Tehran-embedded leadership catalogue, defender co-location) both operated within the predicted band. Neither produced an anomalous outcome.

The Caveat Worth Holding

Three caveats are load-bearing.

First, none of these comparator campaigns is a perfect match for Iran 2026. The closest match is Gulf 1991 by duration and target catalogue; the closest match is Libya 2011 by stand-off-attacker posture; the closest match is OIR by precision-air capability. The historical band is constructed from imperfect comparators and should be read as a band, not as a single benchmark.

Second, the comparator class is constructed through independent named-victim methodologies (HRW, Airwars, IBC, the UN Commission of Inquiry on Libya). Each of these methodologies has its own conservative biases. The historical-band figures are likely floors, not ceilings, in much the same way Hengaw’s Iran 2026 figures are floors, not ceilings.

Third, the historical band tells us where Iran 2026 sits in comparison to other campaigns. It does not tell us whether Iran 2026 was justified, prudent, or strategically sound. Those are policy questions outside the comparator-class analysis. The forensic case is that the campaign-wide ratio is not anomalous against the historical record of stand-off precision air war. The judgment of whether a non-anomalous campaign should have been conducted at all is a separate question that the band does not answer.

This article is a deep dive sub-article of Six to One. For the legal framework on defender co-location and the expected-ratio counterfactual, see Article 58 and the Co-Location Tax. For the methodology behind the named-victim verification, see The Verification Floor. For the apex case, see The Minab Test.

Amiriyah shelter strike, February 13, 1991: approximately 408 Iraqi civilian deaths in single incident. F-117 strike with two 2,000-pound laser-guided bombs penetrated the upper level and detonated in the lower-level shelter. US intelligence had identified the bunker as a military command and control facility based on signals traffic, antenna placement, and reinforced construction. Iraqi government had been using the lower level as a public shelter without coalition notification or external civilian marking. Sources: HRW Needless Deaths report; US Department of Defense after-action review. ↩
Comparator class definition. Stand-off precision air war against a country the attacker does not occupy, target catalogue dominated by leadership, command and control, missile production, and air defense infrastructure. Excluded as comparators (treated as upper-bound references only): Israel-Hezbollah 2006 (dense urban embedded, non-state actor with no national air defense or industrial base catalogue); Israel-Hamas 2014 and 2023-2024 Gaza (same reason); Russia-Syria and Russia-Ukraine (deliberate civilian-infrastructure targeting as political objective rather than collateral to military targeting). The classification follows the framework used in HRW campaign reviews, Airwars comparative analyses, and the ICRC humanitarian law commentary on precision-air operations. ↩
Israel-Hezbollah 2006: 34 days, approximately 12,000 sorties. HRW documented at least 1,109 Lebanese deaths, “vast majority civilian.” Hezbollah claimed 250 fighters killed; Israel claimed 530 identified Hezbollah dead. Best-estimate ratio approximately 1 military per 2 civilian deaths. Alma Research Center: “every third house in Shi’ite villages of south Lebanon is used in some way by Hezbollah for military purposes.” Source: HRW Why They Died report (2007); Alma Research Center 2024 assessment. ↩
NATO Operation Allied Force, Kosovo 1999: 78 days, approximately 38,000 sorties, approximately 10,400 strike sorties. HRW Civilian Deaths in the NATO Air Campaign report: 489-528 civilian deaths in approximately 90 incidents. Yugoslav military deaths approximately 1,000-1,500 (Cohen estimate; widely cited but contested). Civilian share 30-50 percent of strike-attributable deaths; military-to-civilian ratio approximately 1:1 to 3:1. Defining incidents: April 14 Djakovica refugee convoy (approximately 73 Albanian refugees killed by NATO cluster munitions in two strikes intended for Yugoslav military vehicles); April 12 Grdelica train bridge (civilian passenger train); May 7 Chinese embassy bombing in Belgrade. Source: HRW Civilian Deaths in the NATO Air Campaign (2000); RAND Project Air Force Kosovo analyses. ↩
Gulf War 1991 air phase: 43 days, more than 100,000 sorties. Iraqi government claimed approximately 3,000 civilian deaths during air phase. HRW Needless Deaths documented approximately 2,278-3,500 civilian deaths. Iraqi air-phase military deaths estimated by USAF at approximately 10,000-25,000. Military-to-civilian ratio approximately 3:1 to 8:1. Sources: HRW Needless Deaths in the Gulf War (1991); Iraq Body Count cumulative tracking; USAF Gulf War Air Power Survey. ↩
US 2003 Iraq invasion air phase: approximately 42 days. Project on Defense Alternatives estimated approximately 3,750 civilian deaths attributable to air phase. Iraq Body Count cumulative tracking range approximately 7,419 (all causes). Iraqi combatant deaths from coalition forces in invasion phase approximately 9,200 (PDA). Military-to-civilian ratio approximately 1.2:1 to 2.5:1. The campaign’s lower civilian-favorable ratio relative to Gulf 1991 reflects coupling to ground invasion, urban targeting in Baghdad, and Shock and Awe doctrinal acceptance of higher civilian-harm rates. Sources: PDA The Wages of War (2003); Iraq Body Count. ↩
Coalition anti-ISIS / OIR, 2014-2019: approximately 30,000 strikes over roughly 5 years. Airwars cumulative all-time minimum range 8,000-13,000 civilians killed (confirmed plus likely). Tens of thousands of ISIS combatants killed. Mosul (October 2016 to July 2017): approximately 9,000-11,000 civilian deaths from all parties. Raqqa (June to October 2017): at least 1,600 Airwars-and-Amnesty-documented civilian deaths from approximately 21,000 coalition munitions over 5 months. Civilian share frequently exceeded 20-30 percent in dense-urban operations. Campaign-wide military-to-civilian ratio approximately 3:1 to 5:1. Source: Airwars cumulative tracking; Amnesty International War in Raqqa (2018); HRW reporting on Mosul. ↩
NATO Operation Unified Protector, Libya 2011: 222 days, approximately 9,700 strike sorties releasing approximately 7,700 precision-guided munitions. UN Commission of Inquiry documented 60 civilian deaths in 20 investigated strikes. HRW retrospective confirmed 72 civilian deaths. Airwars 2021 retrospective range 223-403 likely civilian deaths. Estimated military deaths from NATO action low thousands. Military-to-civilian ratio approximately 5:1 to 10:1. Most-cited single incident: August 8 Majer strike, 34 civilian deaths (HRW). Structural advantages: geographic concentration, ground-coordinator force providing real-time targeting intelligence, UNSCR 1973 explicit civilian-protection mandate constraining target catalogue. Sources: UN Human Rights Council Commission of Inquiry on Libya report; HRW Unacknowledged Deaths report; Airwars Libya 2021 retrospective. ↩
Iran 2025-2026 anchor: Hengaw Day-40 aggregate of 7,650 deaths, 1,030 civilians, 13.5 percent civilian share, approximately 6.4:1 military-to-civilian ratio. Phase 1 plus Phase 2 combined ratio approximately 4.8:1; with imputation of unclassified Phase 1 deaths approximately 5.5:1 to 6.4:1. Campaign duration 52 days combined (Phase 1: 12 days June 2025; Phase 2: 40 days February 28 to April 8, 2026). Source: Hengaw Organization for Human Rights serial reports; HRANA cumulative tracking through April 7, 2026; Airwars-CIVIC-HRA joint report March 27, 2026. Full forensic: Six to One. ↩