Enable native BCAL multi-turn evaluation

[martinsrui-msft]

Summary

• add typed multi-turn NL2AL entries and strict panel/turn validation while preserving the single-turn path
• run authored conversations through one native BCAL scenario session with fail-closed capability probing, strict result validation, preserved failure metrics, and isolated diagnostics
• integrate the multiturn panel across CLI, bc-eval export, workflows, artifacts, documentation, and focused tests

Current main integration

• rebased directly onto main after Refresh BCAL NL2AL dataset panels #966, Enable BCAL gold and challenge dataset panels #967, and Add BCAL package version controls #969
• preserves Add BCAL package version controls #969's composite BCAL installation, version-mode inputs, installed-version discovery, verification, and agent_version persistence
• intentionally excludes old Bcal/dataset panels #965-only import/release leftovers and unrelated changes

Validation

• 384 focused and relevant regression tests passed, including native scenario, dataset/panel, CLI/evaluation/export, workflow, and BCAL version-control coverage
• targeted strict ty check passed
• Ruff and relevant pre-commit hooks passed
• changed workflow YAML parsed successfully
• final diff against main: 18 intended multi-turn files, with Add BCAL package version controls #969's version-control files byte-identical to main

[github-code-quality]

Code Coverage Overview

Languages: Python

Python / code-coverage/pytest

The overall line coverage in commit f136782 in the martinsrui-bcal-mult... branch remains at 88%, unchanged from commit 1b36165 in the main branch.

Show a line coverage summary of the most impacted files.

File	main 1b36165	martinsrui-bcal-mult... f136782	+/-
src/bcbench/age.../mcp_gateway.py	92%	90%	-2%
src/bcbench/types.py	92%	92%	0%
src/bcbench/commands/run.py	100%	100%	0%
src/bcbench/com...nds/evaluate.py	79%	80%	+1%
src/bcbench/dat...ataset_entry.py	95%	96%	+1%
src/bcbench/res...ceval_export.py	95%	96%	+1%
src/bcbench/age...t/bcal/agent.py	86%	89%	+3%
src/bcbench/com...ands/dataset.py	56%	59%	+3%
src/bcbench/age...cal/scenario.py	0%	100%	+100%

---

_{Updated October 13, 2026 09:27 UTC}