#metrics

JustSext growth plan for Eric: goal, current funnel, constraints, projects, funnel targets, partner price ceilings and three scenarios. Data Aug 22 – Sep 20, 2026.

Per-person Claude usage across chat, Claude Code and Cowork: September 2026 and the last 90 days, ranked by sessions, with real cost vs API-price value and monthly trend.

Content-and-behavior analysis of 153 VoiceLander demo calls: topic mix, the strengthening photo-reveal engagement signal, the persisting (and evolving) end_call fabrication issue, and a new finding on internal team QA chatter appearing in production conversation data.

TrafficJunky experiment (Jul 14-20): same explicit lander + creatives, keyword vs Cams vs Dating. Dating wins on all metrics + the only payment. $534 spent, 94 signups, 1 payment ($4). Bottleneck is monetization; recommend pause and fix signup-to-payment.

Complete funnel analysis of the Sophie Dee VoiceLander demo: ad → lander → call → registration wall → registration → purchase, combining ElevenLabs conversation data with backend BigQuery events (d2c_prod). Resolves the earlier open questions on the photo-reveal mechanic and the 4-photos gate threshold.

Active BD/affiliate deals after the Jun 25 Evan × Erin review, reconciled with HubSpot (Jun 26). Pipeline gated on the free-join → first-purchase funnel (<1%); affiliate traffic held on ice by design. Covers the PH/TrafficJunky cluster, Naughty America (live test; reciprocal paused), Rabbit Reviews, CAM4Pays, FanCentro, a new media-outreach wave (LA Weekly, Georgia Straight, Babepedia, Chicago Reader), ~11 new deals, hygiene notes + next steps.

Why calls end across all agents, plus categorized agent-initiated end_call reasons with verbatim examples, recommendations, and prioritized next steps. (Excludes the max-duration artifact and quota-limit rejections.)

Deterministic analysis of all calls: termination distribution, time-to-end, per-agent profiles, the 'user silent' trailing-silence review, bounce stats, plus recommendations & prioritized next steps. (Excludes the historical max-duration artifact and account quota-limit rejections.)

Performance read across 12 TJ campaigns (4 countries x 3 device types), plus the final read of the one campaign we continued (Canada / Android Mobile). Conclusions by country, device/OS, source, creative, landing; final campaign totals, per-creative, and TrafficJunky vs Everflow attribution. Trends based on registrations.

Which quiz funnel produces better in-product behavior. Part 1: visual scoreboard (paired bars + multiples). Includes click->signup via Everflow paid-click proxy (match_me ~1.5%, creator_funnel_v2 ~1.6% — tied at top of funnel; creator pulls ahead post-signup). Both arms keyed on quizVariant identify trait. Source: BigQuery dbt marts + Everflow. Data as of Jun 4, 2026 (window May 28 to Jun 3). Facts only.

Final implemented event pack for stxtai.com quiz funnel, verified end-to-end against BigQuery prod data (last 7 days). Includes spec source-of-truth, property completeness, killed events confirmation, and 7 issues to address before TJ ramp-up.

End-to-end audit. 65 campaigns, $24K spend, 90-day window. TJ to EF reconciliation, per-campaign tracking status, working/broken inventory, and 3 launch-quality recommendations.

Aggregated audit + loss investigation. $6,342 spend, $55 revenue (ROAS 0.87%), 20.2% click loss. Root cause: Safari ITP and structural iOS WebKit drop. iOS converts 2.5x worse than Non-iOS. Includes recommendations for future campaign planning.