receipts

receipts · every count as of 2026-07-23 · read from schema radar, read-only

What this instrument actually reads.

Every reading the loom publishes is computed from a harvest you can count. This page is the count: each feed named, its items totted, its archive depth dated, its cadence stated as it really runs. Where a feed is gated or waiting, that is written down too. The honesty is the receipt.

Items held 9,456 radar.items · first harvest 2026-07-06 · as of 2026-07-23
Sources active 49 of 51 registered radar.sources · 10 feed kinds · as of 2026-07-23
Deviation scores 2,021 radar.climate_scores · 8 daily runs since 2026-07-16
Baseline rows 2,191,483 sum of the 4 tier 0 corpus tables · coverage back to 1880
01

The live harvest

Eight feed kinds currently hold items. Tiers follow the scan method: tier 1 is upstream raw signal, tier 2 is the interpreter layer, tier 4 is verification and timing. Archive depth is when we captured; where the content itself reaches further back than our first fetch, the second line says how far.

feed tier sources items held archive depth (fetched) cadence, honestly
steamgame reviews per tracked title; the vernacular-taste corpus 1 · upstream 8 6,266 2026-07-09 → 2026-07-23reviews reach back to 2026-05-26 daily, via CI
youtubecomments on culture-essayist uploads, Data API v3 1 · upstream 8 1,108 2026-07-22 → 2026-07-23comments reach back to 2024-10-12 daily, via CI; first harvest landed 2026-07-22, so the fetch window is two days old while the comments are not
rssinterpreter newsletters via their public feeds 2 · interpreters 7 802 2026-07-06 → 2026-07-23issues reach back to 2025-10-10 daily, via CI; the pipeline's first feed
hnHacker News front page plus scoped queries, Algolia API 1 · upstream 4 604 2026-07-09 → 2026-07-23threads reach back to 2026-03-10 daily, via CI
marketKalshi, Polymarket, and Metaculus prediction markets; crowd-priced timing. Metaculus is registered and holds no items yet 4 · verification 3 361 2026-07-09 → 2026-07-23markets opened as far back as 2023-12-28 daily, via CI
redditsubreddit hot and rising, RSS path 1 · upstream 3 218 2026-07-19 → 2026-07-23posts reach back to 2026-03-20 daily, from a local machine at 08:15; Reddit closed self-service API access (Nov 2025) and blocks datacenter addresses, so CI skips this leg on purpose
blueskyJetstream firehose, time-boxed keyword sample 1 · upstream 4 91 2026-07-09 → 2026-07-23a sampling window, not an archive; small on purpose daily, via CI
gmailinterpreter newsletter issues via IMAP, one labeled inbox 2 · interpreters 11of 12 registered 6 2026-07-15 (one batch)issues reach back to 2026-07-08 daily, via CI; the IMAP credential landed 2026-07-10. The six issues held were written by the scheduled run of 2026-07-15; the 2026-07-23 run logged into the inbox and reported all 11 routed sources, 0 new items. The remaining gap is routing, not access: two labeled senders, Blackbird Spyplane and Ted Gioia, match no registered source yet, and the leg only sees what the inbox labels.
total, 8 kinds flowing 48 9,456 2026-07-06 → 2026-07-23 latest item fetched 2026-07-23 15:29 UTC

fig 01 · the live harvest. counted from radar.items joined to radar.sources, read-only, 2026-07-23. the sources column counts active registrations per kind, which can exceed the sources that already hold items. one further source is active outside this table: the anchor-graph collector, which writes edges rather than items and has harvested none yet.

not yet flowing, and why

Are.na is registered but inactive until its channels are named; Telegram's ingester is built but no channels are registered; the TikTok field-capture path was built 2026-07-19 and paused the same day because capture runs through Firefox and this desk does not; the anchor graph's tables went live 2026-07-21 with zero edges harvested so far. Each absence is a decision or a gate, recorded here rather than rounded up to "live."

02

The aggregate layer

The numbers the loom displays are not the raw items; they are deviations from baseline. At the end of every daily harvest, an aggregation pass, pure SQL and Python with no language model anywhere in it, buckets items by their true publish dates and scores each scope as a z-score against that scope's own history. A score counts as measured only after 14 days of baseline; until then it is labeled thin, visibly, and graduates on its own as history accrues.

Deviation scores held 2,021 radar.climate_scores · as of 2026-07-23
Daily scoring runs 8 2026-07-16 → 2026-07-23 · one per ingest day
Latest run 278 scores 139 measured · 139 thin · computed 2026-07-23 15:29 UTC

fig 02 · the aggregate layer. counted from radar.climate_scores, 2026-07-23. these are the same rows behind the monitor's gauges; nothing renders as live that is not in this table.

03

The baseline, tier 0

Everything above is present tense. The baseline layer is what the present is measured against: four batch corpora, loaded on a weekly cron, kept separate from the live harvest so a reading can say how far a signal deviates from normal rather than only ranking today against itself.

batch what it grounds rows held coverage refresh
wikipedia pageviewsmonthly attention counts, 15 watchlist articles collective attention: what normal interest in a topic looks like, so a present spike has something to deviate from 374 2024-07 → 2026-07 weekly cron; the current month is refetched because it is always partial
gdeltdaily share of world news coverage, 4 scoped theme queries: ai-companionship, attention-fatigue, loneliness-connection, work-reset the analogous-event backbone: the longest news baseline in the corpus, the one layer that can already ask whether this happened before 2,811 2024-07-19 → 2026-07-20 weekly cron; scoped queries per theme, never the firehose
ssa baby namesname-year-sex counts, the full national file; 105,966 distinct names a longitudinal record of taste diffusion, 1880 to 2025; calibrates how fast a name, and by extension a style, rises and falls 2,181,032 1880 → 2025 annual; a new SSA file lands each spring
atusBLS American Time Use Survey published averages, 116 activities across 3 demographic cuts revealed behavior: the hours people actually spend, not what they post; the behavioral ground truth under every attention number above it 7,266 2003 → 2025no 2020: BLS suspended fieldwork annual data; the weekly cron checks the catalog and skips when nothing new exists
total, 4 corpora 2,191,483 1880 → 2026 tables radar.corpus_pageviews, corpus_gdelt, corpus_ssa_names, corpus_atus

fig 03 · the tier 0 baseline. counted from the four radar.corpus_* tables, read-only, 2026-07-23. loaders live in pw-radar/batch/, one file per corpus.

04

How to audit this page

Where the numbers come from. Every figure above was read from the pipeline's own database, Postgres schema radar, with read-only queries on 2026-07-23. No count on this page is estimated, projected, or rounded up. Each caption names the exact table it was counted from.

What this page is not. It is not a live dashboard. The counts are a dated snapshot, which is what a receipt is. When the numbers move materially, the page is regenerated and the as-of date moves with it; the queries that regenerate every figure ship in this page's own source, where anyone who views it can read them.

How the feeds run. The harvest is a scheduled daily job plus a weekly baseline cron. Fetching is deliberately dumb: no language model touches the harvest or the aggregation, the scan layer that interprets signals is a separate and stated exception, one failing source never kills a run, and content is write-once. Where a feed is paused, unrouted, or otherwise not flowing, the table above says so in plain words instead of rounding it up to live.