How this is built
Every surface here stands on one owned data layer — the Viral Pattern Corpus. This is how we keep it honest.
Every row is sourced and dated
Each corpus row records the platform, the language regime, the hook pattern, the structural beat map, the format, a source URL, and an observed date. No row exists without a source and an observed date.
Freshness is enforced in CI, not by promise
Virality data decays. A build-time gate fails the deployment if any row is older than its freshness window. Between 60 and 90 days a row degrades gracefully — shown as 'last observed' rather than asserted as current.
Thin pages don't generate
A platform page generates only when at least four fresh, verified rows stand behind it. Below that threshold the page does not exist. This is a hard gate, because thin templated pages are a domain-level liability, not a per-page one.
Structure, not prediction
Nothing here forecasts performance. We report observed structure and how often patterns recur. The three laws — observed-not-predicted, credited-not-appropriated, modeled-not-copied — govern every output.
Corpus
Current corpus: 57 verified rows, version 2026.07.27-2.
Observed, never predicted
We show observed patterns with dates and frequency ranges. Nothing here forecasts how many views you'll get — virality is probabilistic and anyone claiming otherwise is guessing.
Credited, never appropriated
Every pattern links to its source. We analyze structure and use official platform embeds where available — we never rehost a creator's media or present anyone's work as a failure.
Modeled, never copied
Outputs are structural skeletons you fill in yourself. This tool cannot regenerate a near-copy of its source, by design. Structure is shareable; a specific post is not yours to reproduce.