root@sleven:~$ ./why

The market has an opinion. We publish the correction.

S/EVEN is an intelligence layer over AI & technology prediction markets: model-driven fair values, divergence alerts, and the primary-source feeds behind them. Every number logged before the fact to an append-only ledger you can audit.

Open markets tracked

2,168

Fair values computed

162,442

Sources monitored

783

Media→market links

18,031

live counts at page render

§ Scored in public

Every fair value is written to an append-only calibration ledger before outcomes are known, then settled against real market resolutions — Brier-scored against the market itself. Nothing is edited after the fact. Nothing is cherry-picked.

884 fair values published and awaiting resolution. The track record accrues from live signal only; lower Brier than the market is the whole game.

/track — the public record →

Fair-value computations — last 48 hours

08-05 20:0008-07 19:00

§ Where we disagree with the market

Largest modeled divergences, shown with a 24h delay — live values are for members.

MarketMarket priceOur fair valueEdge
What will be the top AI model this month? — claude-opus-5-maxkalshi:KXTOPMODEL-26AUG10-CLAUM0.9350.781-1541 bps
What will be the top AI model this month? — claude-opus-5-highkalshi:KXTOPMODEL-26AUG10-CLAU50.0150.098+825 bps
What will be the top AI model this month? — claude-opus-5-maxkalshi:KXTOPMODEL-26AUG31-CLAUM0.7550.683-722 bps
What will be the top AI model this month? — claude-opus-4-6-thinkingkalshi:KXTOPMODEL-26AUG10-CLAUT0.0150.077+615 bps
What will be the top AI model this month? — claude-fable-5kalshi:KXTOPMODEL-26AUG31-CLAUF0.0050.042+373 bps

§ Three models, one number

Every fair value is an ensemble of three in-house signal models — each component recorded per publication, so when we're wrong we can say which part was.

./PRIOR-1

Base-rate model

Settled-history persistence priors: how often does the incumbent hold? Laplace-smoothed from every resolution we've seen.

./FLOW-1

Market-structure model

Longshot-bias calibration, event-group normalization, and spread-aware thresholds over live order books on both exchanges.

./JUDGE-1

Event-judgment model

Structured AI judgment over primary-source snapshots — leaderboard states, filings, releases — cached, versioned, and logged per call.

§ Primary sources, not sentiment

Resolution feeds

The sources markets actually settle on, polled directly: benchmark leaderboards every 15 minutes, SEC EDGAR filing streams per company, official data releases.

Entity graph

Every market gets an extracted registry of the entities that decide it — auto-wired to scouted news feeds and verified official accounts.

Media linking

Incoming stories attach to the exact contracts they bear on within minutes — auto-linked with provenance, arbitrated when uncertain.

§ Access

./free

$0

  • ▸ full catalog, 24h-delayed
  • ▸ public calibration ledger
  • ▸ personal paper-trading bot
  • ▸ API — 60 req/min

./prosumer

invite

  • ▸ live fair values & alerts
  • ▸ model component breakdowns
  • ▸ media & source attribution
  • ▸ API — 300 req/min

./b2b

contact

  • ▸ WebSocket push plane
  • ▸ 1200 req/min + bulk export
  • ▸ calibration record API
  • ▸ support SLA

§ One request to see it

$ curl https://api.sleven.ai/v1/fair-values \
    -H "Authorization: Bearer sk_live_…"

{
  "ticker":  "KXTOPMODEL-26AUG10-CLAUM",
  "pFair":   0.703,     // our probability
  "pMarket": 0.875,     // what the market thinks
  "edgeBps": -1719      // the disagreement
}

S/EVEN

event intelligence, scored in public

Informational only — not investment advice. S/EVEN publishes research signals; it never executes trades or handles funds. Fair values are model outputs, not guarantees; the public ledger at /track is the complete record, including our misses.