@matt__makes's newest followers

1. Profile-signal score = 35% account age signal + 35% profile completeness signal + 30% lifetime posting-history signal.
2. Account age signal: percentage of sampled accounts created at least 180 days before this report.
3. Profile completeness signal: percentage of sampled accounts with a non-empty bio and a custom profile image.
4. Lifetime posting-history signal: percentage of sampled accounts with at least 100 lifetime posts. This does not imply recent or consistent activity.
5. Bio categories match founder/entrepreneur/startup, developer/software/engineer, or designer/design/UX/UI words.
6. Public engagement rate = (likes + reposts + replies) / observed original posts / current followers × 100. Up to 20 posts; replies and reposts excluded.

• Up to 100 provider-ordered follower profiles are selected, and the latest timeline page of each readable public one is read. This is nonrandom and cannot describe the full audience.
• One timeline page of up to 10 recent public posts per readable follower is analyzed, including replies. Every selected follower counts except Unclear ones: 7 Silent (4 with no posts of their own, 2 with posts too thin to judge, 1 private), scored 40; 0 Unavailable (suspended or deactivated), scored 25; 0 Unclear (0 could not be read after retrying, 0 not reached within the time limit, 0 with privacy not confirmed). A provider failure is retried (a rate-limit response up to three times) before a follower is left Unclear. If more than 20% of the selected followers are Unclear no score is published.
• Category percentages use all selected followers; the score is their weighted mean over the counted ones (at least 5 counted). Pattern counts overlap and show their own denominator.
• Short snapshots cannot establish unsolicited contact, follow churn, coordination, automation, X-rule violations or account authenticity. Ordinary networking and self-promotion can be legitimate.
• Formulaic or repetitive language does not establish AI authorship. Human writing, translations, templates and assisted editing can look similar.
• The model can misinterpret context and languages. No individual sampled account identity, post text or classification is published.
• To keep the model input within a fixed size bound, every follower's post text was shortened evenly and, if needed, fewer posts per follower were supplied.
• Recent-post scan is capped at two API pages (up to 40 returned posts), keeping at most 20 original posts.
• 1 selected accounts are private and are counted as Silent; their posts were not read.
• 4 selected accounts had no posts or replies of their own on their latest timeline page and are counted as Silent.

Mean over every selected follower except Unclear ones: substantive discussion 100, mixed context 60, Silent 40, repetitive promotion 25, Unavailable 25, repeated reply pattern 0. Silent covers followers with no posts of their own on the latest timeline page, followers whose posts are too thin to judge, and private accounts (counted without reading their posts). Unavailable means suspended or deactivated. Unclear followers (the read failed after retrying, they were not reached within the time limit, or their privacy was not confirmed) are left out. At least 5 counted followers are required, and if more than 20% of the selected followers are Unclear no score is given. This is an uncalibrated content-context heuristic, not a probability of authenticity or AI authorship.

Model: gpt-6-luna · Reasoning: high · Assessed 7 Oct 2026