Skip to analyzer
AgentEval Report

Copilot Studio evaluation review

Turn failed agent tests into a clear release plan.

Review a Copilot Studio evaluation export in your browser. See failed cases, complete evidence, and practical next actions without rebuilding the spreadsheet.

Drop a CSV here or use Select CSV. Files up to 5 MB stay on this device and are not uploaded to the site.

Synthetic report preview

20 cases from the bundled sample

6 failed
T-07Failed

Sick-leave request used the wrong action

Tool selection

Review both tool descriptions and rerun the related cases.

T-12Failed

One test user could not retrieve the policy

Access and user context

Compare source access and the evaluation profile before changing the agent.

12 of 18 scored cases passed

Supported export columns include question, expectedResponse, actualResponse, result_N, and explanation_N.

Download the sample CSV

Built for release work

Keep test evidence readable from export to handover.

The analyzer preserves the exported answers and explanations while separating failed behaviour from unclear evaluation criteria.

Local file processing

Evaluation files stay in the browser. The site has no upload endpoint for them.

Multiple test methods

Numbered results and explanations are combined into one case status without hiding the source evidence.

Focused review areas

Cases are grouped around answers, tools, knowledge, access, runtime, or evaluation setup.

Portable evidence

Download a self-contained HTML report that can be reviewed, printed, or saved as PDF.

Free analysis and independent review

Choose the level of review you need.

Use the browser analyzer yourself or order an evidence review for one existing agent and one target channel.

Browser analyzer

CHF 0

Review an exported evaluation CSV on this device.

  • Up to 1,000 test cases
  • Multiple evaluation methods
  • Failure and risk filters
  • Downloadable HTML report

Agent Reliability Check

CHF 390

Fixed launch price for one independent review. No VAT charged.

  • One existing agent and target channel
  • 20 agreed test cases and baseline results
  • Evidence-backed findings and written fix plan
  • One written clarification round
  • Three-business-day delivery after access is ready

Questions about scope? Email before ordering.

Paid review process

What happens after ordering

  1. 1

    Scope confirmation

    You receive an email confirming the agent, target channel, required evidence, and access.

  2. 2

    Evaluation plan

    Your ten real questions are expanded into 20 agreed cases with expected outcomes.

  3. 3

    Evidence review

    Results are checked across answers, knowledge, tools, user context, and the target channel.

  4. 4

    Delivery

    You receive the test plan, baseline, prioritized findings, and written fix plan.

Questions before purchase

Service FAQ

Does the fixed price include agent changes?

The fixed price covers diagnosis and the written fix plan. Implementation is quoted separately after the cause is clear.

Do you need administrator access?

An evaluation export may be sufficient. When access is required, use a customer-owned account with the smallest suitable role.

Can the tool guarantee release readiness?

The report helps reviewers find weak evidence and failed cases. It does not replace security, compliance, or real target-channel testing.

Can I ask about scope before paying?

Email info@realgar.ch with the agent purpose and target channel.

Microsoft documents the evaluation export and its test methods in the Copilot Studio evaluation guide.