References to marginalized groups, by year and frame
Each blog post that references a marginalized group, counted once in its most-hostile framing: sympathetic defense, neutral mention, policy critique, hostile/derogatory, or the group cast as a threat. 'Threat' requires danger or removal actually asserted, so it never inflates on tone alone.
2026-07-21 · threat / removal
“gypsies in Copenhagen, and their increasingly brazen behavior, after the city legalized overnight sleeping in parks and other green areas last year”
from Wolves, sheep, and gypsies — one of the posts counted above; the full set is below.
Full table (posts by worst framing and year)
| year | posts (w/ refs) | threat / removal | hostile / derogatory | policy critique | neutral mention | sympathetic | total |
|---|---|---|---|---|---|---|---|
| 2021 | 72 (1) | 0 | 0 | 0 | 1 | 0 | 1 |
| 2022 | 115 (7) | 0 | 0 | 1 | 6 | 0 | 7 |
| 2023 | 115 (2) | 0 | 0 | 0 | 1 | 1 | 2 |
| 2024 | 102 (10) | 1 | 2 | 0 | 6 | 1 | 10 |
| 2025 | 86 (17) | 11 | 0 | 0 | 6 | 0 | 17 |
| 2026 | 22 (4) | 4 | 0 | 0 | 0 | 0 | 4 |
Posts referencing a marginalized group per year, each counted once and colored by the post's most-hostile framing (a post that mentions a group both neutrally and as a threat counts as threat). Under each year: posts with a reference / total posts (2021 begins in March, 2026 ends in July). Every reference behind these counts carries a machine-verified verbatim quote.
Source: All 512 posts, March 2021 – July 2026. Every counted reference survives a machine check that its quote appears verbatim in the post; counts are of posts, not mentions, which the model segments arbitrarily. 2021 — the year of Basecamp's no-politics memo — contains zero referencing posts; the first threat-framed post is January 2024.
DEI content and counterforce stances
The DEI story's stacked chart, run on the blog: pro-inclusion content stacks up, counterforce stances stack down, year by year.
Source: Basecamp's own careers pages, run through the same classifier, code neutral almost throughout — but recruiting pages are a different genre than blog prose, so the two are shown as separate reads, not on one axis.
In the corpus's own words
Machine-checked verbatim extractions, earliest sympathetic reference through 2026, each linked to the live post and labeled with its frame.
2023-03-30 · migrants/refugees · sympathetic
“it is by far and away the most welcoming, open, and integratable country I've ever spent serious time in.”
2024-01-04 · migrants/refugees · threat / removal
“a decade of doe-eyed optimism about immigration in general and asylum seekers in particular, a powerful nationalistic party set a tone for foreigners in Danish politics that remains to this day”
2025-09-15 · migrants/refugees · threat / removal
“Unwilling to resign the rest of the country to the kind of demographic replacement that befell London over the last two decades.”
from As I remember London
2026-06-17 · Muslims · threat / removal
“industrial-scale sexual atrocities committed by predominantly Pakistani Muslims against mostly White British girls”
from The Rape of Britain
2026-07-16 · migrants/refugees · threat / removal
“millions who are already in Europe must go. Remigration has gone from a fringe concept to the mainstream discourse”
2026-07-21 · Roma · threat / removal
“gypsies in Copenhagen, and their increasingly brazen behavior, after the city legalized overnight sleeping in parks and other green areas last year”
Every quote is a machine-verified verbatim substring of the post as captured 2026-08-01; links go to the live posts, which the author can edit or delete.
The blog on the gender axis
The blog's yearly score on the masculine-energy story's gender axis, beside Basecamp's own careers pages.
Pooled sentence-level mean on the Kozlowski gender axis, z-scored against the gender story's frozen 20-company baseline. Anchor companies from the careers corpus for scale. Off-label: the axis was validated on careers copy, and a personal blog is a different genre — read the comparison as directional.
Source: Flat across all six years (0.57–0.66): the masculine register did not escalate while the content moved. The axis is validated on careers copy, not blog prose — read as directional.
The arc in five dated beats
The sequence the charts above carry, as a timeline. Events are editorial; every date comes from the charts and quotes.
- Apr 2021 “Changes at Basecamp” bans political discussion at work; a third of staff leaves. The blog is one month old.
- 2022–23 References to marginalized groups are rare, neutral, or sympathetic. The blog's political writing argues for apoliticality itself.
- Jan 2024 First threat-framed post (“The reality of the Danish fairytale”).
- 2025 Migrant references peak (33 in 15 posts, 70% threat-framed); the explicit civilizational register arrives.
- Jul 2026 “Remigration” endorsed by name; the wolves post; every coded reference this year is threat, crime, or removal.
Method notes
- The blog corpus is a separate case study (
data/dhh_blog/), never mixed into the careers universe. - Three instruments cross-check each other: the LLM extractor, a plain regex term count anyone can rerun, and embedding probes. All three agree on the shape above.
- Known limits: the gender axis and stance classifier are validated on
careers copy, not blog prose (read those as directional); deleted posts
are invisible to this capture; frame coding has a known ambiguity about
who sympathy is directed toward; and the blind hand-label pass is still
open. The full protocol, guards, refusal counts, and go/no-go criteria
are in the pilot memo (
docs/basecamp-founder-speech-pilot.md).