Two scripts demanding removal share the peak priority score, and 23 of the 50 items coded as calls for violence or forced removal came from 11 accounts on X

The window runs from Sunday 18 October 2026 to Saturday 24 October 2026, 00:00 to 23:59 UTC, 7 days. It lands on Sunday 25 October. The record’s last collection closed at 01:08 UTC on Sunday 25 October, after the window ended.

This is a purposive, outcome-selected monitoring sample framed by a 1,034-account roster spanning 23 ideological categories. The roster is the frame we track, and the collection combines the X timelines of the roster’s handles, the Bluesky feeds we follow with a network search beside them, two 4chan boards, /pol/ and /g/, and a news table. On X the collector reads a nightly pull list and a share of the roster handles each night in turn, and before the record closed a top-up read 46 X handles relevant to this instrument that had gone unread for 36 hours. The news table holds 430 headlines in the window. A beat of American campaign and Congress headlines, collected after the record closed, informs the event descriptions on the weather board and sits outside the ledger’s counts. Roster and board collection is systematic within those limits; off-roster discovery is opportunistic and incomplete, and this week it consists of four items quoted in published reports and the two Bluesky rows the ledger shows without the account. The hunts read 7,132 of the 15,355 posts and headlines collected: every post the ranking classifier scored 0.3 or above or could not score, every headline, every post by an account with 5 or more confirmed items in earlier weeks, anonymous board posts it scored 0.15 or above, and random samples of the rest. The hunts read a random 412 (25 percent) of the 1,618 posts that were anonymous board posts the classifier scored below 0.15 or came from accounts with 1 to 4 confirmed items in earlier weeks; the 412 held 2 hits, which puts the hits among the other 1,206 at about 6 (95 percent interval 0 to 26, derived). The hunts read a random 376 (5 percent) of the 7,393 posts from accounts with no confirmed item in earlier weeks; the 376 held no hit, and over this one week that puts the hits among this week’s other 7,017 at 73 or fewer (the upper end of a 95 percent interval, derived). A hit is language about a group defined by who its members are, or about a political camp as a class of people, said of the group as a whole in the poster’s own voice. Each confirmed item is coded at the most severe of six rungs, and the rungs follow the written rung definitions of our methodology. Biased states a negative trait of the group that stops short of danger or lesser worth. Stigmatizing presents the group as a danger or a problem to be managed, including immigration framed as invasion or as degradation by contact. Hateful treats the group as lesser or as not belonging, including the denial that its members belong to the nation. Threatening warns the group that harm will come to it, or demands that it be silent, hide or leave, without itself asking for the harm. Glorifying violence celebrates harm already done to the group or praises the people who did it. Calling for violence asks for harm to come to the group or sets conditions for it: violence, killing, expulsion by force, or a “solution” to its presence. On the expulsion line, a named program of compulsory removal by origin, such as remigration or the mass deportation of a named people, sits at calling for violence; a demand or warning that a group leave or be deported, with no program of compulsory removal by origin stated, sits at threatening; and a mass-deportation demand with no origin in its words removes people by status and sits below the top rung. A coder may supply the step that makes an item hostile to a group when the words alone do not carry it, and records what was supplied and from what context. Every post the ranking classifier rated 0.9 or above was given an explicit decision, and every decline among them was checked again. A verifier tests every hit. A drift gauge re-codes a sample of the confirmed ledger blind each week, and its readings follow below. This instrument states whether an item cleared our harm bar. It makes no legal finding in any jurisdiction, and the countries on this ledger do not share one standard. Every hit rests on a dated source read in full at collection. Each quote is verbatim from its source, or from our collection record where the source has gone offline, as the board threads have. Language of these kinds does not clear our harm bar: claims about a class defined by conduct; the word Islamist used as a political label; institutions as targets; criticism of named individuals over their own acts, including a defendant who is a private individual and never a proxy for a group; captions whose target sits in a video the hunts could not view; stock partisan idiom; and bare statements of dislike. The ledger holds 627 items across 627 distinct urls, one item per url. Jerm holds final editorial responsibility. This week 19 items cleared the extremism branch of our gate, which covers a political camp cast as subhuman, as a criminal class, or as a body to be removed, imprisoned or destroyed, among them Cernovich’s “revolting, anti-white WORMS” on conservative commentators, JackPosobiec’s announcement of a new “UNHUMANS 2” chapter on the political left, AuronMacintyre’s “Like all demons, they would rather rule in hell than serve in heaven” on the left, hasanthehun’s and OccupyDemocrats’s posts casting Republicans as child abusers, board posts casting MAGA supporters and Democrats as child abusers or their protectors, LauraLoomer’s post that “Democrats think it’s ok to murder people you disagree with”, jacksonhinkle’s and sneako’s posts casting Zionists and supporters of the Israeli military as genocidal as a class, and two board calls for violence against left-wing protesters and right-wing men. Every other confirmed item concerns a group defined by who its members are, as our methodology lists them. Items from other directions are coded under the same standard: board threads casting white people as a disgusting race, men as the lesser sex, British people as rats and inbreds and American and Russian people as evil peoples, and a board call for the genocide of white men for how they vote.

Each week a drift gauge re-codes a sample of the confirmed ledger blind, so a reader can see how far a fresh coder reproduces the published coding. This week two blind codings re-coded a sample of 120 confirmed items drawn across all six rungs from all 627 confirmed items. The two codings agreed with each other on the exact rung 89 percent of the time and within one rung 94 percent of the time. The first agreed with the published coding 91 percent of the time on the exact rung and 95 percent within one, and the second 86 and 91 percent. The published coding places about half the sample at hateful or above, and so does each blind coding. The first blind coding found no hit in 2 confirmed items and the second in 5. Items coded through a recorded inference reproduced about as well as items whose hostility sits in the words: on the 52 items whose coding records an inference, the first coding matched the published rung 90 percent of the time and came within one rung 94 percent of the time, and on the 68 items whose hostility sits in the words, 91 and 96 percent. A step that rests on an image, a thread or a carrier’s earlier posts does not travel with the text alone, and 13 of the 23 calls from accounts on X rest on such a step.

This week's collection reached part of the 1,034-account roster

The roster is the frame we track, and the collection combines the handles on our lists, the two boards, a Bluesky network search and a news table. Every number here is computed from the roster file, the handle lists and the collection database for the window of 2026-10-18 to 2026-10-24.

The roster holds 1,034 accounts across 23 categories, and 136 of them sit in more than one category, so the category column below counts more accounts than the roster holds. Handles collected counts the handles that stored a post by the collection that closed the record on 25 October, in both tables, and handles that returned a post and posts collected count the posts and headlines dated in the window that the same collection stored. A handle the collector read that returned nothing leaves no row. The Bluesky collector also runs a network search beyond the handles we track, so its collected count exceeds its list; the X collector reads only the handles listed.

The table below sorts the 46 of 152 X handles we can match to a roster category by name, and a handle on more than one category is counted in each. The other 106 are handles our records hold without a category, and they are collected and read the same way. 13 categories returned an X post this week.

The window is set in UTC. Posts stamped early on 18 October UTC were written on Saturday evening in North American time, and posts written on Saturday evening 24 October in North American time fall after the window.

Comparisons with last week set seven days beside six. The lines that compare a script with its chain a week earlier set two records side by side. This record covers seven days and last week’s covered six, so a count set beside last week’s compares a longer window, and where we compare volume we compare it per day. The number of scripts, items and calls in each depends on that week’s events, on its length and on how its items were grouped, and a difference between the two says nothing about the direction of the discourse. The two weeks also read different sets of X accounts: X holds posts from 121 roster accounts this week and from more last week. A narrative can be absent because it lost its carriers, because it did not clear the bar in this window or because our collection did not surface it, and these records cannot separate those.

The X record has gaps. The nightly pull list holds a fixed core of handles and the roster handles that appear in the past seven days of our news table; the collector also reads a share of the roster handles each night in turn, and before the record closed a top-up read 46 handles relevant to this instrument that had gone unread for 36 hours. X covers 121 distinct roster accounts with a post dated in the window, of the 152 roster X handles, and 31 carry none. Each read takes at most 200 posts per handle. The collector stopped at that page cap for PDChina on 19 October, XHNews on 22 October, globaltimesnews on 24 October and marklevinshow in the run that closed the record, so posts beyond the cap are absent and their counts in this record are floors. The collector’s runs of 19 and 20 October fell short, reaching 58 and 41 of the handles they were set to read, and the handles left over rolled to the next runs. A post deleted before its handle was read is absent. The X collection reads each handle’s own posts, quote posts among them, and leaves out its replies and reposts, so carriage that lives in reply threads is absent.

The boards and Bluesky are read in part. On 21 October the boards yielded about two fifths of the opening posts of a median weekday (derived), and a thread that has left the board cannot be collected again. A board is read from its catalog, so a thread that opened and died between two reads is missed, and the board items on this ledger are a lower bound on the qualifying threads those boards carried. Bluesky in this record is mostly news outlets, and its two items are a news publisher’s quotation of the President’s rally remarks and one private account’s post. Bluesky publishes no impression count, so the two add nothing to any reach figure.

The hunts read a gated share of the collection, and the rest is measured by sample. The hunts read a random 412 (25 percent) of the 1,618 posts that were anonymous board posts the classifier scored below 0.15 or came from accounts with 1 to 4 confirmed items in earlier weeks; the 412 held 2 hits, which puts the hits among the other 1,206 at about 6 (95 percent interval 0 to 26, derived). The hunts read a random 376 (5 percent) of the 7,393 posts from accounts with no confirmed item in earlier weeks; the 376 held no hit, and over this one week that puts the hits among this week’s other 7,017 at 73 or fewer (the upper end of a 95 percent interval, derived). Both figures are estimates from samples, and the posts they describe were not read. A verifier tested every hit, and every decline of a post the ranking classifier rated 0.9 or above was checked again; a hunter’s decline of a post rated below 0.9 had no second check, so the ledger can miss qualifying items among the posts the hunts read.

Impression counts are capture-time counters. X’s impression counts are capture-time counters, and they can rise after collection. Each figure is the counter as it stood at collection, and handles read by the nightly pulls and by the top-up were read at different ages after posting. A counter tallies displays of a post, so one reader can account for several displays, and a script’s audience can sit above or below its sum. The derived 44,603,196 figure sums the 51 scripts; the 46 items outside them are not in it, and board threads, Bluesky posts and the four report-borne items add nothing to it.

Reach reads X’s impression counts where X publishes them. The 4chan corpus has no view counts and no persistent handles, so board items add nothing to a script’s reach, and any comparison of captured impression counters on this page ranks X carriage only. The 20 scripts that no named carrier touched are scored on their four measurable factors and ranked in their own list, outside the peak, the median and the ranked table. Where a head of state’s remarks reach the record through published reports with no counter, reach reads the office and the national pickup beside the script’s measured X total. Every report-borne item rests on the report’s quotation and our record of it.

A few accounts carry much of the named ledger. LauraLoomer supplies more confirmed items than any other account, across 16 scripts, and RadioGenoa and TRobinsonNewEra the next most, across 12 and 10. LauraLoomer carries 12 of the 13 posts treating Muslim officeholders as an enemy inside the state and all 6 posts on Muslim voters stealing elections, RadioGenoa 7 of the 15 posts denying belonging by descent and 6 of the 7 blaming diversity for national ruin, and elonmusk’s one post more than half the measured impressions of the replacement and removal script (derived). PaulGolding, LauraLoomer and TheGloriousLion supply 13 of the 23 calls from accounts on X. Where one account supplies most of a script, the script’s volume and reach describe that account’s output, and a change in its posting, or in our collection of it, would move the script’s readings with it.

Coding is a judgment. The drift gauge sets out how far a fresh coder reproduces the published coding. Items coded through a recorded inference reproduced about as well as items whose hostility sits in the words this week, and 13 of the 23 calls from accounts on X rest on such an inference. The gauge’s 120 items come from all 627 confirmed items. Each item’s detail gives its coding confidence.

We name no private individual. Private individuals are never named on this page, and no quote that would identify one is published. Several arenas this week turned on private individuals: the man a board thread blamed for the fire at CFB Wainwright and the soldiers killed and injured there; the French student a post pictured with her family background; the jurors and the journalist witnesses at Tommy Robinson’s trial; the man charged over the Hackney assaults and the three women attacked; the relatives of Abdul El-Sayed, the donors and the town-hall questioner in the Michigan race, and his children; the woman filmed confronting looters in Milan and the young man a post accused of running a protest page; the relatives of Amy Acton; the woman arrested at the Utah County meeting and the deputy; the woman killed in Amsterdam, the man arrested, and the girl and her parents in Geert Wilders’s post; the relatives of Ilhan Omar; the suspect, the dead and the wounded in Trappes; the man, his family and the girl in the Bangor video; the man held over the damaged statue at Quantico; the boy and the store owner in the Florida case; the Cornell accuser and the accused men; the victims of the Bend school shooting, their families, the school staff and the community member who shot the attacker; the minors and escorts in the claims about the late Senator Lindsey Graham; the trainee doctor, the boys, the woman and the victim in the posts of 20 and 21 October on Scotland and Birmingham; the people board threads and TRobinsonNewEra’s posts asked readers to identify, and the woman one of his posts gave by full name; the man accused of informing the police; and the officer at the center of an online pile-on. Where an item concerns one of them, the ledger quotes the carrier’s words about the group, and none of our own text identifies the person. Rows from accounts that do not clear our identification bar, this week the two Bluesky rows, are shown without the account or the wording, since either would identify the account. Nine rows keep their quote and coding and are shown without a link, because the thread or post behind each identifies a private person.

Coordination is two readings that answer different questions. The coordination factor measures degree of organization on the 1 to 5 anchors and moves above 1 only on a documented link between accounts: a bridge account, phrasing copied across accounts, or a named hub. The coordination verdict asks what the link shows about how the accounts are organized, and convergence alone never reaches it. All 51 scripts carry no coordination identified and a coordination factor of 1: the record documents no bridge, no hub, no directive and no independent origin. On the invasion frame the Republican Jewish Coalition’s and the Senate Republican Conference’s accounts used one sentence frame, in 2 of its 19 posts; the record shows no event, hub or co-participant linking the two accounts and does not show which wrote it first or whether a shared source supplied it, so the shared wording reads as convergence. On an anonymous board, one actor and an organized handful cannot be told apart, and each convergence signal becomes a coordination check if a linking account appears.

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, and this week has no such sample.

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, the weighting our methodology applies to every script that holds a confirmed item, which is every script here, then take the 5.8th root. The 5.8 is the sum of the weights, which brings the result back onto the 1-to-5 scale. The trajectory multiplier, which scales a score by how fast a script’s volume moved, was not measured for any script this window, so each published score is its base score. Multiplication limits how far any one factor can lift a score. One override runs the other way: any script at H=5 takes a second-tier minimum regardless of its computed score. Our anchors set H=5 at a concrete, explicit call for violence against a defined target, at a concrete target with a place or a time and mobilization, or at violence already carried out, and every script that holds an item adjudicated at the calling-for-violence rung sits there. The minimum keeps a small account that asks for violence against a defined group from scoring its way into the bottom tier. Nine scripts reach H=5 this week, and 6 of them take their tier from that minimum. Tiers rank scripts only. The 46 items that match no script sit outside the ranking and carry no score and no tier; seven of them sit at the calling-for-violence rung, and as individual hits they carry no score. Tier edges: the top tier starts at 4.0, the second at 3.0, the third at 2.0. Scores are published to one decimal place, and the tier reads off the published figure. A score that lands within 0.15 of a tier edge is marked at the boundary, and the mark gives a single factor and the score a one-step move in it would give; 9 scripts are marked this week, and on each of them the step would carry the score across an edge. Reach is the one factor a board-only script cannot supply: an anonymous board publishes no impression count, so a script carried only by board threads is scored on its four measurable factors as (T·C·M1.3·H1.5)1/4.8, marked reach unmeasured, and ranked in a list of its own, away from the scripts whose reach was read. Coordination as a factor reaches the co-platforming and network anchors on a named hub or a confirmed bridge, and a script can score above 1 on the factor and still carry no coordination identified as its verdict; every script sits at 1 on it this week.