Skip to content

Fix orchestration follow-ups and selected model routing - #1463

Merged
Paul Lizer (paullizer) merged 3 commits into
microsoft:paullizer-react-v2-uifrom
paullizer:paullizer-second-question-context-error
Sep 7, 2026
Merged

Fix orchestration follow-ups and selected model routing#1463
Paul Lizer (paullizer) merged 3 commits into
microsoft:paullizer-react-v2-uifrom
paullizer:paullizer-second-question-context-error

Conversation

@paullizer

@paullizer Paul Lizer (paullizer) commented Sep 7, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Fix the second-question failure caused by rejecting otherwise valid resolver JSON with an unused clarification: null. Retain strict history/ID validation and bounded correction of malformed responses.
  • Honor the manually selected model or admin default across orchestration instead of silently choosing the legacy GPT-4o configuration. Pin the authorized answer selection through approval, clarification, replanning and conversational plan editing; keep dedicated planner overrides independent.
  • Persist and stream actual answer-model metadata, fail explicitly on unusable answers, and normalize Anthropic completion reasons without accepting truncated or refused output.
  • Integrate the current V2 inline-answer, approval-persistence and conversational plan-editing changes. Preserve atomic revision/execution claims, close prepared model clients on claim failures, and advance the application to 0.261.103.

Target: paullizer-react-v2-ui, the active integration branch this work depends on. The diff is limited to 23 fix-related files; this is not a promotion of the V2 branch into Development. Merge conflicts with the latest plan editor are resolved in da958455.

Linked issue

N/A.

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 to 0.261.103
  • deployers/version.txt not needed because deployers/ was not changed

Testing / validation

  • python -m pytest -q --tb=short --disable-warnings over 18 targeted orchestration, revision, model endpoint, prompt-fill, elicitation and approval-persistence suites: 394 tests and 291 subtests passed after integration with the current V2 base. The clarification fixture now reloads the revision modules with its per-test store instead of retaining another suite's Cosmos boundary.
  • The conversation-context, plan-editor and browser-to-Flask plan-editor suites passed 39 tests and 3 subtests, including selected Terra model continuity through edits, questions, restoration and execution, plus the visible answering-model label.
  • The existing V2 production build passed with output directed into UI test artifacts, not tracked application assets.
  • Route policy inventory/access/coverage scripts, documentation surface/quality scripts, and adapter/invoke-prompt contracts passed. The regenerated documentation inventory is unchanged.
  • Before this merge-conflict resolution, a two-turn synthetic replay used the real configured Terra SDK endpoint with in-memory conversation/run containers. All three completions used Terra; no production conversations or settings were written. No deployment was performed.
  • Four additional adjacent tabular/export-summary test failures were reproduced against unchanged baseline 6a34c8a6: an exact historical version assertion and three extracted-helper tests missing build_model_endpoint_identity_headers. Those unrelated tests and the export route are unchanged. Details are in the model-selection fix documentation.

Documentation

  • Release notes updated, or not needed — release-note edits remain deferred; this PR does not add its own release-note entry.
  • Feature and admin documentation updated
  • Fix documentation updated, including ORCHESTRATION_MODEL_SELECTION_FIX.md and ORCHESTRATION_CONVERSATION_CONTEXT_FIX.md

No new configuration key, dependency, schema migration or deployment configuration is required.

Security checklist

  • Existing Flask route Swagger/authentication decorators preserved, including editor endpoints; the fix adds no routes
  • No raw settings sent to non-admin frontends; model metadata excludes credentials and endpoint addresses
  • Browser JavaScript remains local-only; no CDN assets added
  • No secrets, keys, connection strings or local-only artifacts included

Accept valid nullable resolver output, preserve authorized model selections through planning and execution, and report the model that actually answered. Normalize Anthropic completion flags and add regression coverage and documentation for version 0.261.101.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Preserve the latest inline clarification and approval preference changes. Resolve models from canonical turn state within authenticated planning context, release clarification claims on model failures, and advance fix documentation and regression headers to 0.261.102.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Preserve selected models through plan edits and clarification while retaining atomic revision and execution claims. Update merged regression coverage and bump the app to 0.261.103.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
@paullizer
Paul Lizer (paullizer) merged commit f7a7d56 into microsoft:paullizer-react-v2-ui Sep 7, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant