Fix conversation context and orchestration execution reliability

[Paul Lizer (paullizer)]

Summary

• Fix orchestrated follow-ups losing pasted conversation sources and silently broadening into workspace searches. Load selected eligible stored text with current authorization, provenance and model-budget checks; clarify missing or ambiguous targets.
• Retain authoritative search results before building bounded previews. Keep source-selection completeness separate from excerpt/corpus coverage, preserve genuine partial-result guards, and avoid offering unchanged retries for historical clipped results.
• Remember explicit Orchestrate choices per account/conversation in the current browser, including draft migration and storage-failure feedback, without overriding capability restrictions or deployment defaults.
• Contain deleted-workflow continuation failures with typed classification and conditional terminal updates, without replaying uncertain effects. Add privacy-safe semantic outcomes for direct/orchestrated execution, distinguish cancellation from failure, and update regression coverage and documentation for 0.261.341.

The reported Control Center role-propagation defect was already fixed in 0.261.329. Read-only inspection confirms that fix in the configured 0.261.337 image; this PR does not weaken authorization or duplicate that fix. Production deployment, live chat retries, permission changes and continuation cleanup have not occurred and require separate approval. Discovery remains bounded, and full selected input must fit the verified model budget; no automatic splitting or global partial-input acceptance is introduced.

Linked issue

N/A. The administrator chose session-only tracking for this investigation.

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

Release-note updates were offered but not approved; the release notes remain unchanged. No new admin configuration is introduced.

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: 0.261.340 -> 0.261.341. No deployer changes.

Testing / validation

• python scripts\check_cosmos_response_boundaries.py --run-contracts - passed: 945 tests and 244 subtests; 1,059 owners, 404 doubles, zero review gaps. Ran network-blocked with manifest-pinned SDK dependencies in an isolated session environment; no application manifest or shared Python changes.
• python -m pytest -q --disable-warnings -p no:cacheprovider functional_tests\test_docs_app_surface_coverage.py functional_tests\test_docs_site_quality.py - passed: 13 tests.
• node --test functional_tests\test_v2_conversation_mode_logic.mjs - passed: all five logic checks in the test file.
• npm --prefix .\application\v2_ui run typecheck - passed.
• npm --prefix .\application\v2_ui run build -- --outDir ..\..\ui_tests\artifacts\orchestration-plan-editor - passed; generated browser-harness assets are not committed.
• git --no-pager diff --cached --check - passed before commit; working tree clean after commit/push.

Additional targeted backend and real-browser results are recorded in docs/explanation/fixes/ORCHESTRATION_CHAT_RELIABILITY_FIX.md. Broader exploratory runs also encountered logging, AST-extraction and dependency-recovery fixture failures; these validators were not weakened. This is not a claim that the entire repository suite passes. Live production metrics/chart execution and saved-attempt retries were not performed.

Documentation

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

Updated orchestration guides, chat-control reference and logging-tag reference. The fix document records root causes, behavior boundaries, operational KQL, configured-image evidence and validation limitations. Generated surface inventory has no semantic change.

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 routes or raw-settings responses. Existing source authorization, revocation checks, masking/moderation, result lineage and once-only effect fences remain enforced. Semantic diagnostics exclude prompts, replies, credentials and raw private identifiers.