gossip

Instruments that read culture, made public

What people in public rooms are saying about eight domains of culture, counted every day, at no cost and with no account. Social listening, sold on, is closer to surveillance. An insight carried back into the room it came from is the root of acquaintance.

The corpus, counted 2026-10-15 17:25 UTC Newest row 2026-10-15 16:09 UTC
gossipThe RoomThe cultural corpus reading
Thursday15October 2026
Sources131of 208 · to 15:47 UTC
A line by sheet date with a gap where there is no sheet, ending at 4.3 per 100.4.314 Sep15 Oct
Work, sources a day per 100 reporting · 14 Sep to 15 Oct

Both layers agree, closing out several days where the domain's climb had nothing underneath it.

The read · bodyBody's climb stays unconfirmed for a fourth straight dayRadar only, and still the most valuable kind of signal because it has not confirmed.
The read · relationshipsRelationships retakes the board's lead the same week its loneliness-economy thread reaches a signed numberA domain move and a sweep resolution pointing the same direction, not confirmed item for item.
The read · techTech's softest reading keeps hearing an echo of its own skepticismGemini 4 Argon's contested benchmark placement lands on radar's own Hacker News board the same day Google ships it, even as tech holds the softest climate reading of the eight domains for a fourth straight day.
the workspacea notebook on a seeded scope · audience decode · the fieldOpen the field
who is on this ground, and how much of it each one is
The Naturalist Posters31% · splitThe Aggrieved Neighbours26% · prospectThe Stewardship Volunteers22% · heldThe Data Hobbyists17% · prospectThe Logbook Keepers14% · held

5 crowds on one ground · read r1, 2026-08-16 · worked example, first read, no harvest behind it

listeningthe sources as they report · 7-day deviation against each domain's own baselineOpen the listening layer
the domains of culture
work-0.38Σ43 items measured
body-0.39Σ43 items measured
relationships-0.68Σ43 items measured
language-0.76Σ43 items measured
tech-0.92Σ43 items measured
fashion-1.04Σ43 items measured
music-1.13Σ43 items measured
food-1.20Σ43 items measured

Each line runs 7-day, 3-day, 1-day, left to right, on one scale with the zero drawn in: where it falls away, the short window has stopped agreeing with the long one.

rising terms, week of 2026-08-31, every one at k ≥ 20
starter+107171 accounts · reddit-sourdough
aita+83199 accounts · reddit-amitheasshole
plant+81105 accounts · reddit-houseplants
espresso+70102 accounts · reddit-espresso
yarn+6388 accounts · reddit-knitting
homelab+6284 accounts · reddit-homelab
pens+6186 accounts · reddit-fountainpens
fresh+5486 accounts · reddit-indieheads
acne+54120 accounts · reddit-skincareaddiction
the directoryevery open source the house reads · 14 kinds countedOpen the directory
sources by kind
rss81 of 84 active
reddit66 of 82 active
anchor12 of 12 active
steam8 of 8 active
youtube8 of 8 active
stackexchange6 of 6 active
arena5 of 5 active
apple4 of 4 active
bluesky4 of 4 active
hn4 of 4 active
market3 of 3 active
wiktionary3 of 3 active
trends2 of 2 active
youtube-trending2 of 2 active
the deep record
radar.corpus_ssa_names2,181,032 rows · 1880 to 2025
radar.corpus_atus7,266 rows · 2003 to 2025
radar.corpus_gtrends24,896 rows · 2004 to 2026
radar.corpus_gdelt2,899 rows · 2024 to 2026
radar.corpus_pageviews389 rows · 2024 to 2026
  1. Read today’s sheet: the eight domains, and what moved under them.
  2. Open the day’s lead, and the domain it sits in.
  3. See how it was counted: the standard every reading is held to, and the worked records.
186,665 rows in the corpus
76,406 rows published in the last 30 days
30 segments, all at k ≥ 20
89 readings published
208 sources active of 230

These counts and the frames above include Reddit’s rooms, shown while counsel reads Reddit’s terms and outside the corpus licence. Why

today in the corpus

104new on the charts · 14 Oct

45 UK videos, 43 US videos and 7 books among them. What the charts carried

272new to Wikipedia’s top thousand · 14 Oct

101 of them for the first time since 5 July. Wikipedia’s top thousand

243stories arrived in 72 hours · 15 Oct

165 of them in music. What arrived

Open data reaches whoever can already read it. What changes that is one person telling another what was found, in their language, in the room it came from. If you take a reading from here, carry it back.

kept until you unsubscribe; nothing else is collected.
Aggregate only  every segment at k ≥ 20 No individual egress  person-level rows never leave the private tier Licence  AGPL-3.0 code, corpus terms in the draft terms What we will not collect  stated in the method
The instruments · four tenses

Four ways to read the same rows

One serves every day from the harvest; one runs on a specimen; two are specified and not built. Every one reads the corpus above; none holds a corpus of its own.

Present

The Room

The daily sheet, and the front door of the listening arm: the eight domains, how far each sits from its own baseline, and what moved under them today.

Serving at /today/
Audience

Decode

Builds a working model of an audience from rule sets and evidence rather than demographics, so a move can be tested before it is spent.

Serving at /decode/
Future

Foresight

Runs ensemble counterfactuals on a Decode model and reports a distribution rather than a single trajectory.

Spec'd, not built
Past

Hindsight

Retraces how a position or a phenomenon came to be, carriers named and hops dated, and scores its own retrodiction.

Spec'd, not built

every count on this page is stamped from public.house_counts and re-read at each deploy. a source that stops reporting, a segment that thins below k ≥ 20, and a run that writes nothing all appear here in the same voice and at the same size as a result.