How Lobi works
Lobi compares market price with multiple evidence layers and only publishes setups that survive scrutiny. Five stages. Deterministic tier gates. Every publication logged to the public ledger.
Large raw edge does not outrank bad structure.
Pipeline
Every signal flows through the same five gates. No fast lane. No exceptions at the process level.
Live market + external signal intake
Polymarket + Kalshi order books pulled every 15 minutes. News sentiment, volume anomalies, and cross-venue quotes collected in parallel.
Evidence layers over raw data
Five signal types run independently: cross-venue pricing, order flow, historical base rates, calibrated synthesis, and publication decision.
Probability + confidence band
Each signal votes on a probability. Signals agree or disagree. Confidence band is a function of agreement, not of the headline probability.
Deterministic tier gates
Auto-reject regulatory, illegal, or NSFW. Apply confidence thresholds. Route to tier. Committee review for <1% of signals (Taiwan, elections, crash, regulatory impact).
Route to Public / Edge / Pro Desk
Every publication logged to the public ledger with timestamp, published price, current price, confidence, and source count. Nothing cherry-picked.
Publication rule
The rule is codified, not negotiable. Confidence — not raw edge — gates the tier. Exceptions are governed, not implicit.
Committee review (<1% of signals)
Taiwan / cross-strait scenarios, elections in markets with active legal interference, financial-catastrophe outcomes that could be self-fulfilling, and signals implicating near-term regulatory change. These route to Pro Desk only while human reviewers log their decision. No committee signal appears in Public or Edge.
Worked examples
Same market-like setups, different outcomes. The difference is not edge. It's structure.
Signal types
Each signal votes on a probability. Disagreement is a feature: it downgrades confidence instead of silencing minority views.
ARBITER
cross-venue pricing
Compares Polymarket and Kalshi prices on the same question. Large spreads trigger a look.
FLOW
order flow + momentum
Reads the order book, not just the last print. Aggressive buying or selling shifts Lobi's view before the headline price moves.
PRIOR
historical base rate
How often has this outcome happened in similar setups? The base rate is the floor under any specific pick.
FORGE
calibrated synthesis
Combines ARBITER, FLOW, and PRIOR into one probability band. Reconciles disagreement instead of averaging it.
JUDGE
publication decision
Runs the tier gate. Rejects if any auto-reject flag trips. Otherwise routes to Pro Desk, Edge, or Public by the confidence rule.
Rejection triggers
Before any confidence calculation, these filters fire first. If any of them trip, the signal never enters the tier gate.
Confidence governance
Published signals can still move between tiers. These conditions force a downgrade, not a silent update.
New here
Lobi compares market price with multiple evidence layers and only publishes setups that survive scrutiny. That's the whole product in one sentence.
Hobby bettor
Large raw edge does not outrank bad structure. A 13pp edge with one aligned signal is worse than a 7pp edge with three. This page shows why.
Regular
The worked examples below are real-shaped. Signals with dissent inside them should scare you more than signals with thinner headline edge.
Professional
Thresholds are codified, not negotiable. Confidence gates the tier. Every publication logs timestamp, published price, current price, and source count to the public ledger. Exception policy is governed, not implicit.
Next step
See the rule applied in live signals, or audit every publication on the public ledger.