A daily crypto, finance, and AI vocabulary puzzle about reserve checks, borrowing risk, agent workflows, and model serving tradeoffs.
Stablecoin Guardrails
These terms describe how a stablecoin's backing is checked, redeemed, and monitored.
- Reserve Ratio: Reserve ratio describes how much backing is held relative to the amount of a stablecoin or similar instrument in circulation.
- Proof of Reserves: Proof of reserves is a report or process meant to show that an issuer or custodian holds the assets it claims to hold.
- Redemption Window: A redemption window is the period when holders can exchange a token for the referenced asset or backing.
- Depeg: A depeg happens when a token moves away from its intended price target, such as one dollar for a stablecoin.
DeFi Credit Safety
These terms describe leverage limits and liquidation pressure in lending systems.
- Collateral Factor: Collateral factor is the fraction of an asset's value that can usually be counted toward borrowing capacity.
- Health Factor: Health factor is a score that shows how safely a borrowing position is collateralized relative to its debt.
- Liquidation Threshold: The liquidation threshold is the collateral level at which a borrowing position becomes eligible for liquidation.
- Borrow Limit: A borrow limit is the ceiling on how much value can be borrowed against a given collateral position.
AI Agent Workflow
These terms cover the steps an agent uses to retrieve context, choose actions, and structure outputs.
- Tool Calling: Tool calling is when an AI model sends a structured request to an external function, API, or service.
- Retrieval: Retrieval is the step where a system finds documents, facts, or records that may help answer a query.
- Planner: A planner is the component that selects or organizes the next action in a multi-step agent workflow.
- Structured Output: Structured output is model output formatted to match a known schema such as JSON or another predictable layout.
Model Serving Efficiency
These terms describe memory reuse, batching, and precision tradeoffs when serving models.
- KV Cache: A KV cache stores key and value tensors so a model can reuse prior attention work during generation.
- Batching: Batching groups multiple requests or tokens so shared compute can be used more efficiently.
- Quantization: Quantization reduces numeric precision to make model storage and inference cheaper or faster.
- Inference Latency: Inference latency is the delay between a request and a model's completed response.