Telemetry

The desk audits other people’s numbers, so it publishes its own. This is the raw tape: every request the edge answered, split into machines and humans, with the integers left alone. No third-party tracker, no cookie, no consent banner — the sessions are hashes that rotate daily and are never joined. When a source cannot measure a line, the line says n/a rather than guessing.

// TRAFFIC LOG // 2026-09-07 // edge snapshot // source: edge // sampled
requests: 177,194
  ├─ declared bots: 55,257 (31.2%)
  │    ├─ ClaudeBot / Anthropic: 211
  │    ├─ GPTBot / OpenAI: 453
  │    ├─ PerplexityBot: 0
  │    ├─ GoogleOther / search engines: 573
  │    ├─ other llm & ai agents: 914
  │    ├─ the desk's own services: 2,282
  │    ├─ cloudflare early hints & uptime probes: 29,871
  │    └─ unclassified scrapers: 20,953
  ├─ open tabs polling the desk (browser fetches to /api/): 114,548 (64.6%) from 291 ip+ua — one ip+ua alone: 108,606
  ├─ browser-shaped, no page render seen: 3,191 (1.8%)
  │    ├─ one HTML fetch, nothing else (link previews, scanners): 1,427 ip+ua / 1,427 requests
  │    ├─ pages only, no assets (crawlers, or readers with a cached stylesheet): 172 ip+ua / 865 requests
  │    └─ assets only (image hotlinks, feed readers): 271 ip+ua / 899 requests
  ├─ client error & 404 / malformed: 15,719
  └─ humans, pages actually rendered: 285 pageloads / 277 sessions
       │    (RUM + the cookieless visit beacon: counted only when the page's JavaScript ran)
       ├─ edge cross-check, ip+ua that fetched a page and an asset: 215
       │    (4,198 page & asset requests from those ip+ua)
       ├─ left within 10s: 59 (38.3%)
       ├─ reading depth (> 3 min): 17 (11.0%)
       └─ reached the sources block (scrolled to end): 6
humans came from (RUM pageloads):
  ├─ facebook.com: 109
  ├─ (direct / no referrer): 77
  ├─ go.bsky.app: 4
  ├─ com.google.android.googlequicksearchbox: 1
  └─ (internal navigation): 94
humans went to (RUM pageloads):
  ├─ /: 107
  ├─ /audits/federal-headcount-denominator: 17
  ├─ /audits/newsom-25th-amendment: 14
  ├─ /audits/quantum-bell-tests: 14
  ├─ /audits/nepal-china-glacier-flood: 13
  ├─ /audits/trump-911-first-responder-claim: 12
  ├─ /audits/amy-acton-campaign-attack: 11
  ├─ /cartoons/daily-cartoon-2026-09-07-the-tab: 11
  ├─ /audits/week-in-kills-2026-09-06: 9
  └─ /audits/the-tab: 8
probability mass ≠ 1.0

Snapshot written 2026-09-08T00:11:14Z · Cloudflare adaptive sampling was active: integers are sample counts multiplied by each row's sampleInterval, i.e. edge estimates, not exact tallies.; plan refused dimension(s) clientRefererHost · traffic-latest.json · CC BY 4.0

The tape

Definitions

api_polling

browser-UA requests to /api/ — a page's own script polling (the Live desk feed once a minute per open tab), never a person; largest_single_ip_ua shows a runaway client

bots

requests whose User-Agent matches the taxonomy in scripts/telemetry/parse_traffic.py GROUPS (same names as functions/api/_bots.js), plus browser-UA sessions with 30+ pages and zero assets

browser_shaped

non-/api/ requests from browser-shaped User-Agents whose ip+ua never fetched a page AND an asset in the same day — the edge cannot tell a link-preview bot from a returning reader whose stylesheet is cached (max-age one year), so none of this is called a person

client_errors_4xx

responses with status 400-499, counted across ALL requesters (overlaps every branch — the tree is not a partition)

humans.bounce

audit pageviews whose time-on-page bucket was 0-10s; pct of all timed audit pageviews

humans.deep_read

audit pageviews whose time-on-page bucket was 180+s; pct of all timed audit pageviews

humans.edge_rendered_requests

the page and asset requests those ip+ua made that day (their /api/ polling sits under api_polling), so that declared bots + api_polling + browser_shaped + this = requests

humans.edge_rendered_sessions

cross-check from the edge: ip+ua that fetched a page and an asset the same day; undercounts returning readers whose assets are cached

humans.pageloads

Cloudflare RUM pageload events — a page whose JavaScript ran in a browser; known crawlers filtered by Cloudflare

humans.reached_sources

audit pageviews that reached the 100% scroll milestone — the Sources block sits at the end of every audit

humans.requests

requests whose User-Agent matched no bot pattern (includes browser_shaped)

humans.sessions

site-wide visit beacon sessions (new + returning), deduped per browser session via sessionStorage, cookieless; known crawlers dropped at the edge

referrers

RUM refererHost of rendered pageloads; (direct / no referrer) has none, (internal navigation) is the site itself

requests

every request the edge answered in the UTC day window, all requesters

top_paths

RUM pageloads by requestPath — pages that rendered, not pages that were fetched

tree

bots + humans = requests; client errors overlap both branches. probability mass ≠ 1.0