{ "schemaVersion": 1, "checkedDateUtc": "2026-10-09", "reviewPeriodUtc": ["2026-10-08", "2026-10-09"], "method": "Manual passage review of official documentation; short claim records and section locators only. No vendor page snapshot or live API execution is redistributed.", "scope": "Documentation is not evidence that a pinned SDK or account exercised the feature. Mutable vendor guidance is dated, not relabeled as a versioned specification.", "claims": [ { "id": "C6-SDK-CONSTRUCTION", "bibliography": "ai-sdk-openai-provider", "url": "https://ai-sdk.dev/providers/ai-sdk-providers/openai", "section": "Provider Instance / baseURL; Language Models / Responses Models and Chat Models", "claim": "Responses and Chat factories select different APIs; a base URL does not provide a missing protocol.", "authority": "mutable SDK documentation", "implementation": "Exact-tag factories and the synthetic route probe establish this selected implementation separately." }, { "id": "C6-SDK-WARNINGS", "bibliography": "ai-sdk-stream-text", "url": "https://ai-sdk.dev/v7/docs/reference/ai-sdk-core/stream-text", "section": "Returns / warnings; Parameters / maxRetries", "claim": "The SDK exposes provider warnings; maxRetries=0 disables request retries at this owner.", "authority": "mutable SDK documentation", "implementation": "The tagged loop sets maxRetries=0 and drops warnings; only the six chosen synthetic paths are newly executed. Other current SDK options are not inferred to exist in the tag." }, { "id": "C6-RESPONSES-REQUEST", "bibliography": "openai-responses-create", "url": "https://developers.openai.com/api/reference/resources/responses/methods/create/", "section": "Body parameters / instructions, input, max_output_tokens", "claim": "Responses accepts instructions and role-based input; optional max_output_tokens bounds visible output and reasoning tokens.", "authority": "mutable provider API reference", "implementation": "The normal-constructor synthetic probe separately verifies selected instruction/history encoding and omission of the optional limit; no reasoning execution." }, { "id": "C6-C9-MESSAGES-REQUEST", "bibliography": "anthropic-messages", "url": "https://platform.claude.com/docs/en/api/messages", "section": "Create a Message; Body parameters / system, messages, max_tokens", "claim": "Stateless multi-turn requests supply prior messages; system is top-level and max_tokens an explicit request field.", "authority": "mutable provider API reference", "implementation": "The exact-tag normal-constructor synthetic probe observes system/history encoding and max_tokens=1024; no live state-continuation proof." }, { "id": "C6-C9-REASONING", "bibliography": "openai-reasoning", "url": "https://developers.openai.com/api/docs/guides/reasoning", "section": "Controlling costs; Preserve reasoning across calls", "claim": "Output budgets can be exhausted before visible text; conversation history and opaque compatible reasoning items serve distinct continuity purposes.", "authority": "mutable provider guidance", "implementation": "No causal diagnosis of the historical empty-output report and no opaque reasoning replay claimed." }, { "id": "C6-STREAM-LIFECYCLE", "bibliography": "anthropic-streaming", "url": "https://platform.claude.com/docs/en/build-with-claude/streaming", "section": "Event types; Error events", "claim": "Messages streams use message/content-block lifecycle events and can emit error events.", "authority": "mutable provider guidance", "implementation": "The normal-constructor probe decodes a selected synthetic lifecycle and in-stream error, not arbitrary stream fidelity." }, { "id": "C8-HTTP-AND-STREAM-ERRORS", "bibliography": "anthropic-errors", "url": "https://platform.claude.com/docs/en/api/errors", "section": "HTTP errors; Request ID", "claim": "Overload is documented as 529; SSE can fail after HTTP 200; request-id is a support correlation field.", "authority": "mutable provider guidance", "implementation": "The probe uses selected synthetic failures; this is not the official Anthropic SDK retry policy or a protected diagnostics implementation." }, { "id": "C7-STRUCTURE", "bibliography": "openai-structured-outputs", "url": "https://developers.openai.com/api/docs/guides/structured-outputs", "section": "Function calling vs response format; Structured Outputs vs JSON mode", "claim": "Structured function arguments and structured final output have distinct configuration; JSON mode is not schema adherence.", "authority": "mutable provider documentation", "implementation": "Not exposed by the tagged neutral request; no structured-output execution claimed." }, { "id": "C7-CONJUNCTION", "bibliography": "anthropic-structured-outputs", "url": "https://platform.claude.com/docs/en/build-with-claude/structured-outputs#feature-compatibility", "section": "Feature compatibility / Incompatible with", "passage": "Returns 400 error if citations enabled with output_config.format.", "claim": "Independent citation and JSON-schema declarations cannot establish the conjunction.", "authority": "mutable provider documentation", "implementation": "Neither feature is exposed by the tagged neutral request; no live conjunction test." }, { "id": "C7-SCHEMA-EXCEPTIONS", "bibliography": "anthropic-structured-outputs", "url": "https://platform.claude.com/docs/en/build-with-claude/structured-outputs#invalid-outputs", "section": "Invalid outputs / Refusals; Token limit reached", "claim": "A refusal or max_tokens termination may not match the desired output schema.", "authority": "mutable provider documentation", "implementation": "No structured-output execution in the reference or this revision." }, { "id": "C8-REPEAT", "bibliography": "http-semantics", "url": "https://www.rfc-editor.org/rfc/rfc9110.html#section-9.2.2", "section": "9.2.2 Idempotent Methods, paragraph 4", "claim": "A client should not automatically retry a non-idempotent method without known idempotent semantics or knowledge that the original was not applied.", "authority": "RFC 9110, June 2022", "implementation": "Informs the book recovery policy; does not provide a remote effect receipt." }, { "id": "C8-WAIT", "bibliography": "http-semantics", "url": "https://www.rfc-editor.org/rfc/rfc9110.html#section-10.2.3", "section": "10.2.3 Retry-After", "claim": "Retry-After accepts an HTTP-date or nonnegative seconds delay.", "authority": "RFC 9110, June 2022", "implementation": "The book decision receives a prevalidated millisecond duration; header parsing is absent." }, { "id": "C8-RETRY-LOAD", "bibliography": "aws-retries", "url": "https://aws.amazon.com/builders-library/timeouts-retries-and-backoff-with-jitter/", "section": "Introduction; Retries and backoff; Jitter", "passage": "A timeout or failure doesn't necessarily mean that side effects haven't happened.", "claim": "Multiple retry layers amplify load; jitter changes scheduling, not effect knowledge.", "authority": "mutable engineering guidance, not a model API specification", "implementation": "Informs trusted effect inputs and bounded policy; no distributed retry runner." }, { "id": "C8-ABORT", "bibliography": "dom-abort", "url": "https://dom.spec.whatwg.org/commit-snapshots/b76da7af5fe35a2a368127cc9ec22449ac8fd559/#abortcontroller-api-integration", "section": "3 Aborting ongoing activities; 3.3 Using AbortController and AbortSignal objects in APIs", "claim": "Promise-returning APIs using this integration contract immediately reject when the supplied signal is already aborted.", "authority": "DOM commit snapshot identified in the official page revision metadata", "implementation": "Does not prove arbitrary tool callbacks cooperate or effects roll back; exact-tag local counterexamples are separate." }, { "id": "C9-LINEAGE", "bibliography": "openai-conversation-state", "url": "https://developers.openai.com/api/docs/guides/conversation-state", "section": "Manually manage conversation state; Using the Conversations API; Passing context from the previous response", "claim": "Manual supplied history, persistent conversation identity, and previous_response_id chaining are different state modes.", "authority": "mutable provider documentation", "implementation": "No provider continuation exposed by the tagged neutral request; no opaque-state replay test." }, { "id": "C10-WEB-ANNOTATIONS", "bibliography": "openai-web-search", "url": "https://developers.openai.com/api/docs/guides/tools-web-search#output-and-citations", "section": "Output and citations", "passage": "inline citations must be made clearly visible and clickable in your user interface", "claim": "URL citation annotations belong to answer output; they are not extracted-source passage locations.", "authority": "mutable provider documentation", "implementation": "Native annotations are not projected by the tagged neutral stream; the book card is not this converter." }, { "id": "C10-DOCUMENT-LOCATIONS", "bibliography": "anthropic-citations", "url": "https://platform.claude.com/docs/en/build-with-claude/citations#document-types", "section": "How citations work / Claude returns its cited response; Document types / example citation objects", "claim": "Text/block indices are zero-based, page numbers one-based, with exclusive end positions. This does not establish JavaScript UTF-16 indexing.", "authority": "mutable provider documentation", "implementation": "The book explicitly owns a different UTF-16 coordinate contract; native conversion is absent." }, { "id": "C10-CITABLE-CONTENT", "bibliography": "anthropic-citations", "url": "https://platform.claude.com/docs/en/build-with-claude/citations#citable-versus-non-citable-content", "section": "Citable versus non-citable content", "passage": "title and context are optional fields that are passed to the model but not used toward cited content.", "claim": "Document metadata is not a citable content passage.", "authority": "mutable provider documentation", "implementation": "No native document input or citation stream execution claimed." } ], "localEvidenceMethods": [ { "evidenceId": "local-composition-fixture-v1", "taskProfile": "two-measurements-v1", "method": "scripted-local-composition", "featureRecords": "Trusted synthetic policy inputs simulating scoped measurements, not adapter feature discovery or streaming-performance measurements.", "source": "code/northstar-case-study.ts and test/northstar-case-study.test.ts", "scope": { "deployment": "primary", "model": "synthetic-northstar-v1", "adapterRevision": "script-1", "taskProfile": "two-measurements-v1", "policyRevision": "policy-1" }, "proves": "Local combination of binding, admission, recovery decision, owned history, and task-specific measurement support.", "doesNotProve": "Live wire acceptance, account feature availability, stochastic task quality, or production authority." } ], "historicalAggregate": { "sourceRevision": "86fb4ebcc08b1f576c4a4a727f9f9b33be73d4f3", "release": "v0.3.0", "artifact": "evidence/test-all/2026-10-08-test-all-summary.json", "method": "retained sanitized owner-console aggregate", "attempts": 12, "fullPerAttemptReportPresent": false, "rerunInThisRevision": false } }