Blob Cu Contracts

Availability and dependencies

Read shared CU ingestion for independent model purposes, necessary checks and non-executable pending-dependency drafts. For portal/required-field questions, use the API/service/helper settings matrix.

Blob requires CU-capable AIServices in a supported region with an embedding deployment, not Search embeddingModel or source vectors. Image verbalization/source chat are disabled. Ingestion does not choose KB reasoning, reranking, query mode or synthesis.

Use the existing Search system-assigned identity with Cognitive Services User on that account; Storage stays ResourceId/keyless with Storage Blob Data Reader. Do not copy File CU credential channels or enable local auth. The Blob documentation’s aiServices keyless configuration omits apiKey. Keys, user-assigned identities, private networking, schedules, asset stores and permission ingestion are outside this standard branch.

Verify CU capability/region/deployments, effective Search identity/RBAC and reachability through exact readbacks. configuration references selected processing/required deployment evidence. Existing dependencies permit planning; no setup-owner or azd prerequisite. Do not ask for a separate account-configuration confirmation. Default mappings need contract/observed-error evidence; never change them. cu-prerequisite-missing: name missing evidence, not a generic setup gate. No provisioning/model tests or default extraction downgrade.

Closed intent

Use the ordinary Blob intent, set processing: standard-cu and add this required object:

Fields required; references: 1–4096 characters. Omit embedding (or use null in the intent) to leave source vectors off. Only when selected, add the closed embedding choice. CU and source-vector endpoints are independently validated; they need not match. Combine CU/auth/source/vector costs and data consent; no manual credential/MI confirmation. Missing setup uses approved native bootstrap. A single trailing slash is immaterial. Custom CU endpoints, URL paths/queries/ports and credentials block. Minimal modes omit content_understanding. Private artifacts retain choices; summaries omit references, which are attestations, not CU calls.

New CU plans emit fingerprint-bound cu_plan_version: "1.0". This CU-specific marker requires the CU choice and exact presence/value binding of optional embeddings. Legacy approved wire-only standard artifacts, including expected_source_absent and expected_generated, retain their original admission, guards and fingerprints. Those generic guards predate CU planning; they do not select the new contract. Never rewrite historical approved artifacts/receipts. Removing choices from a marked CU artifact blocks if CU is missing or the wire still requests removed embeddings; changing/removing its marker invalidates existing approval. No new field is sent in the Search request.

Wire and approval

Both GA 2026-04-01 and preview 2026-08-01-preview expose azureBlobParameters.ingestionParameters with contentExtractionMode: standard, aiServices: {"uri": "<selected-endpoint>"}, disableImageVerbalization: true, null identity/schedule. Set embeddingModel only for selected source vectorization; otherwise omit it. Preview also sets public networking and empty permission options. ADLS remains preview-only in this helper with unchanged path/ACL controls.

Use the same blob_source.py --plan → private unapproved execution_input → one explicit unchanged-plan approval → --input flow. Execution uses conditional PUT /knowledgesources('<name>')?api-version=<selected-version> with If-None-Match: *, exact definition GET and status GET; no direct CU API call. Never overwrite or suffix around a collision.

CU creation also retains approved input and uses --receipt-dir. Timeout/interruption: read-only recheck, not another PUT or direct CU call. Fresh reuse capture carries no creation ownership.

Show combined source/CU/embedding cost, access and processing changes before approval. CU is billable with no free document allowance; selected documents move to CU and, only if selected, source embeddings, possibly across regions. Search retains outputs. The CU skill uses underlying CU 2025-11-01; Search manages that call, not a user-selectable version. Both GA and preview Swagger declare embeddingModel optional/nullable, with no required or conditional standard-extraction vectorizer constraint. CU deployment prerequisites must not be substituted for Search wire-field requirements. Search and CU limits apply, including five-minute analysis timeout with possible charges. No universal quality claim. Acceptance does not prove CU support; see format limits.

Verification and failure

Require exact CU mode/endpoint/auth readback. Allow only absent/null/empty-string aiServices.apiKey and null/absent ingestion identity. Masked/nonempty/malformed keys or another identity return cu-auth-conflict, without echoing values. Known null metadata, a single endpoint trailing slash and the documented resultsProcessing: rerank default do not require manual plan edits. Different endpoints, models, extraction, auth, scope, ETags or generated identities block; rerun planning and discard old consent.

Acceptance proves configuration only. The owning executor requires a relevant completed zero-failure ingestion cycle, not active or old success. Vector verification, only when configured, additionally observes generated dimensions/profile/vectorizer and nonempty zero-skip readiness. Retrieval needs its separately approved query; neither proves extraction quality or corpus-wide relevance. CU-only vector probes return embedding-not-configured, not “CU unsupported.” Fresh exact reuse remains mutation-approval-free, not ingestion/retrieval proof. Tests/mocks are not live evidence.

Keep the first failure/request ID. A failed post-write check is partial with run-owned source/children retained; never automatically delete anything. Use existing separately approved source cleanup only, never Storage, dependencies, shared resources or role assignments.

Authorities

Authorities: failure/conflict/uncertainty only.

Blob prerequisites (opens in a new tab), GA create (opens in a new tab), preview create (opens in a new tab), CU skill (opens in a new tab), GA Swagger (opens in a new tab), preview Swagger (opens in a new tab).