An educational crypto, finance, and AI vocabulary puzzle about market structure, risk controls, search, and model serving.
Order Book Anatomy
These terms describe the best prices on each side, the gap between them, and how trades move the market.
- Bid: A bid is the highest price a buyer is currently willing to pay for an asset.
- Ask: An ask is the lowest price a seller is currently willing to accept for an asset.
- Spread: The spread is the difference between the best bid and the best ask in a market.
- Slippage: Slippage is the difference between an expected trade price and the price a trade actually receives.
Collateral Controls
These terms describe posted security, required buffers, valuation discounts, and forced reduction when risk rises.
- Collateral: Collateral is an asset pledged to secure a loan, trade, or other obligation.
- Margin: Margin is the collateral or equity that must be maintained to support a position or borrowing arrangement.
- Haircut: A haircut is a reduction applied to an asset's value when estimating how much borrowing power it provides.
- Liquidation: Liquidation is the forced closing of a position when losses or collateral shortfalls exceed acceptable limits.
Retrieval Pipeline
These terms cover slicing text, turning it into vectors, narrowing candidates, and fitting prompts to a model.
ChunkingEmbeddingRerankerContext Window
- Chunking: Chunking is the process of breaking large documents into smaller sections for search or retrieval.
- Embedding: An embedding is a numeric vector that represents the meaning of text, images, or other data.
- Reranker: A reranker is a model that reorders retrieved results to surface the most relevant items first.
- Context Window: The context window is the maximum amount of text a model can consider at one time.
Inference Efficiency
These terms describe reducing delay, grouping requests, reusing work, and shrinking model memory use.
- Batching: Batching combines multiple requests or items into one processing step to improve efficiency.
- Cache: A cache stores results so repeated requests can be answered faster without recomputing everything.
- Latency: Latency is the delay between sending a request and receiving a response.
- Quantization: Quantization reduces the numeric precision of a model's weights or activations to save memory and speed up inference.