Half of all captured traffic was automation wearing a human costume — convicted by named, inspectable tells (missing client hints, impossible cadence, path hammering, distributed bursts), never by a black-box score. Your JavaScript analytics is structurally unable to see most of this: agents don't run JavaScript.
| Class | Requests | Share | Evidence | Basis |
|---|---|---|---|---|
| Corroborated humans
Client hints + full fingerprint, human cadence, internal referers; 4 of 7 read local landing pages |
7 | 1.7% | 7 distinct sources, all passing every tell | TIER A · OBSERVED |
| AI agents
Training 73: meta-externalagent 47 · Bytespider 15 · Amazonbot 9 · ClaudeBot 2 ✓ — agentic browsers 7 · live answer fetches 0 |
80 | 19.6% | Self-declared UAs; ClaudeBot IP-verified against operator feed | TIER A · OBSERVED |
| Search, SEO & ads
PetalBot 25 · Bingbot 6 ✓ · Applebot 2 · OAI-SearchBot 1 ✓ · AdsBot-Google-Mobile 2 ✓ · DuckDuckBot, Baiduspider, Sogou + |
44 | 10.8% | 9 rows IP-verified (✓) against published operator ranges | TIER A · OBSERVED |
| Suspected automation
Stealth crawlers 175 (incl. two distributed proxy swarms, 44 exits — Exhibit 3) · exploit scanners 32 |
207 | 50.6% | Every conviction lists its tells; sample: missing-client-hints · fast-cadence · distributed-burst | TIER B · CORROB. |
| Unattributed
Named-but-minor bots 48 · human-claimed with insufficient evidence 23 |
71 | 17.4% | We say "don't know" out loud rather than guess | UNRESOLVED |
| Channel | Requests | Est. serve cost† | Orders touched† | Attributed revenue† |
|---|---|---|---|---|
| Corroborated humans | 7 | $0.01 | 1 | $86.00 |
| AI agents | 80 | $0.37 | 0 | $0.00 |
| Search, SEO & ads | 44 | $0.20 | 0 | $0.00 |
| Suspected automation | 207 | $0.96 | — | — |
| Unattributed | 71 | $0.33 | — | — |
The statement's destination: every machine class carried to cost, every human class carried to revenue, with the attribution tier printed on every join. AI-attributed orders arrive via the order-webhook join (Tier A/B) — never inferred from analytics.
† Illustrative values, for design only. The analytics comparison number arrives via a one-field reconciliation input (or read-only connector); serve costs, orders and revenue arrive with the order-webhook join. All traffic figures on this page are real captured telemetry — a live store on the meter (anonymized), 2026-08-02 03:29–15:25 UTC.
"Corroborated humans" is a floor, not a visitor count — it counts only sessions with positive behavioural evidence (JavaScript execution, or a multi-page visit at human pacing). A single tidy request is what a competent scraper produces, so it is reported as unattributed rather than human. Figures on this page were classified before the evidence bar was tightened (2026-08-02) and will be regenerated.
Attribution basis — TIER A observed server-side · TIER B corroborated by ≥2 independent markers, error bars published · TIER C modeled range, always labeled. Tiers never blend. ✓ = operator IP-verified against published ranges. Humanness verdicts carry named tells and are issued at the report layer; raw telemetry is never rewritten. Components may not sum to 100.0% due to rounding.
Raw IP addresses are destroyed at classification; only organizational network attribution (ASN) is retained.