'Lavender': The AI machine directing Israel's bombing spree in Gaza

Summary

Yuval Abraham, reporting for +972 Magazine in partnership with Local Call, reveals on the testimony of six Israeli intelligence officers with first-hand involvement that the Israeli army’s AI program Lavender marked as many as 37,000 Palestinians as suspected Hamas and Islamic Jihad militants and their homes for possible airstrikes in the early weeks of the war, that officers adopted its kill lists with a human check of about 20 seconds limited to confirming the target was male despite a known error rate of about 10 percent, that tracking systems including Where’s Daddy? signaled when a marked person entered his family home so the home could be bombed, usually at night with the family present, that junior targets were struck only with unguided “dumb” bombs, and that fixed collateral-damage degrees permitted up to 15 or 20 civilians per junior operative and hundreds per senior commander (‘Lavender’: The AI machine directing Israel’s bombing spree in Gaza, lines 88-106).

What the file is

• An investigation in +972 Magazine, in partnership with Local Call, published April 3, 2024 as displayed in the text copy (line 81) and in the original HTML’s structured data (datePublished 2024-04-03), downloaded from https://www.972mag.com/lavender-ai-israeli-army-gaza/ (access basis: publisher-free). The original is the HTML page; the text copy supplied with it (about 8,973 words, lines 1-462) is a plain-text rendering of that page, with no page markers, so locators are line numbers.
• The article itself runs lines 86-390, organized as an opening followed by six numbered steps. The rest of the text copy is the page frame: navigation and support prompts (lines 1-66), photo captions throughout, then a “Most read” frame, tags, the author’s biography, a membership appeal whose text belongs to the page as captured after the war continued, and the magazine footing (lines 392-462).
• Text quality: the article’s prose is clean and complete in both copies. Two printed slips in photo captions are recorded under Open questions, not corrected.

Section by section

• Opening (lines 86-112): the 2021 book The Human-Machine Team, released in English under the pen name “Brigadier General Y.S.” and confirmed to be by the current commander of Unit 8200, argues for a machine that could rapidly process massive amounts of data to generate thousands of potential “targets,” resolving a “human bottleneck for both locating the new targets and decision-making to approve the targets” (line 86). Lavender is that machine, unveiled here for the first time: six intelligence officers say it played a central role in the bombing, its outputs treated “as if it were a human decision” (line 88). Formally it marks all suspected operatives in the military wings of Hamas and PIJ, including low-ranking ones; in the first weeks the army almost completely relied on it, which clocked as many as 37,000 Palestinians (line 90). Officers gave sweeping approval to its kill lists with no requirement to check its reasoning or raw data; a source’s normal check was about “20 seconds,” only to ensure the target was male, despite known “errors” in approximately 10 percent of cases (line 92). Targets were systematically attacked in their homes, usually at night with whole families present, because locating them there was easier; “Where’s Daddy?” and other automated systems tracked them and carried out bombings when they entered their family’s residences (line 94). Thousands of Palestinians, most women and children or people not involved in the fighting, were killed, sources testified (line 98). A.: “the IDF bombed them in homes without hesitation, as a first option” (line 100). Lavender marks people and puts them on a kill list, whereas “The Gospel” — Habsora — marks buildings and structures (line 102). For junior targets the army preferred unguided “dumb” bombs: “You don’t want to waste expensive bombs on unimportant people” (line 104). For every junior operative, up to 15 or 20 civilians were permissible, where in the past no collateral damage was authorized for low-ranking militants; for a battalion or brigade commander, on several occasions more than 100 civilians (line 106). The investigation follows six stages: generating targets, linking them to family homes, choosing a weapon, authorizing civilian casualties, calculating collateral damage, and bombing a family home (lines 110-112).
• Step 1: Generating targets (lines 114-208): in the past a “human target” was a senior operative who could be killed in his private home even with civilians around, marked very carefully, only senior commanders bombed at home, to maintain proportionality (line 118). After October 7 — around 1,200 killed and 240 abducted, as the article gives it — under Operation Iron Swords the army designated all operatives of Hamas’ military wing as human targets regardless of rank (line 120). The old “incrimination” process for a single target was complex and lengthy (line 122); with tens of thousands of lower-ranking operatives, AI did most of the work, and four sources say Lavender marked some 37,000 Palestinians, most junior (the IDF Spokesperson denied such a kill list) (line 126). “We didn’t know who the junior operatives were, because Israel didn’t track them routinely [before the war]” (line 128). Approval to automatically adopt the lists, previously an auxiliary tool, came about two weeks in, after a manual sample of several hundred found 90 percent accuracy (line 130). “At 5 a.m., [the air force] would come and bomb all the houses that we had marked” (line 132); one source nicknamed the low-ranking targets “garbage targets” (line 134). On Palestinian Health Ministry data, some 15,000 Palestinians — almost half the toll so far — were killed in the first six weeks, to the Nov. 24 ceasefire (line 136). Lavender analyzes mass-surveillance information on most of Gaza’s 2.3 million residents and gives almost every person a rating from 1 to 100 (line 142); it learns “features” of known operatives from training data and locates the same features in the general population (line 144). The Unit 8200 commander’s book calls humans a “bottleneck” (line 146) and guides building a “target machine”: features include being in a Whatsapp group with a known militant, changing cell phone every few months, and changing addresses frequently (line 148); “The more information, and the more variety, the better” (line 150). Footage and slides from a 2023 Tel Aviv University lecture by “Col. Yoav,” commander of Unit 8200’s Data Science and AI center, describe a machine first used in May 2021 to identify Hamas missile squad commanders: “eventually, people of flesh and blood take the decisions” (lines 152-162). In practice, sources say, human agency was substituted by mass target creation. Officers were not required to independently review assessments (line 168); “Everything was statistical, everything was neat — it was very dry” (line 170), despite internal checks at 90 percent accuracy, meaning 10 percent of those slated were not members of Hamas’ military wing at all. Mistakenly flagged: police and civil defense workers, militants’ relatives, residents with a name and nickname identical to an operative’s, and Gazans who used a device that once belonged to an operative (line 172); “It’s a vague boundary” (line 174). Phones change hands constantly in war (line 178). “There was no ‘zero-error’ policy. Mistakes were treated statistically” (line 180). B.: “The machine did it coldly. And that made it easier.” (line 182). A defending source accepted the margin of error for junior militants (line 184). Pressure was constant — “Bring us more targets” — and lowering the threshold marked more people: “At its peak, the system managed to generate 37,000 people as potential human targets” (line 188), at times bringing civil defense personnel and police officers (line 188). Training data included employees of the Hamas-run Internal Security Ministry whom a data-science source does not consider militants, including civil defense workers, which made Lavender more likely to select civilians with a “civilian communication profile” (lines 192-194). The IDF Spokesperson denied using AI to incriminate targets, calling the systems “auxiliary tools” and saying an independent analyst examination is required (line 198). Sources say the only protocol for junior targets was the male/female check (line 200): “I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added value as a human, apart from being a stamp of approval.” (line 204). For men marked in error there was no mechanism to catch the mistake; a common error was a target giving his phone to his son, older brother, or a random man, who “will be bombed in his house with his family” (line 208).
• Step 2: Linking targets to family homes (lines 210-244): the IDF Spokesperson’s statement — Hamas “systematically uses the civilian population as human shields,” the IDF “is bound by and acts according to international law, directing its attacks only at military targets and military operatives” — is given at line 216. The six sources echoed it to some degree on tunnels under hospitals and schools, ambulances, and assets near civilian buildings, a characterization human rights groups warn evades Israel’s onus (line 218). But sources explain the unprecedented toll in part by the systematic choice to attack targets in private homes alongside families, because automated systems could mark family houses easily (lines 220-222). Surveillance could automatically “link” individuals to family houses; tracking programs alert the targeting officer when they are home, one revealed here for the first time being “Where’s Daddy?” (line 226). “It’s called broad hunting: you copy-paste from the lists that the target system produces.” (line 228). In the first month, more than half the fatalities — 6,120 people — belonged to 1,340 families, on UN figures, a proportion of entire families bombed at home much higher than in 2014 (line 230). A low-ranking officer added “something like 1,200 new targets” to the tracking system on his own accord when attacks decreased (line 234). Initially inputted in the first two weeks were “several thousand,” including all Nukhba members, all anti-tank operatives, and anyone who entered Israel on October 7 (line 236); then, “In the end it was everyone [marked by Lavender]” — tens of thousands, including, a source said, some minors, as “operatives are over the age of 17, but that was not a condition” (line 238). Combined, the systems collapsed a house on everyone inside as soon as the marked person set foot in it (line 242): “Usually, these 10 will be women and children. So absurdly, it turns out that most of the people you killed were women and children.” (line 244).
• Step 3: Choosing a weapon (lines 246-258): once a target is marked, verified male, and located at home, the munition is picked (line 250). CNN in December 2023, on U.S. intelligence estimates, put dumb bombs at about 45 percent of air force munitions in Gaza; an army spokesperson answered that the army was “devoting vast resources to minimizing harm to the civilians that Hamas has forced into the role of human shields” (line 252). Three sources say junior operatives marked by Lavender were assassinated only with dumb bombs, to save expensive armaments: a junior target in a high-rise would not be struck, because a precise “floor bomb” was not worth spending, but a junior target in a building of a few floors could be killed, with everyone in the building, with a dumb bomb (line 254). C.: “we usually carried out the attacks with dumb bombs, and that meant literally destroying the whole house on top of its occupants. But even if an attack is averted, you don’t care — you immediately move on to the next target. Because of the system, the targets never end. You have another 36,000 waiting.” (line 258).
• Step 4: Authorizing civilian casualties (lines 260-342): one source put the fixed collateral-damage degree for junior operatives in the initial weeks at up to 20, another at up to 15, applied broadly regardless of rank, military importance, and age, with no case-by-case weighing (line 264). A., a target operation room officer, said the international law department had never given such “sweeping approval”: “they directly tell you: ‘You are allowed to kill them along with many civilians.’” (line 266); “Every person who wore a Hamas uniform in the past year or two could be bombed with 20 [civilians killed as] collateral damage, even without special permission” and “In practice, the principle of proportionality did not exist.” (line 268). “it could also be 20 children for a junior operative”; the rationale, A. replied, was “Lethality.” (line 270). The fixed degree saved time and accelerated mass target creation; B.’s figure for the first week was fifteen, varying over time (line 274): “At first we attacked almost without considering collateral damage” — after a week the number dropped from 15 to five, “which made it really difficult for us to attack, because if the whole family was home, we couldn’t bomb it. Then they raised the number again.” (line 276). Sources say the army is no longer mass-generating junior targets, partly due to American pressure, and because most homes were destroyed or damaged and the population displaced (line 280); E. says the junior bombardment lasted only the first week or two and stopped mainly to not waste bombs, under a “munitions economy” and fear of a northern war with Hezbollah (line 282). Senior commander strikes continue, authorizing “hundreds” per target, with no historical precedent in Israel or recent U.S. operations, sources say (line 284). For the Dec. 2 bombing aimed at Shuja’iya Battalion commander Wisam Farhat, “we knew that we would kill over 100 civilians” (line 286). Amjad Al-Sheikh, at a supermarket, heard five blasts (line 290), ran home to rubble and screams (line 292), and found 11 family members dead; his wife and baby daughter survived under a fallen closet; B’Tselem is cited for dozens of buildings destroyed, dozens killed, hundreds buried that day (line 294). For Ayman Nofal, commander of Hamas’ Central Gaza Brigade, a source said approximately 300 civilians were authorized, destroying several buildings in Al-Bureij refugee camp on Oct. 17 on an imprecise pinpointing (line 298). Resident Amro Al-Khatib: “Between 16 to 18 houses were wiped out in the attack” (line 300); around 50 dead bodies were pulled out the first day and around 200 wounded, with five days of digging (line 302). Paramedic Nael Al-Bahisi counted 50-70 casualties the first day: “Entire families with children were killed.” (line 306). A high-rise in Rafah was destroyed in mid-December killing “dozens of civilians” in an attempt to kill Rafah Brigade commander Mohammed Shabaneh (line 308); witness Wael Al-Sir, 55, said the Dec. 20 bombing destroyed an “entire residential block” and killed at least 10 children, most of the injured being children (line 310). D.: the policy was “so permissive that in my opinion it had an element of revenge,” with a calculation — “how many for a brigade commander, how many for a battalion commander, and so on” (line 312). E.: “We’ve killed people with collateral damage in the high double-digits, if not low triple-digits. These are things that haven’t happened before.” (line 314). The comparison is the United States in Iraq, Syria, and Afghanistan (line 318): General Peter Gersten said a collateral damage of 15 deviated from procedure and needed special permission from Central Command head General Lloyd Austin (line 320); “With Osama Bin Laden, you’d have an NCV [Non-combatant Casualty Value] of 30, but if you had a low-level commander, his NCV was typically zero” (line 322). All sources said October 7 and the hostage kidnapping greatly influenced fire policy: “At first, the atmosphere was painful and vindictive” (line 326); a “dissonance” between frustration at not attacking enough and another thousand Gazans dead (line 328); “hysteria in the professional ranks” (line 330). D. was not explicitly told the goal was “revenge,” but a policy so permissive “loses all meaning” (line 334). A.: “We were told: now we have to fuck up Hamas, no matter what the cost. Whatever you can, you bomb.” (line 336). B., in retrospect, believes the “disproportionate” policy endangers Israelis long-term, as bereaved families — “which is nearly everyone” — will raise motivation to join Hamas (lines 338-340). The army’s denial is given: “each target is examined individually” and “The IDF does not carry out attacks when the collateral damage expected from the attack is excessive in relation to the military advantage.” (line 342).
• Step 5: Calculating collateral damage (lines 344-360): the calculation of civilians expected to be killed in each house, examined in the previous investigation, was after October 7 conducted with automatic and inaccurate tools, the previous wars’ thorough verification largely abandoned for automation (line 348). The New York Times in October reported a phone-data system giving a live estimate of how many fled north to south, colored red, green and yellow; Brig. General Udi Ben Muha: “It’s not a 100 percent perfect system — but it gives you the information you need to make a decision.” (line 350). Sources describe a similar system for collateral damage: pre-war residents of each home, from building size and resident lists, reduced by the proportion supposedly evacuated from the neighborhood — a house usually of 10 counted as five if half the neighborhood was estimated to have left — with no surveillance to check actual occupancy, to save time (lines 354-356). “This model was not connected to reality” (line 358); a house was bombed without knowing several families were inside hiding together. The model was adopted because it was faster, the calculation “completely automatic and statistical” — even producing figures that were not whole numbers (line 360).
• Step 6: Bombing a family home (lines 362-390): a substantial gap could open between a Where’s Daddy? alert that a target had entered his house and the bombing, killing whole families without hitting the target: “It happened to me many times that we attacked a house, but the person wasn’t even home,” one source said. “The result is that you killed a family for no reason.” (line 366). Three sources witnessed an incident in which the intended target was not inside, no further verification having been conducted in real time (line 368). A target might have been home earlier, then at night slept elsewhere, “say underground” (line 372). In one incident, the target understood to be home at 8 p.m. had by the 3 a.m. bombing moved to another house with his family; the building bombed held two other families with children (line 374). Bomb damage assessment (BDA) — the previous wars’ post-strike check, in the earlier investigation involving listening to relatives’ phone calls — was abolished at least for junior militants marked using AI, to save time; sources did not know how many civilians were killed in each strike, or, for the low-ranking suspected operatives, whether the target himself was killed (line 376). A source told Local Call for a previous investigation published in January: “You don’t know exactly how many you killed, and who you killed” (line 386). Amjad Al-Sheikh closes the article still searching: “Until now, there are bodies under the rubble,” he said. “Fourteen residential buildings were bombed with their residents inside. Some of my relatives and neighbors are still buried.” (line 390).

Key passages

• “human bottleneck for both locating the new targets and decision-making to approve the targets” (line 86)
• “as if it were a human decision” (line 88)
• “the IDF bombed them in homes without hesitation, as a first option” (line 100)
• “You don’t want to waste expensive bombs on unimportant people” (line 104)
• “Once you go automatic, target generation goes crazy” (line 128)
• “At 5 a.m., [the air force] would come and bomb all the houses that we had marked” (line 132)
• “The more information, and the more variety, the better” (line 150)
• “eventually, people of flesh and blood take the decisions” (line 162)
• “There was no ‘zero-error’ policy. Mistakes were treated statistically” (line 180)
• “The machine did it coldly. And that made it easier.” (line 182)
• “I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added value as a human, apart from being a stamp of approval.” (line 204)
• “It’s called broad hunting: you copy-paste from the lists that the target system produces.” (line 228)
• “most of the people you killed were women and children” (line 244)
• “You have another 36,000 waiting.” (line 258)
• “In practice, the principle of proportionality did not exist.” (line 268)
• “Lethality.” (line 270)
• “In the bombing of the commander of the Shuja’iya Battalion, we knew that we would kill over 100 civilians” (line 286)
• “This model was not connected to reality” (line 358)
• “The result is that you killed a family for no reason.” (line 366)
• “You don’t know exactly how many you killed, and who you killed” (line 386)
• “Until now, there are bodies under the rubble” (line 390)

Evidence the work relies on

• Testimony of six Israeli intelligence officers, all of whom served in the army during the current war on the Gaza Strip and had first-hand involvement with the use of AI to generate targets for assassination (line 88). They are identified in the article only by initials — A., B., C., D., E. — with roles given at points: A. an intelligence officer and target operation room officer, B. a senior officer who used Lavender, C. a user of various automated programs, E. and D. intelligence sources drafted after October 7.
• The 2021 book The Human-Machine Team, under the pen name “Brigadier General Y.S.,” its author confirmed as the current commander of Unit 8200 — five sources in 8200 confirmed the commander is the author, as reported also by Haaretz (lines 86, 146).
• Footage and slides, obtained by +972 and Local Call, of a private lecture by “Col. Yoav,” commander of Unit 8200’s Data Science and AI center, at Tel Aviv University’s AI week in 2023, reported on at the time in the Israeli media (lines 152-162).
• Palestinian testimonies named in the piece: Amjad Al-Sheikh on the Dec. 2 Shuja’iya bombing and his 11 family members killed (lines 290-294, 388-390), Amro Al-Khatib and paramedic Nael Al-Bahisi on the Oct. 17 Al-Bureij strike (lines 300-306), and Wael Al-Sir on the Dec. 20 Rafah bombing (line 310).
• Official statements by the IDF Spokesperson and the army, reproduced at points throughout: the denial of a kill list (line 126), the denial of AI incrimination and the “auxiliary tools” statement (line 198), the human-shields and international-law statement (line 216), the CNN-response spokesperson statement on dumb bombs (line 252), and the denial that each target is not individually examined (line 342).
• Data and reports as the piece credits them: Palestinian Health Ministry data for some 15,000 killed in the first six weeks (line 136); UN figures for 6,120 of the first month’s fatalities belonging to 1,340 families (line 230); CNN’s December 2023 report, on U.S. intelligence estimates, for the 45 percent dumb-bomb share (line 252); B’Tselem for the Dec. 2 Shuja’iya destruction (line 294); satellite footage and videos for Al-Bureij (line 298); the New York Times’ October report on the phone-data evacuation system, with Brig. General Udi Ben Muha quoted (line 350); and a U.S. defense magazine’s 2021 account of General Peter Gersten for the U.S. NCV comparison (lines 320-322).
• A source’s statement to Local Call for a previous investigation published in January, on not knowing how many were killed (line 386), and the magazines’ own previous investigations, in November 2023 and January, referenced for The Gospel, the target-file procedure, and BDA (lines 102, 348, 376).

Analysis

• Method: an investigation built on insider testimony at each stage of a six-stage production line — generation, home-linking, weapon choice, casualty authorization, casualty calculation, and the strike itself — with the mechanism documented stage by stage rather than a single practice asserted, official denials reproduced at the stages they answer, and Palestinian testimonies supplying the named cases for the two senior-commander strikes the piece examines in detail (Shuja’iya, Dec. 2; Al-Bureij, Oct. 17).
• Perspective: Abraham writes as a journalist reporting sources who are themselves divided: B. trusts the statistical mechanism more than a grieving soldier and defends its coldness, a second source defends the margin of error for junior militants, while other sources call the targets “garbage targets,” the boundary vague, and the policy revenge-like. The piece holds those positions in the sources’ own terms alongside the army’s denials, rather than resolving them.
• Limits, on the piece’s own terms: the six central sources are anonymous and identified only by initials, and cannot be checked from the piece; its two fixed collateral figures for junior operatives conflict — one source says up to 20, another up to 15 — and both are recorded (see Open questions); the piece states that sources did not know, for AI-marked junior operatives, whether the target himself was killed or how many civilians each strike killed, because BDA was abolished for them (line 376); the Shabaneh strike’s outcome is stated to be unclear (line 308); and the piece is dated April 3, 2024, covering the war’s early stages, so its tolls and its present-tense claims about current policy are those available then.

People, organizations, places and events

Yuval Abraham, Unit 8200, Amjad Al-Sheikh, Ayman Nofal, Wisam Farhat, Mohammed Shabaneh, +972 Magazine, Local Call, Hamas, Islamic Jihad, Hezbollah, Israel Defense Forces, B’Tselem, Bureij camp, Gaza, Gaza City, Rafah, Operation Iron Swords

Concepts

Lavender, Where’s Daddy?, Habsora, The Human-Machine Team

How it relates to other sources

• ‘A mass assassination factory’: Inside Israel’s calculated bombing of Gaza is the November 2023 investigation this piece follows: it documented Habsora (“The Gospel”), Power targets, target files with pre-calculated civilian deaths, and the bombing of family homes, on seven intelligence sources. This piece names the companion system that marks people rather than buildings, adds the home-tracking systems, the male-only 20-second check, the fixed 15-or-20 degree for junior operatives, the statistical occupancy model, and the abolition of BDA for junior targets, and gives the authorized figures for two named senior-commander strikes — over 100 for Farhat, approximately 300 for Nofal — where the earlier piece reported a case of hundreds approved for a single top commander without naming the target. The earlier piece quoted a January investigation source’s counterpart passage on unknown casualties only in this piece (line 386); both pieces close, in different registers, on official policy continuing.
• Light in Gaza: Writings Born of Fire presents, in Nour Naim’s chapter, Unit 8200’s deep-learning programs — Alchemist, Gospel and Depth Wisdom — and Gaza as a laboratory for AI military technology without international monitoring (PDF p.150-153). Abraham’s piece is a later, insider-testimony account of one such program in operation in a specific war, naming Lavender and Where’s Daddy?, which Naim’s chapter does not name.
• Debunking the Genocide Allegations: A Reexamination of the Israel-Hamas War from October 7, 2023 to June 1, 2025 answers this report in its AI-targeting section: AI in the IDF is, in their account, an intelligence-processing and cross-referencing tool, not an autonomous decision-maker, approvals remaining in human hands, while hasty mechanical use by some soldiers is a grave concern (PDF p.117). Abraham’s sources describe approvals as a 20-second male check and a stamp of approval; Orbach et al.’s answer and Abraham’s account stand, attributed, on the Lavender page. Orbach et al. also discuss the reported upper limit of 20 civilian casualties per main target via +972, holding it is a maximum threshold, not an average (PDF p.111-113); Abraham’s sources describe the 15-or-20 degree as a fixed permission applied broadly to junior operatives.
• A Textbook Case of Genocide argues intent from stated declarations in the war’s first week; Abraham documents, from intelligence sources, a targeting mechanism and casualty policy that operated in the same early weeks, including the atmosphere sources describe as vindictive after October 7 (lines 326-336).

Open questions

• The fixed collateral-damage degree for junior operatives is given two ways: one source says it was fixed at up to 20, another that the fixed number was up to 15 (line 264); the opening gives the permission as “up to 15 or 20 civilians” (line 106), and B. gives fifteen for the first week, varying over time, dropping to five after a week and then raised again (lines 274-276). All figures are recorded as given; the piece does not reconcile them into one schedule.
• The photo caption at line 304 prints “digging with bear hands” and “killed dozens Palestinians,” as printed, confirmed at the byte level; the caption’s subject is Al-Maghazi refugee camp on November 5, 2023, placed in the article amid the Al-Bureij account, and the piece does not connect the caption to that strike.
• The photo caption at line 332 prints “Defence Minister Yoav Gallant,” with the British spelling, as printed, where the article otherwise uses American spelling.
• The slides caption at lines 156 and 158 is printed twice in succession, as printed.
• Whether Mohammed Shabaneh was killed in the mid-December Rafah high-rise strike is stated in the piece to be unclear (line 308).
• The piece gives the October 7 toll as around 1,200 killed and 240 abducted (line 120), where ‘A mass assassination factory’: Inside Israel’s calculated bombing of Gaza, on a Physicians for Human Rights Israel report, gives more than 840 civilians massacred, 350 soldiers and security personnel killed, and around 240 kidnapped (line 118 of that source). Both accounts stand as their sources give them.
• The membership appeal in the page frame as captured (lines 424-428) speaks of “Israel’s genocidal assault on Gaza,” language belonging to the frame, not to Abraham’s article, which does not itself use the word genocide for the campaign it documents.

Coverage

lines 1-462 of 462, read in full.