[Paul Lizer (paullizer)]

Summary

Document-backed Orchestrate runs could reject straightforward reviews even when ordinary chat succeeded. This fixes the plan, native-reader, and answer-checker handoffs without relaxing permissions or factual checks.

• Repair missing document bindings locally only for one explicitly authorized, available selection and an existing eligible reader/action. Preserve explicit IDs, dynamic inputs, disabled work, and multi-source scope; allow at most two model corrections shared across recognized planning errors.
• Route opt-in DOCX/DOCM/DOTX comment inspection through the existing authorized Word reader, retaining supported replies, anchors, resolution information, and exact citation evidence in Classic and V2. Preserve Excel inspection and default ingestion behavior; unavailable or partial evidence never becomes a complete empty inventory.
• Restrict checker references to supplied evidence labels and provide one actionable malformed-report repair without unnecessarily regenerating the answer. Substantive grounding failures still block delivery.
• Add content-free, correlated reader/checker diagnostics with debug logging on or off, regression coverage, and current documentation for v0.261.422.

Native Word inventories remain workspace-only. Comment-only inspection can work without stored style metadata, but combined requests still require their stored sections. This PR does not add an original Word preview or replace Analyze/OCR/vision.

Linked issue

N/A. No associated issue was created or authorized for this work.

Release Notes & Latest Features

• New Feature
• Bug Fix
• UI Enhancement
• Breaking Change
• Internal only

Is this visible to end users?

• Yes
• No

Is this admin-facing (Admin Settings, governance, deployment, config)?

• Yes
• No

Should this become a Latest Feature card?

• Yes
• No
• Already added

Screenshot needed for the card?

• Yes
• No
• Attached

Version bump

• application/single_app/config.py VERSION third segment bumped, or not needed because this is docs-only
• deployers/version.txt bumped, or not needed because deployers/ was not changed

Application version: 0.261.421 -> 0.261.422. No deployer files changed.

Testing / validation

Python checks used:

$env:PYTHONPATH = "$PWD\application\single_app;$PWD\functional_tests"

Command	Result
python .\scripts\check_cosmos_response_boundaries.py --run-contracts	Passed: 2,026 tests and 431 subtests; 1,094 owners and 425 doubles with 0 review gaps. Network blocked.
Coupled regression command below	Passed: 262 tests and 3 subtests; 1 confirmed baseline case deselected.
npm --prefix .\application\v2_ui run build	Passed TypeScript/Vite build.
python .\scripts\build_docs_inventory.py	Passed; generated inventory is current.
python .\functional_tests\test_docs_app_surface_coverage.py	Passed: 7/7 checks.
python .\functional_tests\test_docs_site_quality.py	Passed: 6/6 checks.
git --no-pager diff --check	Passed.

python -m pytest -q -p no:cacheprovider --disable-warnings --tb=short --show-capture=no `
    .\functional_tests\test_orchestration_selected_source_planning.py `
    .\functional_tests\test_orchestration_plan_revision_planner.py `
    .\functional_tests\test_orchestration_document_action_binding.py `
    .\functional_tests\test_orchestration_document_action_end_to_end.py `
    .\functional_tests\test_orchestration_composition_metadata_recovery.py `
    .\functional_tests\test_orchestration_compose_output_retry.py `
    .\functional_tests\test_orchestration_retained_analysis_publication.py `
    .\functional_tests\test_orchestration_search_result_integrity.py `
    .\functional_tests\test_orchestration_result_imports.py `
    .\functional_tests\test_plugin_discovery_identity.py `
    .\functional_tests\test_document_accessibility_extraction.py `
    .\functional_tests\test_excel_document_accessibility_extraction.py `
    .\functional_tests\test_orchestration_registry_contract.py `
    -k "not test_an_endpoint_refusing_json_mode_is_asked_again_without_it"

Known baseline limitations: The deselected test has an unchanged exact-log assertion missing fields already emitted by the endpoint client. A separately attempted legacy planner-diagnostics suite has ten setup errors because its AST-only harness omits imported STORAGE_FAILURE_CODE. Both were reproduced against unmodified HEAD and left unchanged; neither exclusion waives the full storage contract gate. Additional focused and browser evidence is recorded in docs/explanation/fixes/ORCHESTRATION_RELIABILITY_PIPELINE_FIX.md.

Not performed: deployment, live beta chat submissions/replays, or a full Jekyll site build. These local results are not a claim of 100% deployed-model reliability.

Documentation

• Release notes updated, or not needed
• Feature documentation updated, or not needed
• Fix documentation updated, or not needed

Updated orchestration, Document Search, accessibility, plan-editing, citation-control, and logging documentation. Added ORCHESTRATION_RELIABILITY_PIPELINE_FIX.md with root causes, contracts, verified results, and reproduction commands. Release notes were left unchanged pending user approval; no Latest Feature card was added.

Security checklist

• New Flask routes include @swagger_route(security=get_auth_security())
• Settings sent to non-admin frontends use sanitize_settings_for_user()
• Browser JavaScript is served from local SimpleChat static assets only; no CDN-hosted JS
• No secrets, keys, connection strings, or local-only artifacts are included

No new Flask routes or frontend settings-serialization changes. Existing source authorization, current-source/integrity rechecks, factual validation, and inert citation rendering remain in place.