We scored 23 camp-versus-camp narratives across 15 fault lines for the week of September 7 to 13, 2026. Three sit at the top watch tier, two of them tied at 4.37: the two Michigan Senate frames that carried the 9/11 anniversary, with the reckless-war frame behind them. None was fully supported by our checks: five failed at least one check outright and eighteen split. Six narratives read as ignited: four reached ignited for the first time on their chains this week, the demonization-defense frame and the Sánchez frame, whose lineages crossed this week, the Mamdani ceremony-clip frame, ignited in its first window, and the Netflix neck-injury frame, ignited on a single cleared post; two continue, the El-Sayed association frame in its third ignited week and the Clancy sympathy-as-endorsement frame in its second. The anniversary and the Clancy juror interviews handed the harm-clearing items to the other instrument. The trajectory instrument applied one reading, which moved a tier, and held four under its floor. Everything measured here is protected political speech, with one exception this week: items inside six narratives cleared the harm bar, were routed to the harm instrument, and are reported only there. What remains on these cards did not clear our harm gate.
Everything on this dashboard describes legitimate political expression, with one exception this week: items inside six narratives cleared the harm bar, were routed to the harm instrument, and are reported only there, and what remains on these cards did not clear our harm gate. A narrative appearing here is not an accusation of hate, extremism, or bad conduct; it is a description of a framing pattern. Our separate harm methodology, with its own much higher bar, is reported in The Weather Report.
Our weekly harm report measures the narrow band of content that collectivizes, dehumanizes, or incites against groups of people. Most of the culture war never gets near that bar. This report measures that larger remainder: same week, same verification standard, different instrument.
We track framing contests. An event happens, and within hours commentators on each side stop arguing about the event and start arguing about what it reveals about the other camp's motives and character. We state each narrative's claim in a testable form, check how the predictions it generates fare against what is observable, look for the mirror narrative on the other side, and score the whole contest for reach, intensity, penetration, elite adoption, and mirror lock. We also watch each narrative for seven specific markers that it is drifting toward the territory our harm methodology covers.
Three narratives reached the top tier this week among the cards we scored, two of them tied at 4.37. The El-Sayed association frame, in which the Rogers campaign and its super PAC carry Hasan Piker’s 2019 remark into the 25th anniversary as proof that the candidate gives comfort to the hatred behind the attacks, rose from 4.18 on the nominee’s convention speech and the Great Lakes HQ television ad, and reads as ignited for a third week. The demonization-defense frame, the pro-Israel and Republican reading of El-Sayed’s refusal to drop Piker as sharing that hatred, held at 4.37 on the same carriage and reads as ignited for the first time. A new British card sits just below them at 3.95: Rupert Lowe, Tommy Robinson and Lee Anderson recasting the national mask order and the policing of Dover and Portsmouth as a two-tier state that protects migrants, the left and the burqa while it targets patriots, answered at official tier by No 10’s ‘thuggish’ label. The reckless-war frame, carried by Warren, Jeffries, Murphy and Khanna as crude passed $100 and the president promised an end to the Iran war right after the midterms, held at 4.13. Thirteen narratives sit in the active tier and six at W3, none at W4, with a median of 3.73 across all twenty-three against 3.95 across eighteen a week earlier. Eighteen play out in the US arena, two in Canada, one in the UK, one in Spain and one across borders. Eleven cards sit at the top elite-adoption anchor, among them a Senate nominee’s convention speech, the House minority leader on the war, the Treasury secretary and the president on the data-center backlash, the Senate party organ’s open-borders frame in its sixth window, the HHS secretary from the convention stage, and two members of the Commons on the mask order. That is a fact about the narratives we scored rather than a measurement of the whole discourse. The public-health lane carded three narratives, its first since W35, from the Allison vaccine-injury hearings in Ottawa and Kennedy’s convention speech.
The predictions are our operationalizations of each frame. None came back supported this week: every one of the twenty-three narratives either failed a check outright or split, with no frame whose predictions all resolved in its favor. Five came back contradicted, each against a specific prediction stated on its card. The Senate party organ’s claim that a socialist wing’s positions are the Democratic Party’s governing program failed for a sixth consecutive window, its Canada preview naming no Democratic bill, plank or pledge. The Sánchez frame failed in its second window, with the verdict scoped to the policy description we drew from it: a government flooding Europe would keep rather than return the bulk of the Ceuta arrivals, and the great majority were returned on either side’s count, while the Red Cross and the NGOs framed as accomplices do relief inside an enclave the army guards; motive and criminality are not observable to us, and the verdict says nothing about them. The claim that the data-center backlash is a Chinese and far-left manufacture failed in its third window, against a bus tour chaired by an RNC committeewoman and a Treasury secretary who conceded the grievance in the same remarks that called it agitprop. The reading of four seconds of Mamdani and Ocasio-Cortez at Ground Zero as mockery of the dead failed against the account of a Republican security chief who stood beside them and saw officials of both parties joking behind him, none of whom drew the charge. And the reading of Netflix UK’s ‘neck injury’ caption as a deliberate joke about Kirk’s killing failed where the images show the craning look the idiom describes and no left-coded account in our collection used the phrase in-window. Each card carries the prediction, what we observed, and the source. The remaining eighteen came back mixed because our checks reached a frame’s minimal reading and never its strongest. That is a statement about what we could check rather than about the claims. Every one of the eleven continuing narratives was re-tested this week, each lane re-running its predictions against the window’s record, and every one of those verdicts was re-read this week and unchanged: none moved on its own chain.
Eight narratives record a locked mirror this week, four pairs, every one inside the card set. Two come from the Michigan Senate fight on the anniversary: the El-Sayed association frame against the left carriers’ frame that the right does not care about 9/11 and uses it only to spread hatred, and the demonization-defense frame against Piker’s reading that the right exploits 9/11 for wars, surveillance and racial agitation, carded on its own this week so both sides of the pair carry a verdict. One comes from the Kirk anniversary: the death-cult frame, that the memes prove the left celebrates his murder, against the convention-shout frame, that the right loves calling for assassinations and buries the evidence of its own violence. One comes from Jacob Coxon’s resignation: the funded-doomer-op reading against the reading that the AI industry’s oligarchs are buying Congress. The two-tier card records an official-tier answer in place of a lock: No 10’s ‘thuggish’ and the home secretary’s ‘malign intent’ attach to documented assaults on officers and doxxing of volunteers, the gate passes documentation, and a government frame we did not card cannot make a locked pair. Eleven of last week’s narratives persisted under ruled lineage joins and twelve opened new lineages; seven lineages took no card this week, and the record says only that they did not clear the bar or were not surfaced, which it cannot separate. Four moved tier on their own chains: the serve-Israel frame and the Sánchez frame down from W1 into W2, the WNBA panic-as-pretext frame up from W4 to W3, and the Woke Right cabal frame from W2 to W3 on the trajectory multiplier. Rulings on the record: the Kirk anniversary pair stays as two cards, each with its verdict; Piker’s counter-frame stands as its own card; the tech-oligarchs frame opens a new lineage on the Coxon trigger instead of joining last week’s reading of the president’s ‘backwards and poor’ post; the association frame and the demonization-defense frame both stand, with their shared RJC and Loomer carriage noted on each; the Netflix caption card stands on its own beside the death-cult card; Sacks and Bernier sit at the official tier; and a post may carry two narratives when it makes two claims, which is how Occupy Democrats and James Lindsay sit on more than one card each. The lineage record, including the joins we refused, is at hobocode.net/lineage.
The trajectory instrument applied one reading this week and held four. Its measure is a corpus-share rate of change. Volume is matches per thousand collected posts in each window, and the two corpora differ in size, 14,301 posts in the prior window and 11,731 in this one, so a share moves differently from the raw count behind it. The change is a percentage change: this window’s share divided by the prior window’s share, minus one, read against the median change across the week’s cards; the baseline recorded this week is zero, so each card’s percentage change stands as its own reading against the field. The bands are set in advance and recorded as tunable: a change above plus 15 percent reads accelerating and carries 1.3, a change below minus 15 percent reads decaying and carries 0.6, and anything between reads stable and carries 1.0. One multiplier was applied, on the only card where both windows clear the five-post floor. The Woke Right cabal frame matched 23 posts in the prior window of 14,301, 1.61 per thousand, and 9 in this window of 11,731, 0.77 per thousand; 0.77 divided by 1.61 is 0.48, and 0.48 minus one is a change of minus 52 percent, below the minus 15 band, a reading of 0.6 that took its score from 3.52 to 2.11 and its tier from W2 to W3, in the same week its ignition status decayed from near ignition to warming. Four readings were held at 1.0 by ruling, and each of those cards shows the reading it holds unapplied: every one is a first appearance whose prior-window count sits under the floor, and every one read accelerating at 1.3 with confidence marked low. The death-cult frame went from 4 posts to 10, which is 0.28 to 0.85 per thousand and a change of plus 205 percent; the tech-oligarchs frame from 3 to 6, 0.21 to 0.51 and plus 144 percent; the Allison inquiry frame from 2 to 34, 0.14 to 2.90 and plus 1,972 percent; and the Bernier frame from 1 to 27, 0.07 to 2.30 and plus 3,192 percent. Applied, the first two would have crossed into W1 on the multiplier alone, and the rule declines to let a pair of windows that starts on a handful of posts do that. The remaining eighteen carry 1.0 as a default, with either no prior-window base or too few posts in both windows to read. The applied reading says that the Woke Right fight lost half its share of a corpus that itself shrank; the held readings say that two Canadian hearings and an anniversary went from almost nothing to something, which a first window cannot tell from noise. The five-post floor is a chosen threshold below which a count is treated as noise, a standing rule under review rather than a validated cut, recorded as tunable with a date.
Our harm methodology’s gates run first on every item, and this week items inside six carded narratives cleared them. Six narratives read as ignited: four reached ignited for the first time on their chains this week, the demonization-defense frame and the Sánchez frame, whose lineages crossed this week, the Mamdani ceremony-clip frame, ignited in its first window, and the Netflix neck-injury frame, ignited on a single cleared post; two continue, the El-Sayed association frame in its third ignited week and the Clancy sympathy-as-endorsement frame in its second. On the El-Sayed association card, five Laura Loomer posts in the narrative’s arena cleared the bar as the Michigan Senate fight ran into the anniversary, and are scored on the harm side under the handoff references w37-x-unassigned-007 and 043 and w37-x-right-lanes-080, 082 and 092, four of them in the Muslim-official-disloyalty family, f3, and the ban demand in the Islam-eliminationist family, f4. The demonization-defense card lists the four of those items that name El-Sayed as a shared handoff and reads as ignited on them, under the handoff rule as written, by ruling. On the Mamdani ceremony-clip card, ignited in its first week, four items cleared the bar, w37-x-right-lanes-090 and 103 and w37-x-unassigned-063 and 086, filed across the Muslim-official-disloyalty, political-outgroup-dehumanization, Muslim-collective-guilt and descent-belonging-denial families. On the sympathy-as-endorsement card, ignited for a second week, Matt Walsh’s post on the Clancy jurors cleared as w37-x-right-lanes-110 in the women’s collective-guilt family, f17, with a second Walsh post scored beside it and a MyronGainesX post in the women’s civic-incapacity family, f22. On the Sánchez card, RadioGenoa’s post on the Ceuta shantytown fires, which blames the prime minister in its final clause, cleared as w37-x-right-lanes-049 in the migrant-force-celebration family, f10, and the card moved from near ignition to ignited under the handoff rule, with the alternative reading recorded on the card for the review gate. The routing rule is one rule: an item that clears the harm bar is scored and reported only on the harm side; it still sets the ignition status of the narrative it came from, recorded on that card as its handoff reference; and it never counts here as a below-bar example or as evidence for the charge priority score. The card keeps the framing contest and the below-bar remainder, and the harm instrument keeps the harm.
Two narratives sit at near ignition, both new this week and both from the anniversaries. The death-cult frame reached it on three markers: a generic ‘they’ standing in for the meme-makers, the motive hardened into a standing trait of the left, and guilt spread to the Democrats who did not condemn the memes. Piker’s frame that the right does not care about 9/11 reached it on three: a collectivized ‘the right’ in place of Rogers and his super PAC, a trait recode, and a mirror lock tightening on the date. Fourteen narratives sit at warming and two at cold, Kennedy’s silenced-mothers frame and the Bernier frame, with no marker observed in either carrier set. Two decayed from near ignition to warming under the decay rule with their earlier marker instances retained: the Woke Right cabal frame, where none of the carded URLs sits in the confirmed harm corpus even as fresh instances of three markers were recorded and count toward next week’s read, and the China-manufacture frame, where last week’s trait recode did not recur. The WNBA panic-as-pretext frame stays at warming with no new instance. Each marker instance is logged on the card that carries it with its own source and date, and clicking a marker opens that card.
Routing also ran in the other direction. Candidates that cleared the harm gates were routed to The Weather Report, which carries 367 unique posts across 47 scripts this window with nothing at its top tier, and none of them is carded here. Several adjacencies were screened there and sit on no card. Roughly thirty of Loomer’s other in-window posts sit in the confirmed harm corpus, and none is cited on the Woke Right card that carries her ‘Woke Reich’ line. ACTBrigitte’s ‘Communists are merging with Islamists’ sits in that corpus and is cited nowhere on the open-borders card. Rep. Fine’s ‘Jihad Mamdani’, in the same post as his carded line, was flagged for the harm side’s screening and is cited nowhere as evidence, and Loomer’s ‘Islamic communist’ label on Piker was left uncited on the death-cult card and sent to the same screening. Walsh’s ‘demonic leftists’ post on the Netflix caption cleared the harm bar as w37-gap-12 and is a confirmed item in the political-outgroup-dehumanization family, f23, reported on the harm side at 1.1 million impressions; on this side it is credited to no marker and stands as no example.
Near ignition and ignited are review statuses. They escalate our checking cadence and record where an item crossed to the other instrument, and they carry no verdict about anyone. They do not make anyone on these cards an extremist, and we ask readers to leave that use aside as well.
Twenty-one of the 23 narratives drew a response we could find: thirteen record a mirror that answered, and eight record a locked pair. Two ran unanswered: the Netflix caption frame, where the company’s ‘ludicrous’ statement argued the merits and no left-coded carrier framed the critics’ motives, and the Allison inquiry frame, where every in-window carrier of the hearings in our footprint sat on the health-freedom side. Answered and locked are distinct statuses, and the three counts together account for all twenty-three cards. A lock means each side runs on the other’s statements as its primary evidence, and the two anniversaries supplied six of the eight locked cards. Read carefully, the answered count is a finding about this instrument as much as about the discourse: what our gate declines to code is the form the response took, most often merits argument, documentation or reporting rather than a counter-imputation. The Sánchez frame is the cleanest case this week, answered by a government publishing its own intelligence record and arguing scale, and the Clancy frame the next, answered by six newsrooms carrying the foreperson’s account and the holdout’s court record as news while the sharpest counter-imputation sat below the bar.
A polarization monitor that codes one side more than the other needs to say so in the same place it reports its findings. The camp label is derived from each card's carrier set. The label is coarse. The carrier-set paragraph on each card remains the authoritative account of who carried a narrative.
Each card carries its own camp label and the full distribution is above. We do not roll those labels up into a left and right count here, because the labels are written per card rather than drawn from a fixed set, and any two-way split of them would be our judgment rather than a measurement. Reading them, the mix runs in both directions: the Michigan Senate fight, the Kirk anniversary and the Coxon resignation each produced carded narratives on both sides, two cards are fights inside the right, over Carlson in the US and over Poilievre’s Covid record in Canada, and the per-card camp labels are listed above for any reader who wants to count the rest. We do not present that reading as a measurement of where polarizing framing happens. Our handle list is weighted toward right-of-center accounts on X, and an engagement floor sits under every lane, so any balance we see has passed through both. Where the opposing frame existed and we could not card it, the card says so.
7 narratives we carded last week 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. These records cannot separate those, so we report the absence and stop there.
| Narrative | Continues | Tier |
|---|---|---|
| The Woke Right cabal frame hardens into 'an Islamic movement' as Carlson defends El-Sayed on 9/11 | Vance's refusal to disavow Tucker Carlson recast as proof of a 'Woke Right' cabal inside the conservative movement | W2 to W3 |
| The WNBA panic as anti-trans stalking horse, third week: Mother Jones names a profiteer and Touré calls the sport issue a dodge | The WNBA panic as anti-trans stalking horse, second week | W4 to W3 |
| The Rogers campaign and its super PAC carry the Piker association into the 9/11 anniversary as proof that El-Sayed gives comfort to the hatred behind the attacks | AIPAC and allied carriers, now joined by a sitting Democratic senator, convert El-Sayed's continued embrace of Piker into a claim that they share one vision | no change |
| Piker and allied left carriers recast the Rogers anniversary attack and the Mamdani ceremony fight as proof the right does not care about 9/11 and uses it only to spread hatred | Left media critics recast the Fetterman and CBS attacks on El-Sayed as proof the party and press are bigoted, mendacious, and attacking their own candidate | no change |
| Greenwald, Piker, and Blumenthal read the state-level threats against the UK and Jeffries' Moskowitz pick as US politicians serving Israel over their own constituents | Left-populist carriers widen the serve-Israel frame from the Piker condemners to Fetterman, Vance, and both parties in Congress | W1 to W2 |
| Rogers, a pro-Rogers super PAC, and the RJC recast El-Sayed's refusal to drop Piker as sharing the hatred behind 9/11 on the attack's 25th anniversary | AIPAC and Sen. Fetterman recast El-Sayed's refusal to drop Piker as proof he shares a bloc's hostility to America, Jewish Americans, and the alliance | no change |
| Warren, Jeffries, Murphy, and Khanna carry the reckless-war frame as crude tops $100, the Saudi bypass shuts, and Trump promises an end right after the midterms | House and Senate leaders carry the reckless-war frame as CENTCOM resumes strikes and Vance says it is not a war | no change |
| The juror interviews recast as a confession that a female and left-coded camp rigged the trial, then as a leftist smear of a Black holdout | The mistrial holdout recast as one sane man against a female camp that excuses child-killing | no change |
| RadioGenoa, second window: Sánchez and the socialists as criminals flooding Europe, widened to the Red Cross and the NGOs in Ceuta as enemies, against the declassified record | RadioGenoa and Vox: Sánchez recast as a criminal flooding Spain and Europe with migrants, against his own frame of an ultra-right international using Ceuta to divide Europe | W1 to W2 |
| SenateGOP/SenJohnBarrasso: 'Democrat socialists' and a Canada preview around the Dallas convention, with the ICE predicate reduced to one post, sixth consecutive window | SenateGOP/SenJohnBarrasso: a socialist spokesman's wish list recast as the Democratic Party's 'explicit goals' to abolish ICE and borders, fifth consecutive window | no change |
| The data-center backlash as a Chinese and far-left manufacture, restated by the Treasury secretary as Chinese agitprop, by the president as a China PR thing, and syndicated to forty local stations | The data-center backlash as a Chinese and far-left manufacture, now with a named domestic network | no change |
Logic verdicts rarely move between weeks. Rogers, a pro-Rogers super PAC, and the RJC recast El-Sayed's refusal to drop Piker as sharing the hatred behind 9/11 on the attack's 25th anniversary crossed into the harm instrument. RadioGenoa, second window: Sánchez and the socialists as criminals flooding Europe, widened to the Red Cross and the NGOs in Ceuta as enemies, against the declassified record crossed into the harm instrument.
Each line is one narrative compared with itself a week earlier, joined on a key assigned when the narrative was first carded rather than on any similarity between the two weeks' text. The two instruments are reported separately and never combined into a single direction, because they measure different things on different scales.
Every item first runs through our harm methodology's gates. Items that clear them are reported in The Weather Report, under the routing rule stated in the ignition section above. Items that stay protected run through a two-prong polarization gate: the commentary's primary object must be the motives or character of the camp being framed, which may be the carrier's own, rather than the event's merits, and at least one of eight named polarizing operations must be present. Heat alone never codes, and documented, sourced motive claims never code.
Each narrative's charge priority score is CPS = (R·I·P·E·L)^(1/5): Reach, Intensity, Penetration, Elite adoption, and Lock, each anchored 1 to 5, combined by geometric mean so no single factor can dominate. The resulting score is plotted on a 0 to 5 axis. Watch tiers: W1 at 4.0 and above, W2 at 3.0, W3 at 2.0, W4 below.
The ignition instrument is separate from the score and never feeds it. Seven markers, each keyed to a construct our harm methodology already defines, produce a status: cold, warming, near ignition, or ignited. Ignited means an item actually cleared the harm bar and is reported there.
The logic test states each frame's claim as a prediction-generating hypothesis and checks the predictions against what is observable, with sources. The predictions are our own operationalizations of the frame, and reasonable readers may formulate different checks; where a frame supports both a narrow and a strong reading, we test both and say which one a check touches. A card’s verdict aggregates its predictions under the published rule: every prediction is shown with its own result in the drill, and a mixed verdict means the checks landed on both sides or reached only one reading, so a mixed verdict on a compound claim substantiates none of its clauses on its own. Verdicts report how each frame's predictions fared against what is observable, as we operationalized them; they are never a ruling on the underlying policy question, and a camp's broader concern can remain entirely legitimate while its narrative's checks come back contradicted.
Every quoted item passed a two-stage verification: a first check against its source, then an independent adversarial re-check of every quote, link and date. Items that did not meet that standard were excluded before scoring. The screening ledger for this week: 10,081 posts and 606 news items sat inside the window, 5,650 posts cleared the engagement floor, a deterministic narrowing pass fed each of seven research lanes 70 posts, and 23 narratives were authored citing 92 carrier items between them. The narrowing pass ranks the floor-clearing posts by lexicon-and-beat match and then by engagement, capping each lane feed at roughly seventy posts. A post matching more than one lane goes to the lane it matches best. The research cut ran Saturday, September 12, by ruling, after a full-roster catch-up walk widened X coverage; September 12 is a partial day in the collected record, September 13 is not in it, and the Sunday nightly’s posts roll into next week’s collection. Candidates drop out for several reasons, and the polarization gate is only one of them: no codable text, no readable source, an object the identification bar puts out of reach, or a primary object that is the event's merits rather than a camp's motives. Every item was also checked against the harm methodology's gates before it could be carded here; candidates that cleared them were routed to that instrument rather than carded, and the ignition section above records where that happened this week. Clustering requires a shared canonical claim about the same trigger event on the same fault line, and a card requires either two independent carriers or a single carrier at official or major-outlet tier. Research lanes are collection beats; fault lines are the terrain the narratives land on. We ran the same queries for both camps in every lane, and we searched for a mirror before any card recorded one as unanswered. Research and drafting run through Claude under this methodology; every published classification passes the documented verification chain, and Jerm holds final editorial responsibility.
Our collection footprint covers the open web and X. Surfaces we cannot systematically see, including several large closed platforms, are underrepresented, and a narrative living mainly on those surfaces would read here as smaller than it is. Coverage inside our own footprint was narrower than last week: 111 of the 152 tracked X handles have in-window posts, against 117, after a full-roster catch-up walk on Saturday, with 41 handles uncovered at the cut. The collector caps each pull at 200 posts per handle; JackPosobiec sits at exactly that count in-window and RadioGenoa at 182, both active carriers on this queue, so the busiest carriers’ tails may be missing and their in-window volume reads low. The daily shortlists ran September 7 to 11; September 12 is a partial day in the collected record and September 13 is not in it, with the Sunday nightly rolling into next week’s collection. The news table carried no rows for the US stories that drove the week, the anniversary ceremony, the Rogers attack on El-Sayed, the Clancy jurors, the Texas board of education and the Republican convention, which surfaced through the posts table and web search, so news-side carriage of those events reads thin here; the daily shortlists added almost nothing beyond the table, since 445 of their 561 news titles were rows outside the window. Where a first-party post was not retrievable, quotes are cited to the outlet that carried them.
Penetration is scored from carrier-set breadth as a stated judgment; no within-camp sample was drawn this pass, so every penetration score should be read as an upper bound. Reach is scored on qualitative anchors, and platform-reported view counts, where shown, are floors over time rather than counts of unique readers.
The trajectory multiplier moved one score this week. On one card it is a ruling: both windows clear the five-post floor, and the reading was applied. On four it is a held reading: the instrument produced a measurement on a first appearance, the prior-window count sits under the floor, and the card shows the reading with its confidence marked low while the score ignores it. On the other eighteen it is a default with no readable base. The applied reading rests on a two-window pair of 23 and 9 posts, the held readings on pairs that start at one to four posts, and the five-post floor is a chosen threshold below which a count is treated as noise, a standing rule under review rather than a validated cut, recorded as tunable with a date; readings this early say more about small windows than about acceleration.
The published product carries a selection bias of its own. Not-testable verdicts are expected to be common, W4 narratives are logged rather than written up, and false positives are the first-class error we optimize against, so these reports over-represent charged narratives that happened to generate falsifiable predictions, which is no random subsample of charged narratives. That effect lives in the publication layer, apart from any collection or platform effect, and we name it here so readers can weigh it.
Escalation markers this week are recorded as marker instances, each with its own source URL and date on the card that carries it. The status a set of markers produces is still a judgment under the marker rules, and clicking a marker opens the card so the judgment can be read against its evidence.
The charge priority score is an operational priority score for our own review queue. It does not measure polarization. Its reach, penetration, and elite-adoption factors move together in practice (a narrative carried by elites tends to score up on all three), so nearby scores should be read as a tier rather than a ranking; we publish scores to one decimal and treat smaller differences as noise. The factor anchors, the marker list and the tier cutoffs in this instrument are conventions we are still calibrating, and every one of them is recorded as tunable with a date attached, so a later change to any of them is visible against this week rather than silent.
This research is unpaid volunteer work. If you want it to continue, the fundraiser is the way to help.
Support the fundraiserBlind verdict agreement: 25 of 28 as of 2026-07-24