AI Markets
AI markets are prediction markets resolved by the Flap AI Oracle instead of a direct price feed or manual-only admin result. Strike uses AI resolution in multiple pool flows, including public Flap Token Pools.
Current AI Market Surfaces
Standard pool AI markets
Admin / protocol-created
USDT or STRIKE, depending on market
Configured by Strike/admin
Configured by protocol
Flap Token Pools
Public creator flow
Creator-selected BEP20 token
Supplied by creator and stored on-chain
Fixed by Strike; AI fee covered by Strike
For public Flap Token Pools, creators do not choose the AI model and do not pay a model-specific oracle fee. They provide the market prompt and post the 0.05 BNB creator bond.
How AI Resolution Works
Prompt is configured — for Flap Token Pools, the creator prompt is stored on-chain in the native-token pool factory.
Market closes — buys stop at the market's trading close time.
Resolution time arrives — the keeper/resolver requests a FLAP AI answer using the stored prompt and outcome count.
AI proposes an outcome — the oracle returns a numeric outcome choice.
Challenge window opens — a 30 minute challenge window allows review before finalization.
Market finalizes — if unchallenged, the AI outcome finalizes; if challenged, admin review resolves the dispute.
Users claim or refund — winners claim payout; invalid/cancelled markets refund principal.
Public Flap Token Pool Prompt Scope
The current public Flap Token Pool flow is optimized for token-data questions resolvable from current Ave-supported information, such as:
price,
liquidity,
volume,
FDV / market cap,
other token-specific metrics available to the resolver.
Avoid prompts that require historical prices, social/news/web evidence, exchange listing announcements, Discord or X activity, subjective judgments, or private evidence. Those are not suitable for the current public Flap Token Pool flow.
Current Oracle Limitations
The FLAP AI Oracle is not a general-purpose truth engine. For public Flap Token Pools, creators should assume:
Ave data is read at AI invocation time; the public flow is not suitable for historical snapshots or before/after comparisons.
The hosted creator flow uses a fixed Strike-selected model and tool setup.
Evidence outside the supported token-data toolset may be unavailable or unsuitable.
The oracle returns one numeric outcome index, so every valid result must map cleanly to one listed outcome.
Ambiguous prompts, unavailable data, oracle failures, or challenged proposals may require fallback/admin handling.
Writing Good Prompts
Good prompts are precise, bounded, and machine-checkable.
They should include:
Token + chain — e.g. token contract on BNB Chain.
Data source scope — current public Flap pools should use Ave token data.
Resolution-time data rule — say that the market should use Ave data when the market resolves; do not depend on past snapshots.
Threshold and equality rule — define “above”, “at or below”, “greater than or equal”, etc.
Outcome mapping — every outcome must be mutually exclusive and cover all possible results.
Examples
Price threshold
“Resolve using Ave spot price data for TOKEN on BNB Chain when this market resolves. Choose Above $0.10 only if price is strictly greater than $0.10; otherwise choose At or below $0.10.”
Liquidity threshold
“Resolve using Ave liquidity data for TOKEN on BNB Chain when this market resolves. Choose Above $100k only if liquidity is strictly greater than $100,000; otherwise choose At or below $100k.”
FDV threshold
“Resolve using Ave FDV or market cap data for TOKEN on BNB Chain when this market resolves. Choose Reached only if FDV is greater than or equal to $50,000,000; otherwise choose Not reached.”
Avoid
“Will the project go viral?”
Subjective and not token-data-resolvable.
“Will Binance list this token?”
Requires exchange/news evidence outside the current public Flap scope.
“Will the team announce mainnet?”
Requires social/web monitoring, not Ave token data.
“Will TOKEN do well?”
Vague, no threshold or time.
Challenge / Finality
AI proposals are optimistic. During the 30 minute challenge window, a participant can challenge a proposed result by posting the configured challenge bond. A successful challenge can correct the result; an unsuccessful challenge can lose the bond according to the resolver rules.
Verification
AI resolution data is exposed through the indexer where available. The canonical market prompt and chosen outcome are tied back to on-chain events and contract state.
Related Pages
Last updated