A daily crypto, finance, and AI vocabulary grouping puzzle focused on market plumbing and reliability concepts.
Balance Sheet Risk
These terms describe exposure and resilience concepts used in risk management for portfolios, treasuries, and counterparties.
- Duration: Duration measures how sensitive the price of a fixed-income asset is to changes in interest rates, and it is often used to manage rate risk.
- Liquidity Buffer: A liquidity buffer is a set of readily available assets kept to meet withdrawals, redemptions, or margin needs during stressed conditions.
- Counterparty Exposure: Counterparty exposure is the risk that a trading partner, lender, custodian, or service provider fails to meet its obligations, causing losses or delays.
- Basis Risk: Basis risk is the risk that two related prices (such as spot and futures, or two similar assets) move differently, reducing the effectiveness of hedges.
On-Chain Execution
These terms describe how transactions are propagated, ordered, and sometimes exploited in public transaction pipelines.
MempoolMEVSandwich AttackPrivate Orderflow
- Mempool: A mempool is the set of unconfirmed transactions broadcast to a blockchain network that are waiting to be included in a block.
- MEV: MEV (maximal extractable value) is value that can be gained by controlling the ordering, inclusion, or exclusion of transactions in a block.
- Sandwich Attack: A sandwich attack is a pattern where an attacker places transactions before and after a target swap to influence execution and capture value at the target's expense.
- Private Orderflow: Private orderflow is transaction intent routed through private relays or auction systems so it is not visible to the public mempool before execution.
Agent Building Blocks
These are common building blocks for making agents more useful and more testable in real workflows.
Tool UseRetrieval-Augmented GenerationPlanningEval Harness
- Tool Use: Tool use is when an AI agent invokes external functions or services (like search, code execution, or databases) to complete tasks beyond pure text generation.
- Retrieval-Augmented Generation: Retrieval-augmented generation (RAG) combines search over a document store with generation so outputs can be grounded in retrieved context.
- Planning: Planning is the process of decomposing a goal into ordered steps and selecting actions, which can improve reliability for multi-step agent tasks.
- Eval Harness: An eval harness is a repeatable test setup that measures an AI system's behavior on representative tasks, helping detect regressions and failures.
Verifiable AI Infrastructure
These terms describe approaches for increasing trust in AI systems by adding proofs, attestations, or isolated execution.
Trusted Execution EnvironmentZKMLModel AttestationProof of Inference
- Trusted Execution Environment: A trusted execution environment (TEE) is a hardware-backed isolated environment that can protect code and data during execution and produce attestations about what ran.
- ZKML: ZKML refers to techniques that generate cryptographic proofs about a machine learning model's computation, aiming to verify claims without revealing all inputs or internals.
- Model Attestation: Model attestation is producing evidence about the model version and execution environment used for a result, helping downstream systems reason about trust and provenance.
- Proof of Inference: A proof of inference is evidence that a particular model inference computation was performed as claimed, which can be used to improve trust in off-chain or distributed AI services.