A daily crypto, finance, and AI vocabulary puzzle about reserve checks, borrowing risk, agent workflows, and inference tradeoffs.
Stablecoin Mechanics
These terms describe how a stable asset is supported, redeemed, and kept near its target value.
Reserve RatioRedemption QueuePeg BandAttestation Report
- Reserve Ratio: The reserve ratio is the share of reserve assets held against outstanding stablecoin supply.
- Redemption Queue: A redemption queue is the waiting line for users who request conversion back to the reference asset.
- Peg Band: A peg band is the price range a stablecoin is expected to stay near.
- Attestation Report: An attestation report is a third-party statement that describes reserves or controls at a point in time.
DeFi Credit Risk
These terms describe the rules and pressure points around borrowing against collateral.
- Collateral Factor: The collateral factor is the fraction of a deposit's value that can count toward borrowing.
- Health Factor: A health factor is a risk score that compares collateral value with debt and liquidation pressure.
- Liquidation Threshold: The liquidation threshold is the collateral level at which a position can be closed or penalized.
- Borrow Limit: A borrow limit is the maximum amount a user can borrow against posted collateral.
AI Agent Workflow
These terms cover the planning, tool use, and self-review steps inside an agentic system.
- Tool Calling: Tool calling is the act of asking a model to use an external function or API.
- Planner Model: A planner model is a model or component that decides the next step in a multi-step task.
- Memory Buffer: A memory buffer is stored conversation or task state the agent can reuse later.
- Reflection Loop: A reflection loop is a review step where an agent checks its own output and adjusts.
Model Serving Economics
These terms describe the tradeoffs used to make inference faster, cheaper, or more reliable.
- KV Cache: A KV cache is saved key-value attention state reused to speed up token generation.
- Batch Scheduling: Batch scheduling groups requests so a server can process them efficiently.
- Quantization: Quantization reduces numeric precision to make a model smaller and faster.
- Inference Latency: Inference latency is the time it takes a model to produce an answer.