SimpleFunctions

Edge.

Your model vs the market. The runtime kills the weak ones.

Adversarial edge detection across Kalshi and Polymarket. Each candidate runs a disprove-the-thesis search loop, gets walked through the orderbook for your size, classified into one of four edge classes, and ranked by post-slippage realized edge. What you see has already survived the kill chain.

GET /api/public/edges?minEdge=8&depth=highopen endpoint
Taker strategy·Maker strategy·Calibration
Tycho Brahe at his observatory Uraniborg, Hven, 1585 — leaning against the giant mural quadrant, an apprentice recording readings, the 1572 supernova chart on the desk

Tycho Brahe · Uraniborg 1585 — patient measurement against the mural quadrant; the consensus said "fixed stars," he said no.

Edges that survived the loop

▲ 57¢Will Rand Paul vote for the next Fed Chair nominee?
72¢kalshi
▲ 57¢CPI year-over-year in Jul 2026?
9¢kalshi
▲ 55¢DOJ reopens Powell investigation by...?: June 30
6¢polymarket
▲ 55¢Will average **gas prices** be above or below $4.00 by Dec 3
8¢kalshi
▲ 55¢Will average **gas prices** be above $5.40 by Dec 31, 2026?
27¢kalshi
▲ 53¢What will Gold (XAUUSD) hit in May 2026?: ↓ $4,500
40¢polymarket
▲ 53¢Will the 7-day moving average of transit calls through the S
83¢kalshi
▲ 52¢Will average **gas prices** be above or below $5.60 by Dec 3
24¢kalshi
▲ 52¢Iran x Israel/US conflict ends by...?: June 30
87¢polymarket
▲ 52¢CPI year-over-year in Jul 2026?
9¢kalshi

Updated every 15 minutes · data from Kalshi and Polymarket

What rides on every candidate

Seven fields. Adversarial search runs continuously.

Edge detection on prediction markets is not a single number. It is a tree of conditional probabilities, walked through the orderbook, under attack from a search loop trying to disprove the thesis. What survives is what surfaces.

thesisPrice

Probability your causal tree assigns to the contract resolving YES.

marketPrice

Current ask price for YES buyers (or bid for NO buyers).

edge

thesisPrice − marketPrice; positive means buy, negative means short or skip.

depthAdjustedEdge

Edge after walking the orderbook for your size — what you actually realize.

edgeClass

consensus / attention / timing / risk-premium — drives the sizing rule.

adversarialResult

Survived / downgraded / killed by the disprove-the-thesis search loop.

killChain

Price gates, news patterns, time windows that auto-cancel the candidate.

Why adversarial search beats screening

ApproachWhat it gives youWhere it breaks

Eyeballing the screener

Visible mispricing on a tab

No thesis math, no adversarial search, no depth math, no kill chain

Bloomberg news scrape

Best news terminal money buys

No probability decomposition, no contract mapping, no execution path

Generic alpha screener

Cross-asset coverage

Wrong probability calculus for binary event contracts; no settlement awareness

SimpleFunctions Edge

Thesis tree + adversarial + depth + kill

Surfaces only what survives the disprove loop and clears the depth floor

Who runs edge discovery

Six recurring shapes. Same loop, different cadence and acceptance threshold.

Discretionary PMs

Run your thesis through; see which contracts survive adversarial search before you trade.

Quant funds

Edge feed plus depth + regime + flow toxicity = drop-in features for your sizing model.

Stat-arb desks

Cross-venue timing gaps and matched-pair edges; same event, two prices.

Macro researchers

Test a macro thesis end-to-end: tree → mapped contracts → edge → execution path.

Agent-managed books

LLM agent reads context + edges, picks survivors, declares intents through MCP.

Event-driven traders

Earnings, M&A, geopolitical, regulatory — adversarial scan + matched outcome contracts.

Edge endpoints

Same shape across CLI, REST/API, and MCP adapter. Pull live edges, submit a thesis, or let an agent stream the feed.

API reference

Live edges with depth

GET /api/public/edges?minEdge=8&depth=highopen

Edges by class

GET /api/public/edges?class=consensus,attentionopen

Per-thesis edges

GET /api/public/edges?thesis={id}open

Submit thesis

POST /api/thesis { description, horizon, kill_conditions }open

MCP — sf.edges (agent)

mcp call simplefunctions.edges { thesis: "...", min_edge: 8 }open

An edge, end-to-end

From thesis sentence to ranked candidate. Five panels.

1 · thesis-implied price
Causal tree (sf thesis create ...)

  Hormuz closure persists  85%
  Oil stays above $100     91%
  → WTI peaks above $150   74%   (thesis)

Kalshi:
  KXWTIMAX-26DEC31-T150 YES   38¢ (market)

Edge   = +36¢
Class  = consensus
2 · adversarial search
Disprove queries:
  "iran ceasefire 2026"        → no path
  "hormuz reopening timeline"  → insurers stalled
  "spr release march 2026"     → ⚠ possible
     n2 (oil > $100): 91% → 84%
     edge: 36¢ → 29¢ — still viable
  "trump iran exec order"      → no signal

Result: SURVIVED · downgraded
3 · depth-adjust
Orderbook walk · size 200

  ask 39¢ × 1500
  ask 40¢ ×  600
  ask 41¢ ×  300

→ avg fill 39.4¢
   slippage 0.4¢
   depth at fill: $2,400
   liquidity score: ★high

depthAdjustedEdge = 28.6¢
4 · classify + rank
Edge class: CONSENSUS · high conviction
Kelly size:  280 (half-Kelly · 25% depth cap)
Kill chain:
  market above 60¢      (price gate)
  news 'iran ceasefire' (news trigger)
  news 'spr release > 1Mb/d'
  time > 2026-09-30     (window)

Final rank: #2 of 19 surviving edges

FAQ

What counts as an edge in prediction markets?

A pricing gap between thesis-implied probability and market price that survives spread, slippage, and adversarial search. Concretely: thesisPrice − askPrice for YES buyers, walked through the orderbook for your size, debited for worst-case slippage, then run through a search loop designed to disprove the thesis. What survives is a candidate edge.

What are the four edge classes?

Consensus gaps: your model says a different probability than the crowd. Attention gaps: a sub-claim is true and observable but the market has not noticed. Timing gaps: news propagates between Kalshi and Polymarket at different speeds. Risk premiums: the market is correct in expectation but pays you to hold an unloved tail. Each gets a different sizing rule.

How does adversarial search work?

For each candidate edge, the runtime generates 4-8 search queries designed to disprove the underlying thesis claims (e.g., "iran ceasefire negotiations 2026", "hormuz strait reopening timeline", "trump executive order oil"). It scrapes credible sources, checks for evidence, and adjusts node probabilities. Edges that survive adversarial search are higher quality.

How is edge size adjusted for liquidity?

Theoretical edge minus half-spread is the textbook number. We then walk the orderbook for your intended size, score depth (high/medium/low), and emit depth-adjusted edge — the realized edge after slippage. Sizing recommendation is half-Kelly on the depth-adjusted number, capped at 25% of orderbook depth at fill price.

Cross-venue edge — what does that catch?

Same event repriced differently on Kalshi and Polymarket. Common causes: regulatory flow asymmetry (US persons on Kalshi, USDC on Polymarket), settlement risk premium, attention timing. Surfaced as matched contract pairs with the spread, both books scored, and a recommended cross-venue intent (long the cheap leg, short the rich leg).

How often does the edge feed refresh?

Heartbeat 15 minutes for the curated edge list. Sub-second refresh on orderbook depth via WebSocket /v1/ws. News-driven adversarial search runs continuously; a kill or downgrade triggers an immediate edge re-rank. Out-of-band ticks fire on FOMC, CPI, NFP, OPEC, geopolitical breaks.

What kills a candidate edge?

Adverse evidence from search loop, an explicit kill condition firing (price gate, news pattern, time window), depth collapsing below your size, spread widening past floor, or thesis-tree falsification. Each kill is logged with reason and timestamp.

Can I run my own thesis through edge discovery?

Yes. POST your thesis as plain English to /api/thesis; the runtime decomposes into a causal tree, maps to contracts, computes thesis-implied prices, runs the edge loop, and returns ranked candidates with depth + kill conditions. CLI: sf thesis create + sf edges --thesis ID.

How does this differ from a screener?

A screener filters by spread, volume, liquidity. Edge discovery filters by thesis-vs-market mispricing — you need a model. The two are complementary: screener narrows the universe, edge discovery picks the actionable subset. Most workflows run them in series.

How accurate is the edge detection?

Live calibration is published at /calibration: Brier scores by venue, category, and price bucket. Edges that survived adversarial search outperform raw thesis edges in realized P&L by ~30% (sample-size caveat — calibration is an honest dashboard, not a sales pitch).

Related surfaces

Taker strategy

Cross the spread on news-driven directional moves with depth-real edges.

Maker strategy

Quote inside the spread, skew toward conviction side, manage adverse selection.

Calibration

Live Brier scores by venue / category / price bucket — auditable accuracy.

Heartbeat engine

24/7 monitoring + thesis re-evaluation on news and orderbook ticks.

Quant trading

Microstructure features, regime detection, depth-adjusted edge.

Portfolio Autopilot

LLM agent that picks survivors and runs the loop on a schedule.

Prediction market API

CLI, REST/Data API, real-time WebSocket streams, MCP adapter.