Both sides call each other fascists.
Only one side is calling for public hangings.

Since the WHCD, the dominant claim in political commentary has been that one party has spent years calling the other side fascists and Nazis, and that the labeling itself creates the conditions for political violence. The label is the precursor. The rhetoric paves the way.

That claim is serious enough to deserve a real test. So I ran one.

From January 26 through April 26 of this year, every member of Congress active on X posted to a corpus that ended up 122,106 original posts deep. I built a custom classifier — validated twice against blind human review, applied identically to posts from both parties — and measured how often elected members posted six specific kinds of rhetoric. Calls to violence. Dehumanizing language. Eliminationist framing. Threats against named individuals. Enemy-of-the-people designations. Totalitarian-regime identity labels — the "fascist" and "communist" terms at the center of this week's conversation.

The methodology was set before the count ran. Whatever the data showed was going to publish.

Here's what it shows.

The fascist-calling is a tie

The most important finding in the whole analysis is also the one most likely to be skipped if you skim.

When the question is who calls the other side fascists, Nazis, communists, or Marxists, there is no statistically significant difference between the parties. Democratic members apply totalitarian-regime identity labels to Republicans at a rate of 0.055 per 100 posts. Republican members apply them to Democrats at 0.062 per 100 posts. The 95% confidence intervals around those rates overlap. Any honest statistician asked to call the comparison would call it a tie.

This is not a methodological accident. The decision to flag both "fascist" and "commie" labels was made before the counts were run, precisely because excluding either would have pre-baked an answer. If only one direction had been measured, only one direction would have been found. Measuring both symmetrically turned the question into an empirical one rather than a rhetorical one.

It's empirically a tie. Both directions, indistinguishable rates.

Where the asymmetry actually lives

If the discourse stopped at totalitarian labels, this piece would be done. But the rest of the data tells a second story that the first finding does not displace.

Five other rhetoric categories — calls to violence, dehumanizing language, eliminationist framing, threats against named individuals, and enemy-of-the-people designations — do not show a tie. In three of those five, Republican rates are significantly higher than Democratic rates, with confidence intervals that don't overlap.

These three categories drive the overall finding. Aggregate rate of eliminationist and delegitimizing rhetoric: 0.45 per 100 posts among Republican members, 0.10 per 100 among Democrats. The 95% confidence intervals don't overlap. The Republican rate is approximately 4.4 times the Democratic rate overall.

The 4.4× headline obscures the texture, and the texture is what matters. The aggregate gap is not driven by partisan labeling. The labeling is a tie. What drives it is calls to violence, dehumanizing language, and eliminationist framing — different rhetoric, different stakes. The argument that "calling someone a fascist is the precursor to violence" gets cleaner if you can show that the precursor and the violent rhetoric travel together. In this corpus, they don't. The labeling moves equally in both directions. The actual violent rhetoric does not.

One texture worth surfacing in the dehumanizing-language category specifically: Republican Cat 2 flags are dominated by ethnic, religious, and immigration-status class targeting — Somalis, Muslims, Palestinians, immigrants framed as "fraudsters," "barbarians," "terrorists," or "monsters" as identity attributes of the class. Democratic Cat 2 flags are dominated by institutional class targeting, primarily ICE agents and DHS personnel framed as "thugs" or "goons" during Operation Metro Surge. Both meet the dehumanization rule. Both flag. Whether ethnic-class targeting and institutional-class targeting are morally equivalent is a question this analysis does not answer; it measures the rhetoric, not the comparative gravity of its targets. Readers can examine the texture directly in the audit trail.

Names with receipts

Platform-level rates are one finding. Member-level findings are another. The data shows both.

Among the 480 members of Congress with at least 100 original posts on X in the 90-day window, the highest rate of eliminationist and delegitimizing rhetoric belongs to Representative Andrew Ogles (R-Tennessee), who posted 27 flagged items across 427 total posts — a rate of 6.32 per 100. His flagged content concentrates in eliminationist framing (14 of 27 flags) and calls to violence (7 flags).

Second is Representative Randy Fine (R-Florida) at 4.43 per 100, with 36 flagged posts across 812 total. His flagged content is also dominated by eliminationist framing — 24 of his 36 flags fall into Category 3, much of it directed at Muslim Americans and Palestinians as a class.

Third is Representative Tim Burchett (R-Tennessee), whose pattern is different from the first two. Burchett's 22 flags are concentrated almost entirely in calls to violence, and the specific call he repeats is hanging.

"Hang them." "Public hanging." "Now do hanging." "If we would bring back public hangings this would stop." "A short rope and a tall tree." Five posts. One member. One repeated call.

One context note worth surfacing directly. Burchett's "Hang them" was posted as a comment on a quote-repost about convicted child sex offenders. A reader who clicks the link sees that immediately. The classification stands. The methodology, set before the analysis ran, explicitly declined to grant exceptions for calls to extrajudicial execution based on the sympathy of the target — because the harm being measured is the normalization of public hanging as a policy response, not the worthiness of any particular target. Federal courts handle convicted sex offenders. A sitting United States congressman publicly endorsing public hanging is the rhetoric this analysis flags. The same rule applies to every member, regardless of who they are calling for the hanging of.

One Democrat appears in the top ten: Senator Ron Wyden of Oregon, with three flagged posts that apply totalitarian-regime identity labels to current administration officials. The classifications stand. The audit trail for every member in the top ten is published in full alongside this piece. Anyone, including these members' own offices, can examine each flagged post against the methodology and judge for themselves.

Six of the ten members in this ranking — Bernie Moreno, Andrew Clyde, Paul Gosar, Morgan Griffith, Tim Sheehy, and Ron Wyden — are flagged with the small-sample marker. Their post counts in the window fall between 114 and 196, which clears the 100-post threshold for inclusion but produces wide confidence intervals around their rates. The point estimates are real; the precise rank positions among them are not. Distinctions like Sheehy at 1.55 vs. Wyden at 1.53 are not statistically robust — both members posted three flagged items in different post-volume contexts. The headline ranking is most stable at the top: Ogles, Fine, and Burchett are clearly separated from the rest of the field at rates whose intervals don't overlap with positions four through ten.

What this measures, and what it doesn't

The full audit trail — every flagged post in the top-ten ranking, 110 posts in total — is published alongside this piece. Each post links directly to the original tweet, with full classifier reasoning and engagement data. If a member's office wants to challenge a specific classification, the receipt is available.

What the data does and doesn't say

The data does not say that the four-times-higher Republican rate of eliminationist and delegitimizing rhetoric causes political violence. Causation between speech and action runs through too many intermediating variables — media amplification, audience radicalization, organizational infrastructure, individual psychology — for a corpus analysis like this one to support causal claims.

The data does say something more limited but more concrete. Public discourse since the WHCD has placed totalitarian-regime labeling at the center of the political-violence conversation. This analysis tested that placement and found it incomplete. The labeling is a tie. The rhetoric most directly adjacent to political violence — the explicit calls for it, the dehumanizing characterizations of human beings, the eliminationist framing about who does and does not belong in this country — runs predominantly in one direction. That asymmetry is empirically documented, not editorial assertion.

Whether the finding is comfortable depends on which side of the prior week's argument you were making. The Cat 6 tie disappoints anyone who wanted to argue that one side does the labeling and the other doesn't. The Cat 1 through Cat 3 asymmetry disappoints anyone who wanted to argue that the rhetoric and the violence are unconnected.

The methodology was built to deliver findings that would disappoint someone, no matter which way the data went. It delivered.

How to verify everything in this piece

The full methodology, including the classifier prompt, the counting rules, the validation procedures, and the edge case rulings, is published at the methodology page. The complete audit trail of all 110 flagged posts in the top-ten ranking, with direct X URLs and engagement data, is at the audit trail page. The party-blind classifier prompt is reproduced in the methodology page in case anyone wishes to apply the same criteria to a different corpus or replicate the analysis on a future window.

If a specific classification is wrong, the receipt is right there. It can be checked.

That's the standard the piece holds itself to.