We monitored our 1,034-account roster, the anonymous boards we track, and our off-roster finds for the week of August 17 to 23, 2026, reaching 137 of the 152 X handles on the roster. 143 items crossed our line. Ninety-five of them run seventeen distinct scripts and the rest stand alone. Five scripts reached the second tier this week, and four of the five called for a group to be removed, expelled, or killed. The most severe came from a sitting official rather than a fringe account: Israel’s national security minister set a nightly kill quota for Gaza beyond anyone posing a threat and praised a newly built execution facility for Palestinian prisoners, the highest-scoring script on the board at 3.79. A sitting British politician promised an asylum seeker execution without trial; a high-reach account called for Muslim people to be rounded up and deported as an invasive species; and anonymous board threads told readers they could legally kill or maim Indian and Hindu men. Not one of the seventeen showed documented coordination: each was either a single account posting repeatedly or independent accounts converging on the same claim, with no organizing hub we could establish. The longest-running script needed no event at all, casting Jewish people as a single hidden actor whose will explains unrelated outcomes, twenty-six items across three platforms and the longest continuous chain in the record.
The record runs larger than last week: 143 items against 139, and 7,705,963 platform-reported impressions against last week’s 6.5 million. Part of the measured increase reflects wider collection: this window reached 137 of 152 roster handles against 127 last week, so week-over-week volume comparisons are footprint-limited before they are findings. The severity peak is a sitting official: Israel’s national security minister, three items scoring 3.79, just under the top tier. The largest script by volume is the one we can measure least, twenty-six items casting Jewish people as a single controlling actor, most of them anonymous board threads that report replies rather than views. One second-tier script rests almost entirely on a single account, Rupert Lowe; two third-tier scripts do the same, Laura Loomer’s campaign against Muslim officials and the ACTBrigitte anti-Muslim campaign. Each keeps its coordination factor at the floor, because one account posting daily is not a network converging.
Scripts by severity
Click a bar for the full scorecards.
The week's weather, day by day
One dot per real event. Filled dots drew a reaction we confirmed; hollow ones drew none we recorded. Click for the full board.
The 143 items, by harm level
Click for the full ledger.
On the radar
Click a date for the full calendar and the patterns building toward it.
01 · The scripts
Five factors are scored from 1 to 5: reach, toxicity, coordination, mainstreaming risk, and harm potential. Their composite Ember Priority Score is shown on a chart running from 0 to 5. Click one for its scorecard. Peak severity this week is 3.79, and no script reached the top tier. Five scripts sit in the second tier, nine in the third, and three at the bottom. One second-tier script sits there on the severity floor: an anonymous cluster carrying an explicit call for violence takes the floor its computed score would not otherwise reach. The other four clear the tier on their own scores. Several scripts were rescored on their full corpus after the collection passes described in Limitations, and each factor change is stated on the scorecard.
02 · The ledger
Click a row for the quote, the coding, and the source.
03 · The week's weather
The events we recorded no reaction to matter as much as the ones we did. Together they show which stories these narratives picked up this week and which drew no reaction our footprint recorded.
04 · On the radar
Five scripts continue a ruled lineage. The Jewish-collective script is the longest chain in the record, twenty-six items this week; the migrant-replacement script continues the Channel-invasion chain from two weeks back; the Loomer campaign against Muslim officials continues, now aimed at a named Senate candidate; the ACTBrigitte anti-Muslim campaign runs a second week; and the Islam-incompatible script continues, escalated this week to an explicit call for expulsion. Two proposed joins were refused at the grill on the different-predicate test, and the refusals are recorded beside the joins. Seven of last week’s scripts are not carded this week. We do not read that as decay: a narrative can be absent because it lost its carriers, because it did not clear the bar this week, or because this week’s collection did not surface it, and these records cannot separate those. Each line compares one narrative with itself a week earlier, joined on a key ruled when the narrative was carded rather than on any similarity between the two weeks’ text.
05 · Method
This is a purposive, outcome-selected monitoring sample drawn from a 1,034-account roster spanning 23 ideological categories; this week’s coverage limits are stated below. It is not a random slice of the internet. It misses anything posted off our roster, and anything a roster account did not post. A hit is language that collectivizes, dehumanizes, or calls for exclusion of a protected group, or that runs dehumanizing or eliminationist framing against a political camp. Criticism of governments, institutions, and named individuals over their own conduct stays protected, however sharp. Every hit rests on a dated source read in full at collection. Where the source was still live, a second, adversarial pass re-verified the quote against it; where it had already gone offline, which covers every 4chan thread this week, the second pass re-read our collection record and the coding rather than the live page, and the item says so. Items that failed re-verification are not in this report. Research and drafting run through Claude under this methodology; every published classification passes the documented verification chain, and Jerm holds final editorial responsibility. No item this week cleared only the extremism branch of our gate against a political or ideological target: every confirmed item names a protected class, or a group our methodology treats as one, and the nine violence-tier items all do.
Platform-reported impression counts are capture-time platform counters, not audience counts, and they can rise after we read them. Several items showed materially higher counts when re-verified against the live platform later in the week. Read each figure as the capture-time reading rather than an audience count.
Reach is not measurable everywhere. The 4chan corpus has no view counts and no persistent handles, so reply and image counts stand in for reach and almost certainly understate it. Every 4chan thread in this week’s set had already cycled off the live board by the time our verification pass ran, so those quotes rest on our collection records rather than a second live read.
The research cut ran Saturday evening. A Sunday pass screened 870 late-window posts from the week’s carriers and boards. It opened one new script, an anonymous anti-Indian cluster carrying an explicit call for violence, and folded fresh reach into eight existing scripts, including the anti-Muslim script, which the Sunday carriage pushed into the second tier. Two Sunday items were reviewed and refused at the grill on the protected-speech test. Affected scripts were reassessed against the final corpus; factor changes and unchanged readings are stated on the relevant scorecards. The X footprint needs the same honesty: the weekly budget kept nightly collection on a rotating core of about 33 roster handles. Expanded collection before publication reached 137 of the 152 handles we track for the full window, and handles beyond the nightly core again supplied a substantial share of the recorded items. About 15 low-priority handles were not reached, and one high-volume account truncated at the 200-post cap with its tail not recoverable. A narrative living on the handles we did not reach this week would read here as smaller than it is. The final hours of Sunday are uncovered, which is the standing cost of a Sunday-published weekly. Daily prefiltered shortlists were present for six of the seven days; the research also read slice files built directly from the database for the full date range, subject to the collection limits above.
Coordination scoring leans on shared semantic payload, a shared novel claim, coined term or fabricated detail, except where we can identify an organizing hub. No script had an identifiable organizing hub this week. Every one records either repeated carriage by one account or cross-account convergence we can see, without an organizing link we can establish, and those are different findings. Several scripts are closer to one account posting repeatedly than to many accounts converging, and their coordination factor sits at the floor accordingly. Nothing this week was coded as overt coordination.
The Bluesky lane produced one coded item from 5,373 screened posts. Its priority shortlist this week matched mostly benign news chatter, which says the lexicon layer matched weakly there rather than that the platform carried nothing; read the zero as a collection limit. X’s owner drew wide coverage for a post casting Hitler as a left-wing socialist; it drew no coded carriage in our roster, and that finding sits on the radar board.
No hit-rate or prevalence claim here may be read as a comparison across ideological lanes. That comparison requires a lane-tier sample for every lane compared, over a common platform footprint. We did not draw that sample this pass.
Five factors, each scored 1 to 5: reach (R), toxicity (T), coordination (C), mainstreaming risk (M), and harm potential (H). They combine as (R·T·C·M1.3·H1.5)1/5.8: multiply all five, with mainstreaming risk raised to the 1.3 power and harm potential to the 1.5 power because those two carry the most real-world weight, then take the 5.8th root. The 5.8 is the sum of the weights (1 + 1 + 1 + 1.3 + 1.5), which brings the result back onto the 1-to-5 scale. Multiplication limits how far any one factor can lift a score, without eliminating its influence. One override runs the other way: any script whose harm potential reaches 5 takes a P2 floor regardless of its computed score, because eliminationist content with no measured reach is still eliminationist content. One script carries that floor this week, and the chart marks it. P1 starts at 4.0, P2 at 3.0, P3 at 2.0. Coordination scores above the floor only on real signals: a named hub, near-duplicate phrasing, synchronized timing, or a confirmed bridge between actors.