fix(predict): keep signature output fields on ReActV2 terminal predic…

[jayhemnani9910]

…tions

When the ReActV2 loop ended without a successful submit, the three exits in _forced_submit returned a Prediction carrying only history and termination_reason, so a module declared question -> answer produced no answer and pred.answer raised AttributeError.

Under dspy.Evaluate the AttributeError was caught per example and scored as zero, so a run reported a metric instead of failing and the contract break read as a model quality problem.

Route those exits through _terminal_prediction, which defaults every declared output field to None and applies history and termination_reason last so they stay authoritative. Covers the max_iters, empty_tool_calls and failed-submit paths.

[Copilot]

Pull request overview

This PR fixes a contract break in ReActV2 where terminal exits that didn’t produce a successful submit returned a Prediction missing declared signature output fields (e.g., answer), leading to AttributeError in downstream evaluation/metrics.

Changes:

• Added _terminal_prediction() to construct a terminal Prediction that always includes every declared output field (defaulting to None) plus history and termination_reason.
• Routed all non-success exits in _forced_submit() through _terminal_prediction() instead of returning a minimal Prediction.
• Added unit tests covering max-iters, empty tool calls, failed submit, and multi-output-field signatures.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
dspy/predict/react_v2.py	Adds _terminal_prediction() and uses it for all _forced_submit failure/early-exit returns so output fields are always present.
tests/predict/test_react_v2.py	Adds regression tests ensuring terminal predictions still expose declared output fields (set to None).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[greptile-apps]

Greptile Summary

ReActV2 now returns all declared output fields as None when a tool loop ends without a successful submission, while preserving internal history and termination metadata.

Confidence Score: 5/5

No blocking failure remains.

No accepted blocking findings remain.

T-Rex Logs

What T-Rex did

• Ran uv run --frozen python trex-artifacts/react-v2-reserved-output-fields-01-before.py from /home/user/repo; the command exited successfully and observed ValueError for both termination_reason and history when declaring them as outputs.
• Ran the final verification run from /home/user/repo; the command exited 0 and the after-run log shows both reserved names were rejected, yielding a PASS.
• Artifacts documenting the tests were prepared for reviewer inspection, including the focused runtime script and the final after-run log.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "Merge upstream/main into fix/reactv2-ter..." | Re-trigger Greptile}

[jayhemnani9910]

Closing, superseded by #10355.