The Tunnel
Iran International’s open-source investigation team, working with satellite imagery, leaked Iranian engineering documents, and ground-level photography from Tehran residents, traced an underground tunnel network that ran beneath four civilian sites in central Tehran: the Shahid Shourideh medical center, the Hejrat Girls’ Elementary School, the Tohid Mosque, and the Karimeh Ahl-e Beit dental and medical clinic. The tunnels connected to Ali Khamenei’s residential compound and to associated Office of the Supreme Leader facilities. The construction documentation predates the 2025-2026 war by years.1
A mosque. A girls’ elementary school. Two medical clinics. The places a state is supposed to set apart, used as cover for the state’s own concealment. Each of the four sites was, by Iranian and international law, a protected object. The presence of Khamenei-compound infrastructure beneath them did not, under any reasonable reading of international humanitarian law, transform them into legitimate military targets. It transformed them into civilian objects whose protection had been compromised by a defender’s decision to use them as physical cover for military assets.
The four tunnels are not a marginal case. They are the apex case of a documented pattern. The pattern is what this article is about. The ratio that pattern produced (the approximately 1,030 Iranian civilian deaths attributable to coalition strikes during Operation Epic Fury) is the subject of the rest of the cluster. The legal framework that governs both sides of the strike-civilian-harm relationship is what this article walks through.
Iran ratified Additional Protocol I to the Geneva Conventions in 1977. Article 58 of that protocol places affirmative obligations on the defending party. The IRGC’s documented conduct is what those obligations look like when they are systematically refused.
The Treaty Iran Ratified
Additional Protocol I, adopted in 1977 and ratified by Iran, contains the modern legal framework for protecting civilians during international armed conflicts. Three articles do most of the work, and they are addressed to different parties.2
Article 51 paragraph 7 prohibits a party from using “the presence or movements of the civilian population or individual civilians” to render military objectives immune from attack. This is addressed to the defender. It is the core anti-shielding provision.
Article 51 paragraph 8 specifies that any violation of the anti-shielding rule “shall not release the Parties to the conflict from their legal obligations with respect to the civilian population, including the obligation to take the precautionary measures provided for in Article 57.” This is addressed to the attacker. It is the rule that prevents a defender’s misconduct from licensing an attacker’s misconduct.
Article 58 imposes affirmative duties on the defender. To the maximum extent feasible, the defender must endeavor to remove the civilian population, individual civilians, and civilian objects under its control from the vicinity of military objectives, avoid locating military objectives within or near densely populated areas, and take other necessary precautions to protect the civilian population, individual civilians, and civilian objects under its control against the dangers resulting from military operations.
The structure of these three provisions is deliberate. Article 51(7) prohibits the worst defender conduct. Article 58 requires the affirmative defender conduct that prevents the bad defender conduct from arising in the first place. Article 51(8) ensures that the prohibition on attacker misconduct cannot be eroded by defender misconduct.
For the IRGC, the affirmative Article 58 obligation runs directly against operational doctrine. The Foundation for Defense of Democracies’ 2024 assessment of IRGC force structure described the doctrine as centered on “the concealment of strategic assets in places the enemy cannot target.” Iran International’s open-source investigation describes the same pattern from the empirical end. The Khamenei tunnel network beneath civilian infrastructure is what that doctrine produces in the field. Article 58 is what international humanitarian law obligates instead.3
What the Pattern Looks Like
The documented record of IRGC co-location of military assets with civilian infrastructure during the 2025-2026 conflict period covers at least twelve incidents across nine cities, plus the Tehran tunnel network. The fuller forensic catalog lives in Twelve Incidents, Nine Cities. A summary of the most heavily documented cases, drawn from Hengaw, Iran International, FDD, AP, Mnemonic, and the US State Department:4
Sardasht (West Azerbaijan): IRGC forces occupied the Kosar Girls’ Elementary School during the June 2025 air campaign.
Kermanshah: at least seven IRGC weapons sites identified in the Dizelabad and Dolatabad districts adjacent to Farabi Hospital.
Ilam: Imam Khomeini Hospital subjected to a 24-hour-plus IRGC siege January 4, 2026, with forces firing into the courtyard, deploying tear gas in wards, beating medical staff. The Rababeh Kamali School across from the hospital was commandeered as a staging area.
Tehran: Sina Hospital breached by security forces; tear gas deployed inside; January 8, 2026 saw 217 deaths across the Tehran hospital network where IRGC had evacuated facilities to treat its own forces.
Tehran: Khatam al-Anbiya Hospital used by IRGC commanders for non-medical meetings, accompanied by security details (Iran International February 2026 documentation).
Tehran: Mofatteh 2 School (District 15) used as a Basij/IRGC staging point ahead of Phase 2 operations.
Tehran: the Hamdaman rehabilitation institute (an orphan-girls-with-disabilities facility) partially commandeered by IRGC as documented by FDD March 26, 2026.
Arak: Imam Ali high school used as a Basij/IRGC staging point.
Minab: the Shajareh Tayyebeh elementary school stood approximately sixty meters from the IRGC Naval Forces compound; the Shahid Absalan IRGC Navy medical clinic stood approximately 240 meters from the school; commercial satellite imagery showed people in the school courtyard in December 2025, two months before the strike.
The Tehran tunnel network: Khamenei-compound tunnels documented running beneath the Shahid Shourideh medical center, the Hejrat Girls’ Elementary School, the Tohid Mosque, and the Karimeh Ahl-e Beit dental and medical clinic.
The pattern’s load-bearing feature is not the dozen identified incidents. It is the systemic Article 58 violation those incidents indicate. Hengaw’s standing assessment, repeated across multiple serial reports, characterizes the practice as “Iranian military forces have relocated into civilian spaces, including schools, student dormitories, and mosques in residential areas,” which “constitute the use of human shields under the Geneva Conventions.” The legal framework on which this assessment rests is straightforward; the framework’s application to the documented conduct is straightforward; the empirical record supporting both is robust.
What the Catalogue Predicts
The expected-ratio counterfactual for Operation Epic Fury rests on Collateral Damage Estimation methodology. The methodology is governed by Chairman of the Joint Chiefs of Staff Instruction 3160.01 and operates in five escalating levels, each progressively more detailed when civilian risk cannot be resolved at a lower tier. The methodology is the subject of How Targeting Actually Works; the application to the Operation Epic Fury target catalogue is the subject of this section.5
The campaign’s documented target mix, reconstructed from CENTCOM, IDF, ACLED, Hudson Institute, CSIS, and TWZ campaign reporting through April 2026, breaks down as follows.
Approximately sixty to seventy percent of strikes hit IRGC bases, missile-production complexes, air defenses, and naval assets in semi-rural or industrial settings. Khojir, Shahroud, Parchin, and Hakimiyeh are Iran’s four major missile-production complexes; the IRGC Naval Forces are concentrated at Bandar Abbas; the Bushehr region houses additional naval and missile infrastructure. For this category of target, historical civilian-incidence rates run approximately three to six percent under CDE Level 4 precautions.
Approximately twenty to twenty-five percent of strikes hit Tehran-embedded leadership, intelligence, and command nodes. The Pasteur compound (Office of the Supreme Leader), Malek-Ashtar university (associated with the Aerospace Force Research Institute), the IRIB headquarters, the Resalat-neighborhood Basij-affiliated building, IRGC headquarters in Saadat Abad. For this category, historical civilian-incidence rates run approximately ten to eighteen percent under standard CDE Level 4 precautions.
Approximately five to ten percent of strikes hit dual-use or contested targets. The Karaj B1 bridge (struck on Sizdah Be-dar, the thirteenth day of the Iranian new year, when Iranian families traditionally picnic outdoors); Iran University of Science and Technology; the Minab compound. For this category, historical civilian-incidence rates run twenty to forty percent even with maximum precaution because civilians are present by definition.
Weighted-average expected civilian share for the documented catalogue: approximately nine to thirteen percent under CDE Level 4 precautions.
Hengaw’s observed thirteen point five percent civilian share sits at the top of the predicted band. Not above it. The catalogue’s composition is what predicts a civilian share at approximately this level. Two factors plausibly pushed the observed rate to the top of the predicted band rather than the floor: the documented gutting of the Pentagon’s Civilian Harm Mitigation and Response apparatus by approximately ninety percent in 2025 under Defense Secretary Hegseth, and the documented IRGC pattern of Article 58 violation cataloged above. Both factors operate within the predicted band. Neither produces an anomalous campaign-wide outcome. The campaign sits where the catalogue predicts it would sit.
The Reverse Counterfactual
A second test runs the calculation backward. If the IRGC had garrisoned its bases away from civilian neighborhoods and had not operated underground tunnel complexes beneath Tehran schools, hospitals, and mosques, what civilian-share rate would the same campaign have produced?
The closest historical analog is NATO Operation Unified Protector in Libya 2011. The Libya campaign benefited from a defender (the Qaddafi regime) that concentrated military assets along an open Mediterranean coast and in identifiable garrisons rather than systematically embedding them in urban civilian infrastructure. The civilian-share rate under those conditions ran approximately four to eight percent of strike-attributable deaths, per the HRW retrospective and the Airwars 2021 retrospective.6
Applying the Libya 2011 norm to Operation Epic Fury’s approximately five thousand to eleven thousand documented strikes against a comparable munition mix would predict approximately two hundred to four hundred Iranian civilian deaths.
Hengaw documented approximately one thousand thirty.
The gap (six hundred to eight hundred Iranian civilian deaths above the no-co-location counterfactual) is the empirical measurement of what the Article 58 violation cataloged in the previous section produced in the field. Each of those deaths is, under standard IHL analysis, the product of a defender decision to embed military assets in civilian infrastructure rather than to separate them as Article 58 obligates.
Imagine being told the school across the street from your house had been selected, without your knowledge or consent, as physical cover for a military command node by your own government. Imagine learning this only after the bombs fell. The tunnel under the Hejrat Girls’ Elementary School in Tehran was not built without engineering plans. The plans existed. The decision to route them under the school was a decision. Article 58 obligates Iran’s defense apparatus to make a different decision, and it did not.
Two clarifications are essential.
First, the gap is a counterfactual estimate, not a precise measurement. The Libya 2011 norm itself rests on conservative HRW and Airwars accounting; it could be too low if Libya 2011 itself under-counted. The gap could therefore be smaller. But the directional finding (a substantial portion of Iran 2026 civilian deaths fall above what the no-co-location counterfactual predicts) is robust to plausible variations in the Libya benchmark.
Second, the gap is a forensic attribution, not a moral one. Under standard IHL analysis (the Schmitt-ICRC consensus), the attacker bears full proportionality liability for civilian deaths even when the defender violates Article 58. Article 51(8) is explicit: defender misconduct does not relieve the attacker of obligations under Article 57 (precaution) or Article 51(5)(b) (proportionality). The reverse counterfactual measures what the defender’s conduct caused in the field; it does not relieve the attacker’s legal duty. The legal and political framings on this point diverge sharply, and the next section walks through that divergence.
The Scholarly Debate
International humanitarian law scholars disagree about whether the proportionality calculation under Article 51(5)(b) should be adjusted when human shields are present. Three camps exist; the disagreement is sharp and substantive.7
Yoram Dinstein argues that “the actual test of excessive injury to civilians must be relaxed” when shields are present. The Conduct of Hostilities Under the Law of International Armed Conflict, fourth edition, treats human shields as carrying “less weight than other civilians in the proportionality calculation.” The U.S. Department of Defense Law of War Manual and the U.K. Joint Service Manual both treat the defender’s responsibility as a relevant factor in proportionality assessment. This is the relaxation camp.
Beth Van Schaack of Stanford, formerly U.S. Ambassador-at-Large for Global Criminal Justice, rejects relaxation. Efforts to “relax” proportionality should be “rejected,” in her formulation, because they create incentives for defenders to use shields and because they shift accountability from the more legally accessible party (the attacker) to a party that operates outside the rule of law (the shielder). The ICRC’s commentary maintains the same position: all shields retain full civilian status and full weight in the proportionality calculation. Amnesty International and Human Rights Watch hold the same position. This is the no-relaxation camp.
Michael Schmitt of the Lieber Institute at West Point, lead author of the Tallinn Manual on the International Law Applicable to Cyber Warfare and one of the most cited scholars in the field, occupies the middle ground. Schmitt distinguishes voluntary from involuntary human shields. Voluntary shields, in his analysis, may constitute direct participation in hostilities and may therefore be excluded from proportionality calculations. Involuntary shields retain full civilian status; their proportionality weight is not relaxed. Schmitt’s principal objections to the relaxation argument are practical (it raises the unanswerable question of how much relaxation would be appropriate and under what conditions) and principled (it depends on a reading of IHL that the field has long rejected). This is the middle ground.
Geoffrey Corn and Laurie Blank apply a continuous-combat-function and reasonable-attacker standard. The attacker’s good-faith assessment, made under operational time constraints, is the relevant legal benchmark for evaluating proportionality compliance. This framing is consistent with the U.S. military’s operational implementation of CDE methodology.
What all camps agree on: the party using human shields commits a violation of international law regardless of whether the shields are attacked. The ICTY’s Blaškić Trial Chamber held that using civilians as shields constitutes inhuman treatment under customary international humanitarian law, and that the crime is complete upon the act of placing civilians at risk regardless of whether the shielded sites are actually attacked or harmed. Iran’s IRGC has, by the documentation cataloged above, completed the Article 51(7) violation. The completion of the violation does not depend on coalition strikes producing civilian harm at the shielded sites.8
Who Owns the Dead
The legal framing and the political framing produce different answers to the same question. The question is who bears responsibility for the civilian deaths produced when an attacker strikes a target the defender has placed alongside civilian infrastructure.
The legal framing, under the Schmitt-ICRC consensus, is that the attacker bears full proportionality liability. Article 51(8) holds attacker obligations under Articles 51(5)(b) and 57 undiminished by defender misconduct. Under this framing, the approximately one thousand thirty Iranian civilian deaths attributable to coalition strikes are the legal responsibility of the coalition, with the IRGC bearing a separate legal liability for its Article 51(7) and Article 58 violations.
The political framing, under the Dinstein-leaning analysis, is that the defender absorbs much of the moral culpability for civilian deaths attributable to its own Article 58 violations. Under this framing, the approximately six hundred to eight hundred deaths above the no-co-location counterfactual are politically attributable to the IRGC’s documented embedding doctrine, with the coalition bearing legal liability for the proportionality assessment but not the moral weight of the deaths the embedding produced.
Both framings operate on the same factual record. The disagreement is normative, not empirical. Both framings agree on the underlying numbers: approximately one thousand thirty civilian deaths in total; approximately two hundred to four hundred predicted under no-co-location norms; approximately six hundred to eight hundred above that counterfactual.
The forensic case has nothing to add to the normative disagreement. The forensic case establishes the numbers and the documented Article 58 conduct that produced them. The judgment of how to allocate responsibility between attacker and defender is a separate question that depends on which scholarly framework the reader applies. The Law Is Clear walks through the broader IHL framework on human shields and the asymmetric enforcement environment that follows from it.
The Caveat
Three caveats are load-bearing for the article’s claims.
First, Article 51(8) is explicit and uncontested across all three scholarly camps. Defender violation of Article 58 does not relieve the attacker of obligations under Articles 51(5)(b) and 57. The campaign’s specific strikes, including Minab, remain subject to attacker-side legal evaluation regardless of the defender’s conduct. The full forensic on the Minab strike lives in The Minab Test; the institutional vetting failure documented there is real and consequential under any reading of the IHL framework.
Second, the reverse counterfactual produces a directional finding rather than a precise measurement. The actual gap between predicted and observed civilian deaths could be larger or smaller than the six hundred to eight hundred range depending on which historical benchmark is applied. The directional finding (a substantial portion of Iran 2026 civilian deaths fall above the no-co-location counterfactual) is robust to plausible variations in the benchmark.
Third, the documented Article 58 violations cataloged in this article are a floor rather than a ceiling. The twelve documented incidents and the Tehran tunnel network are what open-source investigators and named-victim documenters have been able to verify. Iran’s internet blackout during Operation Epic Fury, the longest state-imposed shutdown ever recorded, prevented documentation of additional cases that almost certainly occurred. The full extent of the IRGC embedding pattern is not known and may not become known for years.
What is known is enough. The documented record of Article 58 violations is sufficient to establish that the campaign’s civilian-share rate of thirteen point five percent reflects two factors operating inside the predicted band: the gutting of the Pentagon CHMR apparatus in 2025, and the systematic IRGC pattern of embedding military assets in civilian infrastructure. Both factors are real. Both produced civilian deaths. The legal allocation of responsibility for those deaths is what scholars disagree about; the empirical measurement of how each factor pushed the rate from the predicted floor to the predicted ceiling is what this article has tried to make visible.
This article is a deep dive sub-article of Six to One. For the historical comparator class against which the expected-ratio counterfactual is measured, see The Historical Band. For the methodology behind the named-victim verification that produced the 1,030 figure, see The Verification Floor. For the documented catalog of co-location incidents, see Twelve Incidents, Nine Cities. For the broader IHL framework on human shields, see The Law Is Clear. For the apex case, see The Minab Test. For the US targeting process and CDE methodology, see How Targeting Actually Works.
Footnotes
-
Khamenei tunnel network: Iran International OSINT investigation documented underground tunnels running beneath Shahid Shourideh medical center, Hejrat Girls’ Elementary School, Tohid Mosque, and Karimeh Ahl-e Beit dental and medical clinic, connecting to Khamenei residential compound and associated Office of the Supreme Leader facilities. Construction documentation predates the 2025-2026 war by years. Source: Iran International February-March 2026 OSINT investigation; corroborated by FDD Iran Update reporting and CSIS Iran satellite analysis. ↩
-
Additional Protocol I to the Geneva Conventions, adopted 1977, ratified by Iran. Article 51(7): prohibition on use of civilians to render military objectives immune from attack. Article 51(8): defender violation does not relieve attacker of obligations under Article 57 (precaution) and Article 51(5)(b) (proportionality). Article 58: defender obligation to remove civilians from military vicinity, avoid locating military objectives in densely populated areas, take other necessary protective precautions, “to the maximum extent feasible.” Source: ICRC AP I commentary (1987); Sandoz, Swinarski, Zimmermann. ↩
-
IRGC operational doctrine described as “concealment of strategic assets in places the enemy cannot target” per Foundation for Defense of Democracies 2024 assessment of IRGC force structure. Hengaw standing assessment: “Iranian military forces have relocated into civilian spaces, including schools, student dormitories, and mosques in residential areas,” constitute “the use of human shields under the Geneva Conventions.” Source: FDD IRGC force structure assessment 2024; Hengaw serial reports March-April 2026. ↩
-
Documented IRGC co-location incidents during 2025-2026 conflict period: Sardasht (Kosar Girls’ Elementary School, Hengaw); Kermanshah (Farabi Hospital vicinity, Hengaw); Ilam (Imam Khomeini Hospital and Rababeh Kamali School, AP/Mnemonic/US State Department); Tehran (Sina Hospital, The Lancet/AP); Tehran (5+ hospitals on January 8, AP/IHR); Tehran (Khatam al-Anbiya Hospital, Iran International); Tehran (Mofatteh 2 School District 15, multiple); Tehran (Hamdaman rehabilitation institute orphan-girls facility, FDD March 26 2026); Arak (Imam Ali high school, IHR/Hengaw); Minab (Shajareh Tayyebeh elementary school adjacent to IRGC Naval Forces compound and Shahid Absalan IRGC Navy medical clinic, multiple). Tehran tunnel network: Khamenei compound tunnels beneath Shahid Shourideh medical center, Hejrat Girls’ Elementary School, Tohid Mosque, Karimeh Ahl-e Beit dental and medical clinic (Iran International). Full forensic catalog: Twelve Incidents, Nine Cities. ↩
-
CDE Methodology under Chairman of the Joint Chiefs of Staff Instruction 3160.01 (Levels 1-5). Operation Epic Fury target catalogue composition derived from CENTCOM, IDF, ACLED, Hudson Institute, CSIS, and TWZ campaign reporting through April 2026. Predicted civilian-share rates by category: ~3-6 percent for IRGC bases / missile production / air defense in semi-rural settings; ~10-18 percent for Tehran-embedded leadership and command nodes; 20-40 percent for dual-use or contested targets even with maximum precaution. Weighted-average expected civilian share: ~9-13 percent. CHMR apparatus reduction approximately 90 percent in 2025 under Defense Secretary Hegseth, against recorded objections of CENTCOM commander Gen. Erik Kurilla. Sources: Politico, ProPublica, NPR. ↩
-
NATO Operation Unified Protector, Libya 2011 civilian-share norms: ~4-8 percent civilian share per HRW retrospective and Airwars 2021 retrospective. Operation Epic Fury strike count approximately 5,000-11,000 documented strikes through April 2026 across 24 of 31 Iranian provinces. Reverse counterfactual: applying Libya 2011 civilian-share norm to Iran 2026 strike count predicts approximately 200-400 Iranian civilian deaths under no-co-location conditions. Hengaw documented approximately 1,030 civilians. Gap: approximately 600-800 deaths above the no-co-location counterfactual. AP I Article 51(8) holds attacker obligations under Articles 57 and 51(5)(b) undiminished regardless of defender violation. Sources: HRW Unacknowledged Deaths report; Airwars Libya 2011 retrospective; ICRC Customary IHL. ↩
-
Scholarly debate on proportionality with shields. Yoram Dinstein, The Conduct of Hostilities Under the Law of International Armed Conflict, 4th ed. (relaxation camp). Beth Van Schaack, Stanford, formerly U.S. Ambassador-at-Large for Global Criminal Justice (no-relaxation camp). Michael Schmitt, Lieber Institute West Point, lead author Tallinn Manual (middle ground; voluntary vs. involuntary shield distinction). Geoffrey Corn and Laurie Blank (continuous-combat-function and reasonable-attacker standard). U.S. Department of Defense Law of War Manual and U.K. Joint Service Manual treat defender responsibility as relevant factor. ICRC commentary maintains all shields retain full civilian status. Amnesty International and Human Rights Watch maintain same position. Source: International Review of the Red Cross 872 (2009); Schmitt various; Dinstein Conduct of Hostilities 4th ed. ↩
-
ICTY Prosecutor v. Blaškić (IT-95-14), Trial Chamber: using civilians as shields constitutes inhuman treatment and cruel treatment under customary international humanitarian law; mental suffering alone suffices for the violation even without physical harm. The crime is complete upon the act of placing civilians at risk regardless of whether the shields are attacked. ICTY Prosecutor v. Galić (IT-98-29), Trial and Appeals Chambers: defending party’s failure to abide by Article 58 obligation does not relieve attacking side of duty to abide by principles of distinction and proportionality. Specific paragraph references in the judgments left to the secondary legal literature; the principle is uncontested across IHL scholarship. ↩