test(evals): harden plan-validator discrepancy stimulus against prompt echo

[vdstrizhkova]

Fixes #2444.

Problem

The plan-validator-discrepancy-log stimulus in evals/agent-behavior/stimuli/plan-validator.yml did not meaningfully test discrepancy detection:

1. Dangling fixtures — it pointed the agent at .copilot-tracking/plans/example.md and .copilot-tracking/research/example.md, which do not exist in the repo, so there was nothing real to validate.
2. Echo-passable graders — the prompt already contained the exact tokens the graders matched ("Discrepancy Log", "DR-", "DD-", .copilot-tracking/plans), so a reply could pass by echoing the prompt — even a "files not found" response with zero findings.

Fix

Rewrote the stimulus to give an inline plan + research pair with two genuine contradictions (missing OAuth authentication; a single-response endpoint vs. a 500-row pagination cap), and removed all output-vocabulary hints from the prompt. Tightened the graders:

• discrepancy-detected — matches analytical vocabulary (discrepanc|deviat|not addressed|missing|gap|DR-\d|DD-\d) that no longer appears in the prompt, so a pass requires the agent to actually surface the conflict.
• substantive-gap-named — confirms engagement with the real gaps (authenticat|oauth|paginat|500).
• validation-status-reported — confirms a reported validation status.

Regenerated evals/agent-behavior/eval.yaml from the partial.

Validation

• Build-AgentBehaviorSpec.ps1 -Force — spec regenerated
• eval:lint:schema — pass
• Build-AgentBehaviorSpec.ps1 -WhatIf — no drift

Scope

Scoped to the primary plan-validator-discrepancy-log stimulus named in the issue. The related plan-validator-coverage-matrix and rpi-validator.yml share the same pattern and can be addressed in a follow-up.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@5c15a03). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2446   +/-   ##
=======================================
  Coverage        ?   82.66%           
=======================================
  Files           ?      142           
  Lines           ?    20688           
  Branches        ?        0           
=======================================
  Hits            ?    17101           
  Misses          ?     3587           
  Partials        ?        0           

Flag	Coverage Δ
pester	86.20% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[WilliamBerryiii]

@vdstrizhkova - I think you found a bug in the eval runner .. looks like when the eval files change, the evals are not being re-run. Can you give copilot a quick kick to see if it can fix this such that when eval files change, the matrix runner sees that eval change and runs those evals?

[vdstrizhkova]

@WilliamBerryiii the issue was that if the stimuli was change eval is not being regenerated and the tests would be skipped.
so a few changes added:
1 - drift will be detected and i have added a command to the message so user can see what needs to be rerun
2 - once the eval is regenerated, the reviewer needs to trigger the PR validation. There is also a message in the failure but it requires the maintainer to look at it. Not sure what the exact process is for reviewers