ShadowGraph Intelligence · 22 July 2026

“Benjamin Netanyahu is a war criminal.” The first fifteen hours of a six-word claim

The mayor's video statement captured in situ on X and Bluesky at the three-hour mark, on a dark intelligence-report cover.

At 8 PM Eastern on July 21, the official X account of New York City's mayor posted a two-minute video opening with six words. By midday the next day the mayor's accounts displayed roughly 118 million combined views, the Instagram reel had overtaken the X post, and the news agencies had taken the story worldwide. This report follows how the claim moved — and who moved it.

The window, by the numbers

Fifteen hours, four platforms, one six-word claim

205M
combined platform-displayed views at the day-two check: X 88M, Instagram 106M, Facebook 11M; Bluesky adds an estimated ≈2.4M more (§12)
1.35M
likes on the X post alone, with 283K reposts; the Instagram reel adds 8.2M likes and 1.39M reshares
436,766
posts in the cross-channel capture window, 2.74M engagements; the ≈52K-per-hour peak decayed into an overnight floor that never fell below ≈20K per hour
0
Community Notes on the post, two days in: proposed notes from hour two died in the queue, and no fact-check or platform action surfaced

Figures are platform-displayed counts and analyst-labeled samples, reviewed through the latest July 22, 2026 re-check. Every figure in this report is a floor, not a census (§16). Stance labels are hand-applied by me; no automated sentiment scores are used anywhere in this report (§12).

§ 01 · The Bottom Line

Each faction kept the half it needed

Two hundred five million views and counting, and nobody was in charge of any of it. The video carried a two-part message — a legal concession and a rhetorical escalation — and each faction kept the half it needed and discarded the other: supporters ran the moral-clarity half by retweet, opponents ran the concession half. The platforms' ranking mechanics did the rest, manufacturing consensus on one app and a brawl on another from the same 129 seconds. The most-viewed voice in the whole war was a reporter who simply typed what the video said. No bot army pushed it, and no referee ever showed up — the spread was fast, adversarial, and by every test I ran, human. Five findings, then the evidence. Dotted-underlined phrases link to the actual posts on X.

  1. Most impactful: one scheduled post out-distributed the world's newsrooms: ≈118M displayed views by midday, ≈205M by the day-two check from a four-platform drop, with topic volume up tenfold and never falling back.
  2. Most surprising: the plain transcript beat every spin. A reporter's verbatim quote is still the most-viewed post of the war: 3.12M impressions, fifteen hours running.
  3. Most meaningful: same video, opposite realities. Consensus on Instagram (98.9% of comment like-mass), a brawl on X (41% hostile quotes). The ranking mechanics, not the audience, decide what each viewer thinks the country thinks.
  4. Fastest-moving: the othering frame went commentators → podcast host → sitting member of Congress in under three hours; the mayor's own left flank grew ×127 overnight.
  5. Least of all: no significant coordination, and no referee. Two clean authenticity passes, and still zero Community Notes, fact-checks, or platform action on a two-hundred-million-view claim.
The anchor post on X at T+15:30: 41.7M views, 1M likes, 222K reposts.
The anchor on X the morning after (counts have since grown). The same video runs on all four of the mayor's surfaces:
§ 02 · The Primary Source

What the video actually says

Nearly every downstream frame is a compression of this text — from the video's own captioning, verified against three city-hall reporters' quotes filed within six minutes.

  • “Benjamin Netanyahu is a war criminal… responsible for the killing of more than 73,000 people… There is a reason the International Criminal Court issued a warrant for his arrest.”
  • “It is clear we do not have the independent legal authority to enforce this warrant. The federal government, however, does, and I call on them to join the ICC and execute this warrant.”
  • “Benjamin Netanyahu is not welcome in New York City. Nor is any other war criminal at large.”
⚑ For the record

The platform's own auto-captions dropped the “not” from the pivotal sentence — rendering it “we do have the independent legal authority” (verifiable on the post with captions on). Every journalist quoting the audio kept the “not.” Even the machine layer compressed the claim, in the most consequential direction possible.

§ 03 · The Run-Up

This did not start at 8 PM

The video is the third beat of a four-day exchange — a walk-back of the arrest option and an escalation of the language at once. That duality is what fractured downstream.

Nov 2024ICC issues the warrantWarrants for Netanyahu and Gallant; the US is not a Rome Statute party. The mayor cites the claim's origin; he does not invent it.
Jul 18–20The stage is loaded“Looking into” an arrest → the President vows it won't happen → Israel's UN envoy dismisses the mayor's jurisdiction.
Jul 21 · 8 PM ETThe staged dropX at 8 PM sharp · Instagram 8:08 · Bluesky 8:13 · Facebook the same evening. A scheduled distribution, not a tweet that traveled.
8–11 PMThe wave≈52,000 posts/hr at the 9:15 PM peak; fans first, institutions by minute 40, five languages by minute 60.
OvernightThe plateauThe floor never drops below ≈20,000 posts/hr, Bluesky-led. The story changes carriers, not size.
MorningThe agencies take itNews layer grows 80 → ~1,800 items; the duality headline syndicates 140+ times; ≈118M combined views by midday.

Landing at 8 PM sharp is a scheduled-publication signature, not an off-the-cuff post.

§ 04 · Who Carried It

The carriers, fifteen hours in

The four official anchors, then the biggest carriers in each lane of the X quote pool, ranked by the reach each quote displays now. Real accounts, real faces, current numbers.

The virality map: every tracked engagement, in time
x = when it engaged (square-root time scale, so the first hour has room) · y = lane · size = prominence — the larger of the account’s following and the quote’s displayed reach · accounts over 80K prominence wear their real profile picture (faces have a small legibility floor) · hover or tap for details
  • The cascade is a hub, not a chain. Every arc connects directly to the anchor. No relay layer of middlemen — which is what organic spread around an official source looks like.
  • Both sides posted on the same clock; their firepower ran on different clocks. Median arrival is ~50 minutes for both lanes. But support's reach came early (half posted by T+1:29), while opposition's came late (half after T+2:13) because its biggest carriers were the principals who kept arriving through hour eight. The amplification had its power early; the argument got stronger all night.
  • How to read it: solid arcs are quotes of the anchor, dashed arcs are the operator's own cross-posts; nodes are the labeled quotes sized by prominence. The labeled record thins after T+8; §05 carries the full window.
@ShadowGraphIntel · @talk2trav

Who qualifies: every hand-labeled account whose quote of the anchor displays at least 25,000 impressions at the ~T+14:30 refresh, capped at the six biggest per lane; the chip under each column counts the labeled accounts below that bar. Reach, not follower count, is the criterion, because reach is what puts a frame in front of an audience. The pool holds 12.6K quotes, so these are the visible head of a much longer tail. Lanes are my hand labels; overnight entrants were labeled on collection. The gray lane's leader is the story of the whole event: a reporter's verbatim transcription out-reaching every partisan frame.

§ 05 · Velocity

Two waves, then a plateau

Full-population volume in 15-minute buckets: a pre-video baseline of 1,000 to 1,600 posts per 15 minutes, a 12,929-post peak at 9:15 PM (about 52,000 an hour), then an overnight floor that never drops below about 5,000 per 15 minutes through late morning.
  • Wave one (minutes 0–10): fans, first responders, the night's most-viewed hostile quote (“LOL.,” 92K impressions at minute 4), and reporters posting the exact legal language by minute 7.
  • Wave two (minutes 40–60): institutions. Israel's UN ambassador, a sitting congressman, national columnists, relays into Arabic, Spanish, and Japanese. Five languages by minute 60.
  • Then the plateau: from the 9:15 PM peak of ≈52,000 posts per hour, the full population settled to a floor that never dropped below ≈20,000 posts per hour, all night and all morning. The topic was already running ≈5,000 an hour before the video; the drop multiplied it tenfold.
  • The morning handoff: the news layer thickens toward morning as the agencies take the story: 436,766 results and 2.74M engagements in the capture window. The story did not burn out; it changed carriers.
§ 06 · The Media Spectrum

Every outlet, one grid: politics × treatment

One event, one grid. Every rated outlet that covered the video, placed by two independent judgments: where the outlet sits politically (which you can know before reading a word), and how its actual piece treated the video, coded from the headline and lede rather than the outlet's label.

Coverage details
Rated news sources35
Leaning left13
Center8
Leaning right14
Bias distribution40% lean right
Last reviewedJuly 22 re-check
Placements follow widely used media-bias ratings; unrated aggregators and state-run agencies are excluded from the rated set (§16).
Bias distribution · 40% of rated sources lean right
Left4
Drop Site HuffPost The Guardian The Mirror
Lean left9
Al Jazeera NPR CNN ABC7 NY Yahoo News Haaretz TIME Politico Forward
Center8
AP CBS News TheWrap Bloomberg France 24 The Hill JTA Times of Israel
Lean right8
Daily Mail Forbes Ynet The Jerusalem Post Israel Hayom Fox News The Express Yeshiva World
Right6
Washington Examiner NY Post JNS GB News HotAir PJ Media
Hover or tap a logo for the outlet; click to open the piece it was coded from. Left-right buckets are the outlet's rating, not the piece's treatment — the grid below crosses the two.
@ShadowGraphIntel · @talk2trav
How each side covered the same 129 seconds
  • Led with the charge itself: claim-forward headlines quoting the six words, treating the video as the story's moral center.
  • The legal concession ran deep in the piece where it appeared at all; no rated left outlet led with "backs off."
  • The harshest left frame was the concession story — the mayor's own left flank, not the press, carried the "do your job" critique (§07).
  • The duality frame: both halves in one headline — "calls Netanyahu a war criminal, concedes New York cannot arrest him."
  • The agencies settled on this wording by morning and it syndicated 140+ times, making duality the default frame worldwide (§03).
  • The agency core held the neutral line exactly (AP, Bloomberg, The Guardian, CNN, NPR, Yahoo); other center-rated outlets split mildly both ways — claim-forward at CBS and TheWrap, solid criticism at JTA and Times of Israel — but none reached the attack edge (§06 grid).
  • Concession-forward: "backs off," "admits he can't," "abandons plan" — the climb-down is the story, the charge is context.
  • At the edge the treatment slides from criticism into attack framing; the harshest headlines attack the mayor, not the claim.
  • One outlier ran the claim itself as a tabloid splash — claim-forward because the claim is the story, not out of support.
  • Same video, three stories: which half of the two-part message leads the headline tracks the outlet's prior politics almost perfectly.
  • The cross quadrants are nearly empty: no rated right outlet leaned supportive, no rated left outlet attacked (§06 grid).
  • No outlet in the rated set adjudicated the claim itself — and neither does this report (§17).
@ShadowGraphIntel · @talk2trav

Takeaway: treatment tracks rating almost perfectly — the cross quadrants are nearly empty, the agency core sat exactly on the neutral line, and the lone right-rated supportive-lean piece is a tabloid that led with the claim because the claim is the story. 35 rated outlets; ring = the piece's treatment, coded from headline and lede; state-run agencies and unrated aggregators are excluded from the rated grid and covered below.

  • The tell is the verb. Claim-forward headlines came overwhelmingly from international and state-run outlets; US center outlets went uniformly neutral; right-rated outlets led with the concession or moved into attack. First-night tally: concession 22 · neutral 25 · claim-forward 5.
  • The morning belonged to the agencies, and the agencies took the duality. News items grew from 79 to ~1,800; the agency duality headline syndicated 141 times (a second agency's 73, the concession syndicate 63), with the same split in five languages.
  • One sequencing note: the President's “will not be arrested” and the Israeli PM's office's “kangaroo court” statements both pre-date the video. Morning posts recirculating them as reactions are mis-sequenced — a timeline compression worth its own rung in §11.
⚑ One flag worth its own line

Three items in the capture came from a domain belonging to the “Pravda” aggregation network, which independent researchers have widely documented as a Russian-aligned content-laundering ecosystem. It carried the harshest framing collected anywhere in the window, fusing the concession story with ethnic othering (“Ugandan NYC Mayor… BENT THE KNEE TO TRUMP”). Single-source observation, no amplification measured, but it shows state-adjacent aggregators harvesting the concession frame within hours of the post. Spectrum placements above follow widely used media-bias ratings; outlets without established ratings were classified only by their own headline language.

§ 07 · The Asymmetry

Who said what, and how far it went

The X quote lane at fifteen hours: opposed 6.64M displayed impressions (37%, top quote a congressman at 2.32M), supportive 5.73M (32%), news 3.16M on three quotes (18%), left-flank critique 2.51M (14%).
41% vs 29%
of the formation sample's quote posts, opposed vs supportive. Structural, not sinister: supporters amplify by retweet, opponents contest by quote.
18% on 3 quotes
the news lane's share of tracked reach at T+15, almost all of it one verbatim transcription. Transcription beat both factions' framings, all fifteen hours.
14%
of tracked reach now sits on the left-flank lane, on a handful of quotes. Its lead post out-reaches every opposition quote except the congressman's.
◆ Does this mean opposition beat support? No, and the distinction is the point

This chart measures the quote war: the argument surface, which on X structurally belongs to the opposition, because opponents contest with quotes while supporters amplify with retweets and likes, which create no new posts to label. Zoom out to the amplification measures and the picture inverts: the anchor itself holds 1.35M likes and 283K reposts, the biggest hostile quote holds 45.5K likes, Instagram's comment like-mass runs 98.9% supportive with 1.39M reshares, and Bluesky is a supporter-aligned channel amplifying rather than arguing (§12). Read together: support dominated distribution; opposition dominated argument. This chart shows the argument.

§ 08 · Same Video, Opposite Surfaces

The Instagram comment lane

Same video, opposite surfaces: on X the opposition holds 41% of quote posts and 35% of impressions; on Instagram supportive comments hold 55% by count and 98.9% of comment likes, with opposition near zero visibility.
≈159K
comments on the reel at the day-two re-pull (8.2M likes, 1.39M reshares, 106M displayed views). The labeled sample is the like-ranked surface captured the first night (180 comments); Meta's wall blocks deeper pulls (§16).
98.9%
of the sampled like-mass sits on supportive comments (35,408 likes). By visibility, it is not a contest at all.
8 likes
the combined total held by all 35 hostile comments in the sample. The top supportive comment alone holds 15,396.
  • X hands dissent its own distribution: a hostile quote travels to the quoter's audience, which is how opposition held 41% of quote posts.
  • Instagram ranks dissent out of sight: hostile comments exist, but the engagement-ranked surface buries them below thousands of supportive ones.
  • Same video, opposite realities. Audience composition explains some of it; the ranking mechanics alone guarantee most of it.
★ Tech reality: ranked comments manufacture consensus

Instagram does not show comments in order; it shows the ones its ranking system scores highest — engagement-heavy, early, and algorithmically favored. A viewer scrolling this thread sees a wall of high-like supportive comments and concludes "everyone agrees." That impression is an artifact of ranking, not a measurement of opinion: the 98.9% figure describes what the ranked surface makes visible, not what the ≈159K commenters believe. Dissent exists in the thread — it is simply rendered invisible, holding 8 likes across 35 hostile comments in the sample. The same mechanism in reverse governs X, where the quote mechanic hands dissent standalone distribution. Neither surface is a poll.

◆ Caveats and the gap, stated plainly

180 comments sampled of 21,590 on the reel at first-night capture — under 1%, but it is the like-ranked surface, which is exactly what an ordinary viewer sees, and that visibility is the thing being measured. Likes stand in for reach (Instagram exposes none); stance labels are mine. Re-sampling was re-attempted with the reel at ≈159K comments: the collection route now returns a single comment — Meta licenses no third-party comment data — so the first-night sample stands, and the stance mix of the other 159K comments is unmeasured (§14, §16). A logged-in harvest of the ranked thread at depth is the only way to close this.

§ 09 · The Narrative Leaderboard

Seven stories, one video

Each story is scored by √(share of quotes × share of current reach): a blend that only rewards narratives both widely told and widely seen. Not a contradiction of §07: the opposition out-reaches support as a lane but splits it across four stories, while support concentrates everything in one. One coalition tells one story; the other tells four.

The narrative leaderboard at fifteen hours: moral clarity 43.2 (31 quotes, 5.73M reach), blood libel 13.4, the verbatim words 10.0 on three quotes, arrest him anyway 8.9, attack the messenger 7.0 (climbed from last place), backed down 5.6, do your job 4.8.
  • 1. “Finally, moral clarity” — 43.2 (formation: 25.9). One story carrying support's entire weight: 31 of 55 mapped quotes, the 1,788-repost Bluesky hub, 98.9% of the Instagram like-mass. Message discipline is the supportive coalition's defining feature.
  • 2. “Blood libel / it's a lie” — 13.4 (17.0). The opposition's strongest story, carried by its most institutional voices (Israel's UN ambassador, a former US ambassador, columnists). It attacks the figures and their sourcing: the report's second contested axis.
  • 3. The verbatim words — 10.0 (15.9). Not a narrative at all, and still out-traveling every framing per quote. When raw text leads the reach table, the war is over compression, not invention.
  • 4. “Arrest him anyway” — 8.9 (7.2). The only riser besides moral clarity: the left flank's story concedes nothing to the mayor's critics; its target is his caution.
  • 5. Attack the messenger — 7.0 (3.7). The night's fastest climber by trajectory: last place at formation, fifth by morning, on two quotes — one a sitting congressman's.
  • The biggest faller: “all talk, he backed down” — 5.6 (14.4). The concession mockery that dominated hour one collapsed as the escalation stories scaled. Mockery burns hot and short; identity attack compounds.
Flow diagram from the video's two message halves (the concession and the escalation) into the competing narratives each faction built from them.
Takeaway: the mayor's own two-part message wrote both sides' scripts. The escalation half fed support's single story and the opposition's “it's a lie” counter; the concession half fed three separate opposition stories; and the raw, unspun text out-traveled every framing of it — the widest flow on the chart.

Below the cut: “do your job” — high count, minimal reach. A low score with rising carrier size (like #5 above) is the profile worth watching, not the ones already peaked.

§ 10 · The Escalation Ladder

From pundits to principals to Congress

×9.2
X views since T+3 (9.6M → 88M)
×12.0
Instagram views since T+3 (8.8M → 106M)
×26.5
Facebook views since T+3 (415K → 11M), the biggest relative jump
×10.9
combined displayed views since T+3 (≈18.8M → ≈205M)

The size story is above; this section is the institutional climb: the same message moved up the ladder hour by hour.

  • Hour two: principals replaced pundits. A former US ambassador to Israel (min 72), an actor with a seven-figure following (min 98), a podcast host escalating to “denaturalized and sent back to Uganda” (min 104) — and a sitting member of Congress by T+2:13 (“You are a Muslim terrorist,” now 2.32M impressions).
  • All night: the verbatim quote never lost the lead. The city-hall reporter's transcription grew 141K → 595K → 3.12M impressions, ahead of every partisan frame at every checkpoint.
  • Overnight: the left flank became the second-biggest quote of the night. Nerdeen Kiswani (@NerdeenKiswani), who led the lane at hour two, now holds 2.50M impressions, ~47× her hour-two figure. Proposed community notes appeared at hour two; none ever surfaced.
Share of tracked quote reach, formation vs T+15: opposed 40% to 37%, supportive 30% to 32%, news 28% to 18%, left-flank critique 3% to 14%.
Growth is table stakes in a viral event; composition change is the intelligence. The left flank went from 3% to 14% of tracked reach (+11 pts) while news gave up 9 points as partisan frames scaled. Samples in §16.
§ 11 · The Distortion Ladder

Each frame, graded against the primary text

Each rung grades a circulating frame against the primary text in §02, and the second chip says whether the frame took hold. Only frames with measurable circulation are laddered; two fringe frames are kept because they are the ancestors of the escalation that later reached elected office. This measures fidelity of propagation, not truth of the underlying claim. Tap a row for the exemplar and note.

R1“Mayor says NYC lacks authority to arrest Netanyahu; calls him a war criminal; not welcome in NYC”Took hold · the reach leaderFaithful

Exemplar · first two hours

City-hall reporters, news outlets, min 6–60.

Verdict vs. the primary text: FAITHFUL. Carries both halves of the message: the concession and the escalation, with attribution intact.

R2“Mamdani concedes / backs down / caves to Trump”Took hold · 63-copy syndicateCompressed

Exemplar · first two hours

“cede ante Trump… no cumplirá con su promesa” (Spanish-language account, min 53).

Verdict vs. the primary text: COMPRESSED. Drops the federal demand and the “not welcome” escalation.

R3“He counted dead Hamas terrorists as innocent victims”Fringe · 2 accounts, low reachMixed: grounded + overstated

Exemplar · first two hours

Two accounts posting near-identical text, min 24 and 43.

Verdict vs. the primary text: MIXED. The grounded half: the video cites “more than 73,000” as a total, and that total indisputably includes combatants; the share is contested, but pointing out that the figure is not all civilians is fair critique. The overstated half: the video never characterizes the dead as innocent, so “counted them as innocent victims” attributes a claim the text does not make. The observation is grounded; the attribution goes beyond the text.

R4“He is signaling his foot soldiers / followers to commit violence”Fringe itself · its family escalatedInvented

Exemplar · first two hours

Two conservative commentators, min 17–19.

Verdict vs. the primary text: INVENTED. No such language anywhere in the transcript.

R5“Blood libel courtesy of Hamas propaganda”Took hold · institutional voicesContested

Exemplar · first two hours

Israel's UN ambassador, a city councilwoman, min 49–53.

Verdict vs. the primary text: CONTESTED CHARACTERIZATION. This is the second contested axis (the figures' sourcing), argued rather than fabricated.

R6“He refuses to use the powers he still has”Took hold · ×127 overnightGrounded

Exemplar · first two hours

Left-flank activists, min 51–62.

Verdict vs. the primary text: GROUNDED EXTENSION. Builds on the video's own admission, then argues policy beyond it.

§ 12 · Off X Entirely

The Bluesky surge, and what collection can't see

An earlier draft charted a platform split showing Bluesky "out-carrying" X. I pulled it: it measured my collection lens — Bluesky's fully open firehose vs. a licensed slice of X vs. almost nothing from closed platforms — not the event. What survives honest scrutiny:

  • X remains the largest measured surface, by a wide margin. X-native collection logs 283K reposts and 18.6K quotes on the anchor alone — dwarfing every off-X channel's captured volume. Any capture row that ranks X below another platform reflects sampling limits, not spread.
  • Bluesky's footprint is real and unusually large for the platform. 2,283 captured posts in the first ~2h40m, and the mayor's verified @mayor.nyc.gov native post drew 86K likes · 24K reposts at the day-two check — huge numbers for Bluesky's scale. The posture there is amplification, not argument: the audience is heavily aligned with the mayor, and a single fan hub accounts for 1,788 reposts. Organic virality around an official anchor, not distributed coordination.
  • The news layer's growth is directional, not precise: online-news items grew from 2% to 11% of the capture window as the agencies took the story — a trend the same lens saw at both checkpoints, so the direction holds even if the levels are soft.
  • Facebook, YouTube, Truth Social, and TikTok are blind spots, not small players. Keyword capture reaches almost none of them; the Facebook reel alone displays 11M views. Their footprints are unmeasured — logged as collection gaps in §14, not ranked here.
◆ What Bluesky won't show, estimated: ≈2.4M views

Bluesky displays no view counts, and its API exposes no impressions at all, so industry “average engagement rates” for the platform are follower-based and cannot supply a like-to-view divisor. The defensible route is this video's own like-to-view ratios on the platforms that display both, recomputed at the day-two full-metrics pull: X (1.5%) implies ≈5.7M views from 86K likes, Instagram (8.6%) implies ≈1.0M. My central estimate is ≈2.4 million views, taken between those bounds and weighted toward the lower half because Bluesky's smaller, aligned, chronological-feed audience engages at higher rates per view. That is an analyst estimate from stated assumptions, not a platform figure; it is excluded from the ≈205M displayed-counts total, which would read ≈207M with the estimate included.

◆ One deliberate omission

No machine sentiment scores appear in this report. The tool graded 86% of the sample maximally negative, including plainly supportive posts, because “war criminal” is negative-valence whatever the poster's stance. Stance has to be labeled, not scored (§07, §08). The cross-channel pull is a single-vendor keyword capture; counts are floors.

§ 13 · Account Integrity

Bots and foreign influence, tested

Two full authenticity passes on this post's engagement pools (466 X accounts hydrated first night, 98 more at T+16; highest-risk re-verified live; my own collection). The spread is organic:

✓ No coordination edges ✓ Zero suspensions 2.6% individually inauthentic 0.01% of sampled reach
🔁Quote theft on both sides

A supporter's viral joke (4,105 likes, min 4) reposted verbatim at min 68; a pro-Israel commentator's attack line (min 20) reposted verbatim by a French jewelry-designer account at min 22; the former ambassador's denunciation reposted word for word by a 2026-registered account by hour two. Engagement farming, not coordination — and it ran in both directions.

📡The Arabic relay

A semi-automated Arabic-language news account — North African app-store registration, Washington DC display location — relayed the story faithfully twice within the hour (min 59, 60). Posts US politics around the clock in bursts; classified a cyborg relay, operator unknown. Its two posts drew 30 impressions combined.

🤖One AI-persona quote

An account styled as an AI commentator posted a 2,000-character denunciation at min 48 that reads as generated text. It reached 10 impressions.

🌐The “foreign reposts” claim, tested

A hostile-lane account claimed the top reposts are mostly foreign. Re-tested (300-retweet overnight sample, 98 hydrated): it is ~60% non-US — what organic time zones produce while America sleeps — with zero suspensions, no creation batching, no shared templates; 5–10% show individual signals with negligible reach. Descriptively half-true; the coordinated-amplification inference is unsupported. Negative finding, both windows.

§ 14 · Intelligence Gaps

Where this analysis is blind

Every number in this report comes from keyword-bearing text posts that commercial listening tools can see. That makes all of it the tip of the iceberg, and the honest version of the file says exactly where the iceberg is. Each card is a gap, why it exists, and what would close it.

Surface onlyFacebookMeasured by one number

I hold the view counter on the mayor's own cross-post (11M) and 71 keyword posts. The comment threads, shares, and group discussion behind that number are unmeasured: Meta licenses none of it and the page blocks automated collection. Closing it takes page-owner analytics or a logged-in browser harvest.

Not collectedTikTokAbsent entirely

The claim travels there as speech inside video, which caption-keyword capture cannot see, so a platform where this story plausibly reached millions contributes zero rows. Closing it takes a dedicated hashtag/sound pull plus speech-to-text on the top videos.

One itemRedditThread discussion unmeasured

The keyword capture caught a single message-board item. The actual conversation, in city and politics subreddits, never enters the corpus. Closing it takes a dedicated subreddit pull; the tooling exists, it simply was not in this collection.

StructuralSpeech inside video & audioThe blind spot that shapes everything

Text-keyword listening cannot penetrate video or audio: YouTube commentary, streams, and podcasts discussing the video without typing the keywords are invisible, which is why YouTube shows only ~340 items on a hundred-million-view story. Closing it takes transcript-level collection, a different toolchain from social listening.

Dark by designClosed channelsTelegram, WhatsApp, Discord, DMs

Invisible to all commercial collection, and precisely where narratives harden after the public phase. No lawful third-party tool closes this; it stays labeled dark rather than quietly omitted.

Ranked surface onlyInstagram comments180 of ≈159K

The labeled sample is the like-ranked surface an ordinary viewer sees, captured the first night; the reel now displays 159K comments and Meta's licensing wall blocks a deeper third-party pull — a July 22 re-attempt returned a single comment (§08). Closing it takes a logged-in harvest of the ranked thread at depth.

Still watching from the resolved indicator set: the documented Russian-aligned aggregation network that harvested the concession-plus-othering frame appears in both captures; its amplification remains unmeasured (§06).

§ 15 · The Takeaway

What moved and what did not

The structure of this event was set within two hours and held for fifteen — after the first night, only the size of the carriers changed, never the shape of the fight.

  • The verbatim text outperformed every spin of it, start to finish, and each faction's compression was legible against the primary source within minutes.
  • No referee ever arrived: by morning the escalation lanes had climbed to elected office and the agencies had settled on the duality headline — with zero community notes, fact-checks, or platform action on a two-hundred-million-view claim.
  • The bot layer was present, tiny, and irrelevant. What moved this claim was authentic audiences on four differently-ranked surfaces, plus a four-day institutional run-up (warrant, presidential rebuttal, ambassador's dismissal) that had loaded the stage.
  • By ~11 AM ET the mayor's accounts had posted no follow-up. The story no longer needed him.
Mechanics beat messaging

The same video read as consensus on Instagram and as a brawl on X, because one platform ranks dissent out of sight and the other hands it standalone distribution. Before you read a comment section as public opinion, ask what the ranking mechanic is hiding.

Verbatim outruns spin

The single most-viewed post in the conversation was a reporter's word-for-word transcription, roughly five times the most-viewed hostile frame. When the raw text leads the reach table, the narrative war is over compression, not invention.

Count lanes, not sentiment

Machine sentiment graded 86% of this sample maximally negative, including plainly supportive posts. Stance is a judgment about the poster's position, not the text's valence; it has to be labeled, not scored.

★ Before you believe or share a viral claim

Check the source: who is it, how old is the account, how accountable is it, who funds it. Check the original: what did the document or footage actually say, and with what caveats. Check who benefits from the loudest version, and what got added or stripped along the way. If a post makes you feel certain and furious within five seconds, that is exactly the moment to slow down.

★ Reach out, I read everything

I am Travis Hawley. If you want a claim, a cascade, or a narrative traced the way this one was, find me at shadowgraphintel.com/contact, @talk2trav on X, and @talk2trav on Instagram. And scrutinize whatever you read next, including me.

§ 16 · Coverage & Known Gaps

The tip of the iceberg

Every figure in this report is a floor, not a census. The honest version of “cross-platform” is this table: what I hold, what exists, and why the difference exists. Channels are graded on what this investigation could actually observe; the narrative version of these gaps is §14. Holdings as of each row's latest July 22 check.

ChannelWhat I holdCoverage gradeWhy the gap
X — anchor + engagementAnchor metrics across five collections (last: a later July 22 in-situ re-check), 466+98 hydrated engager profiles, 392 labeled quotes, a 336-quote refresh, 300 replies screenedGood, sampledFast-mode pools cap at 300; full exports available if needed
Bluesky9,477 keyword-matched posts in the latest export sample; the mayor's native anchor captured in situ (86K likes at the day-two check)Good for volumeNo authenticity/profile pass run on Bluesky accounts
Instagram — anchorFull post metrics, caption track, live re-pullGood
Instagram — comments180 of the first night's 21,590 (<1%), the like-ranked visible surface; the reel now shows ≈159KSparse, surface-onlyMeta licenses no third-party comment data; only the ranked surface is collectable. A logged-in browser harvest could deepen this to several thousand
Instagram — wider postsNone in the cross-channel captureNot collectedSame licensing wall; keyword capture excludes Instagram
FacebookCross-post captured in situ (11M views at the day-two check); 71 keyword postsSeverely under-collectedMeta licensing; page blocks automated fetching outright. Comment/share data unobtainable without page-owner access
TikTokNone in this captureUnknown volumeThe claim travels as speech inside video; caption-keyword capture cannot see it. A dedicated TikTok pull is possible but was not run
RedditOne message-board itemNot meaningfully collectedNot in this keyword capture; a dedicated pull is possible
YouTube~340 keyword itemsBlind spot — true volume unknownSame speech-inside-video blindness as TikTok; the low count measures the lens, not the platform
Online news~1,800 headlined items + a verified sweep of major outletsModerateAggregator-heavy capture; syndication chains visible and noted
Truth Social7 items (first night)Blind spot — true volume unknownVendor coverage is near-zero for the platform; the low count measures the lens, not the platform
Telegram, WhatsApp, Discord, private groupsNothingDarkClosed channels; invisible to all commercial collection

Swipe sideways to see coverage grades and gaps →

Sampling notes: the labeled quote sample covers 184 of the first 110 minutes plus the largest hour-two entrants; stance and narrative labels are mine; reply-duplication screening covered 300 X replies; counts should be read as minimums and shares as estimates from labeled samples.

§ 17 · Scope, Method & Caveats

What this is, what it is not, and how it was done

◆ Note on scope

This report traces how a claim moved. It does not rule on the underlying claim. Whether Benjamin Netanyahu is a war criminal is a question before international courts and contested by governments; nothing here adjudicates it. Two contested axes run through this story and I keep them separate: (1) the characterization, an ICC arrest warrant is an allegation and not a conviction, and the video's critics and defenders fight over exactly that gap; (2) the underlying figures, the video's casualty count of “more than 73,000” traces to Gaza health authorities, which critics dispute and supporters treat as authoritative. Both disputes are part of the propagation story. Neither is resolved here.

◆ Method, in brief

This report covers the video's first ~16 hours across X, Instagram, Bluesky, Facebook, and the news record, collected July 21–22, 2026. Platform-displayed counts were captured in situ and treated as floors. Quote posts and comments were hand-labeled for stance; account authenticity was tested twice on my own collection. Every published media claim was checked against at least two independent outlets before inclusion — which is why two widely recirculated “reactions,” the President's arrest vow and the Israeli PM's office's “kangaroo court” statement, are excluded: both pre-date the video. Sample sizes, refresh times, and per-channel coverage grades are in §16; collection blind spots are in §14.

★ Standard disclaimer

Evidence-based assessment of publicly observable data — correlations and patterns. Certain claims remain unverified and this is not a determination of wrongdoing.