Last 60 days (May 31 – Jul 29, 2026), all sources, humans only: datacenter crawlers, the Meta catalog scraper, and international click-farm junk on US-only ad campaigns are removed. Source: PostHog (project Homa), person-level.
Unique people per day after bot removal. The step-change from Jul 6 is real: the X prospecting flight, then Meta LAB scaling (edge ads + dynamic property ads from Jul 19) and the TX budget ramp.
Leads use the canonical LAB-dashboard definition: a person whose first property-detail view (/property/…/details) lands on that day and who has also used search — each person counts once, on the day they first went deep. Registrations = unique people completing onboarding that day. Both climb from Jul 20 with the Meta dynamic-ads scale-up.
Each person attributed once per week by their first tagged touch (falling back to referrer). Weeks marked * are partial. Small buckets (Referral, Organic Social, Reddit, Other tagged) fold into Other.
Full weeks only — the partial first and current weeks are excluded so the ends of these lines aren't fake dips. Rates sag in the weeks of Jul 6 and Jul 13 because the X flight poured in ~28k shallow visitors (1.3 pages/user); they recover as Meta LAB property-page traffic takes over. Engaged = 2+ property views in a day.
Per-person over the full window, first tagged touch. Definitions match the LAB live dashboard: lead = first property-detail view in-window + has searched; registration = first onboarding_completed; qualified = first-ever qualified_lead landing in-window (newly qualified — repeat qualified activity is not re-counted). Read qualified with a lag — it surfaces days-to-weeks after the engaged-lead cohort forms, so recently scaled channels (Meta LAB, late-July Monochrome) under-show it. Two more caveats: a person leads once ever (on their first deep view), so retargeting-heavy channels re-reaching old leads (Meta LAB dynamic ads) show leads < property viewers by design; and ~61% of registrations carry no first-touch UTM, inflating Direct/Organic.
28,707 fake "users" (19.2% of all tracked) were excluded. The Meta catalog crawler executes JavaScript from a real Chrome, so PostHog's bot classifier scores it as regular traffic — it is identified here by fingerprint: meta-source, no campaign tag, desktop, Meta datacenter geos (Prineville OR, Forest City NC, Fort Worth TX, Altoona IA, Springfield NE, Gallatin TN, plus Ireland/Sweden).
Generated 2026-07-30 from PostHog HogQL (person-level, project timezone America/Los_Angeles). Companion saved insight: “Real Visitors by Source (bot-filtered) — weekly”. Lead/registration/qualified definitions match the LAB live dashboard: lead = first /property/…/details view + searched; reg = onboarding_completed; qualified = qualified_lead. Filter: NOT $virt_is_bot · NOT (meta/reddit source + no utm_campaign + Desktop) · NOT (paid source + geo outside US).