llms.txt
+8 −8
− - [Listing an API in an x402 Directory](https://apiosk.com/articles/listing-an-api-in-an-x402-directory/): Listing an API in an x402 directory works when one resource record joins a stable endpoint, task description, executable request contract, output boundary and current payment requirements. Search intents: list my paid API in an x402 directory; check the metadata required for an x402 listing; make an x402 endpoint callable from a directory; diagnose an incomplete x402 directory entry.− - [Price Versus Reliability Tradeoff for APIs](https://apiosk.com/articles/price-versus-reliability-tradeoff-for-apis/): API buyers should evaluate price and measured reliability together because the cheapest listed call may cost more per completed outcome after retries. Search intents: compare API price with reliability; calculate API cost after retries; choose a paid API provider by completed outcome; understand API reliability scoring.− - [Sample Size in API Benchmarks: When to Show Results](https://apiosk.com/articles/sample-size-in-api-benchmarks/): Sample size in API benchmarks is sufficient only when the observations match the intended workload and support the reported metric within a declared uncertainty rule. Search intents: determine enough samples for an API benchmark; decide when to publish API latency measurements; evaluate confidence in API success rate; handle insufficient API benchmark data.− - [Semantic API Search for Agents](https://apiosk.com/articles/semantic-api-search-for-agents/): Semantic API search for agents works when retrieval metadata states the job, accepted inputs, output boundary, exclusions, and natural query variants without changing the endpoint's contract. Search intents: make an API discoverable through semantic search; write API metadata that matches agent task language; test whether agents can find an API by intent; avoid keyword stuffing in an API listing.− - [Switching API Providers Mid-Workflow Without Rebuilding](https://apiosk.com/articles/switching-api-providers-mid-workflow/): Switching API providers mid-workflow is safe when a canonical request maps fully to the replacement, provider-specific state is excluded, and settlement is validated before selection. Search intents: switch API providers without rewriting a workflow; define a safe API provider fallback boundary; check input compatibility before API failover; preserve payment state when − - [Cold Start for a New Paid API: Getting First Calls](https://apiosk.com/articles/cold-start-for-a-new-paid-api/): The cold start for a new paid API is solved by making the endpoint eligible for suitable calls first, keeping unmeasured dimensions explicit, and using routed traffic to build comparable evidence. Search intents: get first agent calls for a new paid API; list a paid API without performance history; understand how agents discover new API providers; build measured latency and success− - [Deterministic vs Model-Driven Provider Selection](https://apiosk.com/articles/deterministic-vs-model-driven-provider-selection/): Deterministic rules should make the final provider selection when requirements are measurable, while models should interpret ambiguous tasks and hand structured requirements to that rule. Search intents: choose between deterministic and model-driven API routing; make agent provider selection reproducible; decide where an LLM belongs in API selection; design a rule-− - [Why API Providers Cannot Compare Themselves](https://apiosk.com/articles/why-api-providers-cannot-compare-themselves/): API providers cannot compare themselves because they control only their own terms and telemetry, while a credible ranking requires equivalent observations across candidates, visible weighting, and an incentive independent of any one seller. Search intents: understand why vendor API benchmarks are not neutral; decide whether to join an independent API comparison; learn what m+ - [Company Registration Number vs VAT Number](https://apiosk.com/articles/company-registration-number-vs-vat-number/): A company registration number identifies a business in a particular register, while a VAT number relates to tax registration. Check each against its relevant source, preserve the country and identifier type, and establish that the evidence concerns the intended supplier. A valid VAT response cannot replace company identity verification. Search intents: difference between a compa+ - [Latency Percentiles for API Comparison: Why the Median Misleads](https://apiosk.com/articles/latency-percentiles-for-api-comparison/): Because agent workflows chain many calls, the slow tail of each provider's latency distribution becomes a common outcome, so hard latency ceilings should be set on a tail percentile matched to the number of calls per task rather than on the median. Search intents: choose between p50, p95 and p99 when comparing APIs; set a latency ceiling for an agent tool call+ - [Supplier Identity Verification Before Onboarding](https://apiosk.com/articles/supplier-identity-verification-before-onboarding/): Supplier identity verification starts with the exact entity you plan to contract with. Match its name, jurisdiction and registration number against the relevant official record, preserve the source and date, and resolve discrepancies before relying on the result. Registration alone does not establish financial health, account ownership or a complete compliance asse+ - [Liveness Checks in API Discovery: What a Probe Can Prove](https://apiosk.com/articles/liveness-checks-in-api-discovery/): An unpaid liveness probe can confirm that a paid API endpoint resolves, answers and returns a well-formed payment challenge that matches its listing, but it cannot show that paid calls deliver, how fast they are or how often they succeed. Search intents: check whether a listed paid API is still alive; remove dead endpoints from an API directory; probe an x402 endpoint with+ - [Provider Selection for Batch Agent Jobs: Once, Per Call or Per Chunk](https://apiosk.com/articles/provider-selection-for-batch-agent-jobs/): For batch agent jobs, deciding the provider once per chunk and re-deciding only on defined triggers such as error spikes, price changes or rate limits keeps output consistent and overhead low while still reacting to drift. Search intents: choose an API provider for a large batch job; decide whether to route each call or the whole batch; switch providers + - [Comparing x402 API Marketplaces: Five Questions to Ask](https://apiosk.com/articles/comparing-x402-api-marketplaces/): x402 API marketplaces should be compared on what they measure and from where, whether their rankings can be recomputed, whether they show rejections and unmeasured dimensions, and where their reviewed boundary sits, rather than on how many listings they hold. Search intents: choose which x402 marketplace to list a paid API on; evaluate how an API marketplace ranks providers; + - [Deduplicating API Listings Across Directories Before Comparison](https://apiosk.com/articles/deduplicating-api-listings-across-directories/): API listings from different directories should be merged on strong identity signals such as the normalized resource URL, payment recipient and request schema rather than on names or descriptions, because duplicates distort relative scores and create fallback options that are not real. Search intents: stop the same API appearing twice in agent search res+ - [Tie-Breaking Between Equal API Providers: Publish the Rule](https://apiosk.com/articles/tie-breaking-between-equal-api-providers/): When two API candidates tie, the tie-break rule rather than the measurements chooses the provider, so the rule should be deterministic, published in advance and recorded with each decision. Search intents: decide which API provider wins when scores are equal; design a deterministic tie-break for agent routing; explain to a provider why it lost a tied decision; ma