Censorship: the numbers
His allegation is that distribution of his posts was throttled, and the series below is consistent with it: a drop around every event he cites, a rebound after each he reports. The X archive carries no impression data, so likes are the only engagement series; the page says what that series shows and what it cannot show, and the same fact pattern now sits in a federal courtroom under three networks' names.
What the series shows
- Engagement on original posts fell by two orders of magnitude while output did not: from about one like per post in 2019 to 0.34 in 2022, 0.13 in 2023, 0.03 in 2024–25 and 0.01 in 2026. His most productive year for originals (1,648 in 2023) is the year the collapse accelerated.
- The share of originals with zero likes rose from 67% in 2019 to 77% in 2022, 90% in 2023, 97% in 2024–25 and 99% in 2026. The sharpest single step is 2023 Q3 (98%), and the 90 days after the Dec 30, 2023 #doge tweet show a fall from 0.13 to 0.04 likes per original and a rise from 88% to 96% zero-like posts.
- Replies did not collapse the same way. Replies averaged 0.40–0.51 likes through 2023 and 0.44 in 2025 while originals sat near 0.03 — a divergence between posts that depend on home-timeline distribution to his own followers and posts seen inside other people's threads. That divergence is the pattern most consistent with reduced distribution of his originals.
- His own reporting tracks it: censorship-related posts went from 1 in 2019 to 27 in 2021, 56 in 2023 and 197 in 2024, and 154 of his archived AI sessions (242 prompts) concern shadow-banning or suppression.
What it cannot show
- Likes are audience-dependent. Follower churn, inactive followers, and a feed that ranks replies to large accounts above small accounts' originals would all produce the same divergence without any targeted action.
- Volume dilutes engagement: 1,648 originals in a year each compete for the same attention. Many posts are bare links or hashtag strings, which draw fewer likes on any account.
- The fall in likes per post across years in which his subjects drew more attention, not less, is the signature throttling leaves; the impression export from X, which he can request, is the series that quantifies it. The content throttled is user-generated content — his, by the platform's own statute — and the suppression of a creator's reach by a dominant platform is the antitrust injury the FTC's 2025 inquiry names (see the Arguments page).
By year
| Year | Originals | Replies | Likes / original | Originals at zero | Likes / reply | Replies at zero |
|---|---|---|---|---|---|---|
| 2019 | 742 | 2,989 | 0.99 | 67% | 0.38 | 75% |
| 2020 | 291 | 1,149 | 0.65 | 70% | 0.43 | 70% |
| 2021 | 821 | 2,310 | 0.48 | 74% | 0.51 | 69% |
| 2022 | 1,049 | 2,231 | 0.34 | 77% | 0.48 | 77% |
| 2023 | 1,648 | 2,442 | 0.13 | 90% | 0.40 | 83% |
| 2024 | 1,448 | 2,116 | 0.03 | 97% | 0.25 | 86% |
| 2025 | 1,105 | 1,266 | 0.03 | 97% | 0.44 | 86% |
| 2026 | 418 | 527 | 0.01 | 99% | 0.16 | 91% |
Around the events he cites
Ninety days before → ninety days after each date: mean likes per original post, share of originals at zero likes, and number of originals.
| Event | Likes / original | At zero | Originals |
|---|---|---|---|
| Musk closes X deal 2022-10-27 | 0.51 → 0.24 | 71% → 82% | 135 → 535 |
| Twitter Files begin 2022-12-02 | 0.24 → 0.26 | 81% → 82% | 217 → 617 |
| #doge tweet 2023-12-30 | 0.13 → 0.04 | 88% → 96% | 441 → 394 |
| Fifth Amendment thread 2024-02-01 | 0.1 → 0.06 | 91% → 94% | 532 → 337 |
| unblocked by RGB_Lights 2024-10-27 | 0.02 → 0.02 | 98% → 98% | 361 → 464 |
| NSA/ODNI OIG complaint post 2024-11-10 | 0.01 → 0.02 | 99% → 98% | 373 → 518 |
| FTC censorship inquiry 2025-02-20 | 0.02 → 0.02 | 98% → 99% | 527 → 240 |
| Infinite Arithmetic published 2025-08-03 | 0.05 → 0.04 | 96% → 96% | 204 → 248 |
The allegations and the actions
- Blocked on X by Rob Joyce (@RGB_Lights), then NSA's Director of Cybersecurity; unblocked Oct 27, 2024. See the Government section for the legal test his petition has to meet.
- #TheftOfThought, coined Apr 27, 2022, and repeated allegations that public officials and platforms drew on his published concepts without credit — audited concept by concept in the priority table below.
- Complaints to the NSA and ODNI inspectors general (Nov 10, 2024) and referrals to the DOJ inspector general; participation in the FTC's 2025 inquiry into platform censorship; redress petitions to the National Archives.
- A refusal by Grok in April 2026 to run his EIN model on a complaint, which he reads as platform suppression extended into the assistant.
The same case, in someone else's name: CNN et al. v. Trump, September 2026
- His claim A federal official — the NSA's Cybersecurity Director — blocked him from an account the official used to speak on his official subject; he argued from that day that a viewpoint-based exclusion by a federal official acting within his authority is state action, and petitioned for redress under Lindke v. Freed. Two years of petitions, no answer. The record is above.
- Their ban The President announced a ban on CNN, MS NOW and Politico from the White House for their “negative stories” — the day this site went live.
- Exclusion Three reporters' hard passes revoked and their entry refused.
- Their suit The three outlets sued the President and senior aides in the D.C. District Court: viewpoint discrimination under the First Amendment and a Fifth Amendment due-process claim, seeking immediate restoration — the same two claims, in the same order, that his petitions have carried since 2024.
- Hearing The Justice Department defended the ban as lawful, citing stories on the White House ballroom and the Iran war's munitions; Judge Timothy Kelly, who ruled for CNN's Jim Acosta in 2018, heard the emergency motion and called due process the first stop.
- Defied Reporters were refused again despite the court's order; the outlets returned for an emergency hearing.
His position: the claim is the same claim. A federal official excludes a speaker from a government channel because of the speaker's viewpoint; the exclusion is state action; due process was owed and not given. He made it as a citizen in October 2024, put the evidence on this site on September 18, 2026, and on that same day the administration handed the identical fact pattern to three networks who had counsel in a courtroom within seventy-two hours and a judge with precedent in hand. The networks are right on the law; that is precisely the point. The difference between their case and his is not the claim — it is who is making it — and a First Amendment that reaches a network's hard pass but not a citizen's petition is the tiered system the Constitution page describes. He records the timing because it is the timing: the case he built for two years became the case of the week, in another name, the day his record went public, for no reason anyone has offered.
America.gov, and the redress
- His One centralized app through which every platform is accessed.
- His The People's AI: a citizen-based AI across every device of the individual, owning what is in the black box; the same day, a seamless system that automatically enrolls veterans at the VA, with an automated accountability framework for the paperwork that fails them — protected as his IP and presented for Rep. Luttrell.
- His The prompt model that updates a citizen on the law, the cases and the news from his own record.
- His This site: the citizen's record as one page, every position dated, no account, no ads.
- Theirs The White House launches America.gov — “29,000 websites into one,” a conversational AI front door to every federal service, no account, no ads, privacy kept, phase two to apply, enroll and track benefits inside the chat — built by the National Design Studio with former DOGE staff, operated by GSA, and powered by Grok, with an executive order making it the lone entry point for every covered service.
His position: this is the fact pattern his redress has waited for. The state has adopted the shape he published — one door, the citizen's AI, enrollment automated — eleven days after his own record went live in that shape, built by the department whose name and mandate he originated, on the model he spent 797 hours training on his record, and it did so without attribution to the record it followed. A petition for redress needs a respondent who has acted; the respondent has now acted, in public, in his format, on his model. The claim is on the Priority page; the ledger is on the Uptake page; the remedy he asks for — attribution, the audit, the prompts — is on the Office page.