> For the complete documentation index, see [llms.txt](https://docs.swanchain.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.swanchain.io/swan-inference/glossary.md).

# Glossary

* **Swan Chain:** A Layer 2 blockchain network that enhances the capabilities of an underlying blockchain (like Ethereum) for decentralized computing and storage, offering scalability and efficiency.
* **UBI (Universal Basic Income):** A feature within Swan Chain that ensures providers receive a minimum payment, encouraging consistent availability of decentralized services.
* **MP (Market Provider):** A platform that facilitates the allocation of computing jobs to providers, utilizing an auction engine for job distribution and a payment engine for financial transactions. It connects job requesters with providers, ensuring efficiency and transparency in job assignments and compensations.
* **FCP (Fog Computing Provider):** Offers a layered network that extends cloud capabilities to the edge of the network, providing services such as AI model training and deployment. This provider utilizes infrastructure like Kubernetes (K8S) to support scalable, distributed computing tasks.
* **ECP (Edge Computing Provider):** Specializes in processing data at the source of data generation, using minimal latency setups ideal for real-time applications. This provider handles specific, localized tasks directly on devices at the network’s edge, such as IoT devices.
* **Public Service Jobs:** Tasks generated by Swan Chain to maintain provider engagement and system stability, especially when the UBI distribution is higher than a set target.
* **Creator Grants:** Rewards issued by Swan Chain for contributing new tasks or services, promoting innovation and expansion within the network.
* **Reputation Score:** A metric used within Swan Chain to assess the performance and reliability of providers, potentially influencing their eligibility for UBI and tasks.
* **Block Time:** The time interval Swan Chain takes to finalize and add a new set of transactions to the blockchain.
* **Blacklist:** A registry on Swan Chain of providers who have been deemed unreliable or who have failed to secure jobs within a specific timeframe, affecting their network privileges.
* **zk-proof (Zero-Knowledge Proof):** A cryptographic technique used on Swan Chain that allows a party to prove possession of certain information without revealing the information itself, enhancing privacy and security.
* **Treasure DAO:** The decentralized autonomous organization on Swan Chain tasked with managing funds and allocations for UBI, network tasks, and creative incentives.
* **Committee:** A group within Swan Chain, likely comprising stakeholders like token holders and developers, responsible for decision-making regarding fund management and network governance.
* **Collateral Reward:** Incentives provided to participants who lock their tokens as a form of security or investment in Swan Chain, supporting its operations.
* **Target UBI Rate:** The desired proportion of providers receiving UBI within Swan Chain. If exceeded, additional network tasks may be generated to balance the system.
* **Average Job Duration:** The typical length of time a task or job takes to be completed by a provider on Swan Chain.
* **Job Price:** The amount of compensation set for the completion of a specific task within Swan Chain.
* **Swan AP (Accounts Payable) and Swan AR (Accounts Receivable):** Financial terms adapted for Swan Chain, indicating modules that manage outgoing and incoming transactions, respectively.
* **Swan Hub:** A central node or sector within Swan Chain that facilitates interactions and transactions between users and providers, possibly through regional hubs.
* **Swan 2.0 / Inference Cloud:** The evolution of Swan Chain from a UBI-subsidised computing network to a market-driven AI inference marketplace at [inference.swanchain.io](https://inference.swanchain.io): an OpenAI-compatible API in front of WebSocket-connected GPU providers, paid per token. Proposed in SIP-002; UBI ended under SIP-003.
* **Contribution Score:** A merit metric proposed in SIP-002 for weighting SWAN provider rewards by real work during the UBI taper. With UBI ended, provider income is the per-token payout.
* **Credit balance:** A consumer's prepaid, USD-denominated balance on Swan Inference, funded by card (Stripe) or crypto deposit (USDC, USDT, SWAN). Every request is deducted from it per token. Larger single deposits are credited with a bonus (see **Deposit bonus**).
* **Computing Provider (CP):** The unified provider role in Swan 2.0 that replaces the separate ECP and FCP classifications. A CP runs a model server plus the open-source `computing-provider` agent, which connects outbound to Swan Inference over WebSocket — no public IP, domain or TLS certificate required.
* **Benchmark:** Periodic quality verification tests run by the Swan Inference platform (default: every 24 hours) to ensure provider inference quality. Tests include math accuracy, code generation, and response latency. Failing benchmarks can result in suspension or collateral slashing.
* **MerkleDistributor:** A smart contract used for gas-efficient batch settlement of provider earnings. Provider balances are aggregated into a Merkle tree, the root is submitted on-chain, and providers claim earnings via Merkle proofs.
* **Inference Marketplace:** The decentralized marketplace within Swan 2.0 that matches AI model consumers with GPU providers for real-time inference workloads, as opposed to the existing AI Computing Marketplace which handles training tasks via auction.
* **Consumer API key (`sk-swan-…`):** The key a developer sends as `Authorization: Bearer` to call `https://api.swanchain.io/v1`. Created in the dashboard under Keys.
* **Provider API key (`sk-prov-…`):** The key a provider puts in the `computing-provider` config so the agent can register its models with Swan Inference.
* **Deposit bonus:** Extra credit added to a deposit: **+2%** from $100, **+4%** from $500, **+6%** from $2,000, or a flat **10%** for SWAN deposits on Swan Chain. The two schemes do not stack. Live schedule: `GET /api/v1/subscription/deposit-bonus-tiers`.
* **Token Plan (Pro):** A retired monthly subscription, no longer sold. Subscriptions bought before it was withdrawn run to the end of the period already paid for and are not renewed; everything else is pay-as-you-go.
* **Two-price model:** Every catalog model publishes two prices per million tokens — what consumers pay and what providers are paid. The platform margin is the spread; there is no percentage commission.
* **Model tier (`standard` / `premium`):** A catalog grouping published in the `tier` field of `GET /api/v1/models`. It groups open-source models served by the provider network apart from frontier gateway models; it does not change how a request is billed, which is always per token at the model's own price.
* **Context window:** The maximum prompt-plus-completion length a model offering accepts. It is a property of each provider's serving configuration, not just of the model; providers must report the window their backend really serves (see the [Context-Window Integrity notice](/swan-inference/become-a-provider/provider-context-window-faq.md)).
* **Canonical serving config:** The precision (e.g. `AWQ-4`, `FP8`, `BF16`) and engine (vLLM, SGLang, llama.cpp) a catalog model is advertised at. Reference nodes and verification compare providers against this configuration, not against the lab release.
* **Verification:** Ongoing checks that a provider is serving the model it claims — benchmark prompts, fingerprint and logprob comparisons, latency and context probes. Results feed a trust score that influences routing; persistent failures can lead to capped traffic or slashing of collateral.
* **Collateral (Swan 2.0):** A refundable deposit a provider account makes before it is activated for paid traffic; it backs the slashing rules. Distinct from the CU-based collateral of Swan 1.0.
* **Explicit provider selection:** Pinning a request to one provider's offering with the `X-Swan-Provider` header (optionally refusing fallbacks with `X-Swan-Allow-Fallbacks: false`). Billed from the credit balance at the catalog price like any other request; the response carries a receipt (`X-Swan-Route-Mode`, `X-Swan-Fallback-Reason`, `X-Swan-Billing-Type`) saying who served it and how it was billed.
* **Guaranteed context length:** The smallest context window known across a model's online providers — what any auto-routed request can rely on — as opposed to the maximum any single provider reports.