1. Capital at risk

Prediction-market positions can lose their entire principal. Binary contracts settle to zero on the losing side. A confidence label of "high" does not make a position safe — it reports the council’s internal agreement, not the outcome probability.

Kelly sizing recommendations on lobi.bet are capped at 25% of notional for institutional-risk hygiene. Sizing at the Kelly cap on every signal will still produce drawdowns. Most high-payoff dark-horse trades will lose.

2. We are not the counterparty

Trades happen on third-party venues (Polymarket, Kalshi, or others). lobi.bet does not custody funds, accept bets, or execute orders. Settlement, liquidity, spread, slippage, withdrawal limits, KYC, and jurisdictional access are all venue-controlled and venue-risk.

If a venue freezes withdrawals, changes its resolution rules, delists a market, or is sanctioned in your jurisdiction, we have no mechanism to recover your capital. Read the venue’s terms before transacting.

3. Calibration has limits

Our track record is published at /track-record, and the full calibration curve at /track-record/scorecard. These are the honest snapshot of how well our stated probabilities matched realised outcomes over the window shown.

Past calibration does not guarantee future calibration. Market regimes change. Rare events are rare. Under Chachie’s Law we stay in Stripe test mode until at least 50 resolved predictions have accumulated; before that threshold, published metrics are small-sample. After graduation, Brier skill versus the naked market is still subject to sample-size uncertainty.

When similar resolved markets are thin (< 3 comparable precedents), we label the market as a "novel type" on its deep-dive page and state explicitly that the reasoning is from first principles. Treat those calls with proportionally more caution.

4. Signals can be wrong in specific ways

The council publishes an invalidation condition on every call — the specific event or data update that would make the thesis wrong. If that condition fires, the call is invalidated immediately. We do not remove stale calls from the public ledger; they stay visible as misses.

Failure modes we track explicitly:

• Overconfidence — confidence score ≥ 0.85 on a losing call
• Direction miss — called the wrong side entirely
• Extreme miss — posterior far from 50% on a losing call; heavier-tailed distribution than modelled
• Close miss — barely worse than the naked market; noise rather than skill regression

Postmortems are written on every prediction where our Brier score was worse than the market’s; the editorial ledger ships publicly in Wave 1.5.

5. Jurisdictional variance

Prediction-market legality varies by country and state. In some jurisdictions, trading on Polymarket or Kalshi is restricted, taxed differently, or prohibited outright. lobi.bet does not geofence access to signals — we publish analysis — but you are responsible for understanding and complying with the laws and regulations applicable to you.

Subscribers in the US, UK, EU, Brazil, UAE, China, and other jurisdictions should review local rules before transacting on any venue we cover. When in doubt, consult a qualified local lawyer or financial adviser.

6. Data timing and market drift

Our probability estimates are computed at publication time. Market prices on Polymarket and Kalshi move continuously. By the time a signal reaches a subscriber’s inbox, the market may have already repriced toward our posterior — reducing or eliminating the edge we reported.

The deep-dive page for each market (/market/[id]) refreshes every 15 minutes; the edge and Kelly numbers there reflect the latest snapshot. Signals sent by email are timestamped and represent a point-in-time view.

7. Our own skin-in-the-game

lobi.bet is built and operated by a small family-owned team. Everyone who builds the product also trades on our own signals. We publish wins and losses in the same public ledger, at the same cadence. "Skin in the game" is not a marketing line — it is the operating constraint.

This alignment does not make us right more often. It means we feel the losses too, and we published the calibration so you can verify the math yourself.

8. Contact

Risk concerns, data corrections, or disclosure questions: cmundim@aikoda.dev.