Live Kalshi odds vs History · Blend · Top buys · Listen
Model desk
Top buys today
Ranked by |History − Live| × sample confidence × Blend agreement. Paper signals only — not advice.
Prior-day and settled boards. Headers start collapsed — expand to review. Score chips show paper call grades for that event.
Pending = snapped before Kalshi settles. After settle: Correct/Wrong + paper PnL (¢ per $1). Qualified calls need |Δ| ≥ 8pp and n ≥ 5.
Trump lane · paper
Separate lane · paper only · not in shared six scorer.Model v2: primary number is p_mention (EB shrinkage, k=10) — not raw History%.
Gates: theme agenda_confirmed · sample ≥ok (n≈4+, no hard n≥10 cliff) · |Live−p|≥8 · briefing ready · agenda_confidence≠low · event-fit floor · on slate · ≤1 Top-buy/cluster · then Confirm (manual).
Side: Live > p → paper No (market hot); Live < p → paper Yes. Speech/blend_trump lane-only — never core six.
Trump Top buys (Call-ready only · paper)
Slate ∩ Call-ready only · cluster-capped · Confirm required. Side from Live vs p_mention. Not mixed into core Top buys.
Paper grade book (this lane)
How to read this desk
Order of trust
Live — what traders price right now (Yes / No pills).
Compare — minimal spark: solid green Live (always in front), purple History, yellow Blend; shade spans min–max. Hover for exact %.
History % — purple pill. Kalshi settle rate for similar phrases.
n — sample size behind History (how many settled markets). Larger n (about 20+) is more trustworthy; under ~10 is noisy. This is the trust signal — not Blend.
Δ — Live minus History in percentage points. Positive = board hotter than history; negative = history hotter than the board. Default sort is largest |Δ| first.
Blend % — yellow pill. Estimated say-likelihood (0–100), same scale as Yes %. ~50% History + ~30% speech + ~20% event fit. Not a trust weight — use History’s n= for trust. When speech is thin, prefer History vs Live.
Past — prior-calendar-day / settled betting boards (auto-collapsed). Header chip shows paper call score (e.g. 3/5 right) instead of Listen.
Historical grades — after settle, each snapped call is graded Correct/Wrong with paper PnL. Pending = awaiting Kalshi result. This feed trains which indicators actually predict.
Trump lane — opt-in paper-only tab with Open | Past sub-tabs. Open = live boards; Past = settled boards (format priors). Separate History/briefings/grades. Not a 7th figure in the shared six scorer.
Top buys — paper edge score: |History − Live| × √(n/25) × Blend agreement (1.15 if Blend agrees with the side, 0.85 if it fights). Needs |Δ| ≥ 8pp and n ≥ 5. Yes = History above Live (history says more likely than the board). No = Live above History.
Situation
What to trust
History and Blend both well above or below Live
Stronger signal — gap is more interesting
History high, Blend much lower
Often thin speech dragging Blend — prefer History vs Live
History thin (low n), Blend higher
Speech-led only — treat as experimental
No coverage / missing History
Skip for edge hunting until more settles land
Figure selectorWho we’re tracking. Tap one to filter boards.
MarketKalshi contract phrase for that show/event.
ReadQuick gap label vs history. Hover ⓘ for the one-line why.