I thought this was interesting - so did some initial quick analysis on this. Even for yourself (which is quite high compared to others), the posts that aren't seemingly on the Primal server seem to be fairly random (in terms of content, not necessarily the critical ones of Primal) - although @fuckstr looks more content based :-) Source code:

GitHub

GitHub - BenGWeeks/censtr: Tools for analysing whether Nostr clients are censoring users

Tools for analysing whether Nostr clients are censoring users - BenGWeeks/censtr

Chart:

Primal Missing Posts by Month

--- CLAUDE ANALYSIS (who has been known to be wrong!): I ran a deeper analysis using scripts that query Primal's actual caching service (wss://cache2.primal.net/v1) and compare the results against the relays Primal claims to index (from their open-source config on GitHub). I tested the 4 named users (semisol, fuckstr, ralf, onyx) plus controls (jack, fiatjaf, jb55) and myself. The code and full results are at

GitHub

GitHub - BenGWeeks/censtr: Tools for analysing whether Nostr clients are censoring users

Tools for analysing whether Nostr clients are censoring users - BenGWeeks/censtr

The posts missing from Primal's cache don't appear to be content-targeted. For the allegedly censored users, the missing posts include mundane "gm nostriches" messages, emoji replies, and technical discussions — mixed in randomly with the critical ones. Notably, jb55's harshest anti-Primal posts (calling Primal "the biggest existential risk to nostr", accusing them of being "an op", and naming onyx, ralf, semisol and fuckstr as censorship victims) are all kept on Primal's cache. If they were filtering bycontent, those would be the first to go. The missing posts instead correlate with post age (older posts evicted first), postingvolume (high-frequency posters like fuckstr at 71% vs my 3%), and reply status (replies dropped before originals). These patterns affect all users including the controls. That said, Primal's architecture remains a legitimate structural concern. Their open-source code confirms infrastructure exists for blocking pubkeys at ingestion and retrieval, Web of Trust scoring (PageRank-style), and content moderation rules applied at read time. Users can't add relays to work around it. Whether or not it's being used for political censorship today, the capability is there — and that's worth monitoring.