How it works

• For you (default): ranked. New users get popular content; once you follow tickers, the feed personalizes.
• Following: every squawk that tags a ticker you follow, newest first. No ranking, no algorithm.

Looking for the raw firehose? Use Feed · New.

For each squawk in the last 48 hours that's not yours and either (a) tags a ticker you follow or (b) is by an author with a Motion above 50, we compute:

score = (likes + 2 × replies + 5 × tips)
      × (1 + 0.10 × author.Motion / 100)
      × 1 / (hours_since_squawk + 2)^1.3

• Tips count more than replies, which count more than likes.
• High-Motion authors get a small boost, capped at +10% (Motion 100 = 1.10×), never dominant. A new or low-Motion author is never pushed below baseline.
• Recency decays fast: a 12h-old squawk needs ~2× the engagement of a 2h-old squawk to outrank it.
• If you have zero follows, you get pure Hot (below) so the page is never empty.

• Hot (default): same shape as For You, no personalization, no follow filter:

score = (likes + 2 × replies + 5 × tips)
      / (hours_since_squawk + 2)^1.5

• New: chronological, all non-hidden squawks.
• Top week: last 7 days, by likes then replies.
• Top all-time: same sort, no time filter.

Every squawk is auto-classified by Claude when it's created. The classifier looks for scams, coordinated manipulation, and bot spam specifically: all-caps shouting, "to the moon" rocket language, urgency ("get in NOW"), fake insider claims, price targets without rationale.

• Auto-hidden at confidence ≥ 97% on manipulation, scam, or spam. Hidden squawks disappear from public feeds; the author still sees them with an "Under review" banner and can appeal.
• Sent to admin review at lower confidence, or for "suspicious" classifications.
• Real DD never gets hidden by accident. Squawks with balanced reasoning, risks alongside catalysts, sources, or technical analysis are classified clean.

• Default sorts never allow pay-to-rank. The formulas above are the whole story for For You, Hot, New, Top week, and Top all-time.
• If we ever add paid boosts, every boosted squawk will be visually labeled and excluded from the recency / engagement math. Boosted ≠ ranked.

These formulas may change as we tune for signal vs. noise. Material changes will be surfaced in-app, and this page is updated to match. What is published here is the whole ranking rule set: the exact weights, the time windows, the score thresholds and the caps that limit how far any one account can be boosted. Our source code is not public.