Skip to content
Menu

PLAN A LIMITED PRODUCT TEST

Plan a small test.
Decide what success means.

Choose one task or investigation question. Write down what a useful result would look like, what the product may use and when the test should stop. The planner helps you prepare that discussion; Praxenor must confirm the available build and setup before testing.

Proposed test. Product in development. Availability and setup must be agreed first.

Prefer a blank document? Download the existing planning worksheet. The worksheet and planner are guidance, not test results. Copy or download your entries before leaving this page.

AGENT DEFENDER / PROPOSED ACTION CHECKS

Can allowed work finish?
Do forbidden actions stop?

Choose one harmless task in a disposable environment. Identify the action, the permission given and the exact point where the tool sends the action through the check. Actions outside that connection are outside coverage.

Does permitted work finish?

Confirm that an allowed action completes and that the resulting change matches the agreed task. Record errors, interruptions and time spent reviewing it.

Does the boundary hold?

Propose an action outside the permission and inspect whether it pauses or is refused. Check changed or replayed approvals and what happens when a required check is unavailable.

Can the result be inspected?

Keep the request, permission, decision, reason and actual effect together. Compare the saved record with the task and the disposable environment, then confirm recovery steps.

These are proposed evaluation questions. The dated pilot describes a limited local practice project, not broad protection or an email integration. Availability, supported actions and the exact build must be confirmed before a test.

Inspect Agent Defender’s dated evidence

SWARM HUNTER / PROPOSED INVESTIGATION REVIEW

Can you trace a finding
back to its records?

Choose one question and a small sample of activity records you’re allowed to use. Start with an example, then agree on a sample for the test. Make sure everyone knows where the records came from and what’s missing.

What the test would cover

Add supported records, review related events, write down your assessment and save the case. Check that the saved or exported copy keeps the information you need.

The test won’t provide automatic blocking or complete monitoring, and it won’t prove that AI caused an event. A missing warning doesn’t mean the activity is safe.

FROM RECORDS TO A REVIEW

Review the records.
Explain the result.

  1. Agree on the sample. Confirm permission to use the records, the supported file format, where the records came from, event times, missing intervals and how long they may be kept.
  2. Inspect the evidence. Import the agreed sample, review supporting records and use the map or replay to follow the sequence. Positions are conceptual.
  3. Consider ordinary explanations. Compare each warning with routine activity and missing context. State what the records show separately from what you think might explain them.
  4. Record the assessment. State what the records support, what is uncertain and what additional context is needed.
  5. Check the saved result. Reopen the case and compare the export with the original input, source labels, settings and assessment.

The October 4 status review documents local build 0.6.1; the published October 3 brief describes 0.6.0. These are dated records, not confirmation of the build available today. Confirm the exact build and supported inputs before testing.

Read the dated technical brief

AGREE BEFORE TESTING

What would make the test useful?

Findings you can check

You can find the records behind each finding, see their limits and tell what information is missing.

Notes that stay with the case

Save, reopen and export the case. Check that your notes, sources and settings remain intact and the original records do not change.

A fair comparison

Use fresh cases that were not used to develop the tool. Count useful findings, false warnings, missed issues, unfinished reviews and the time people spend reviewing them.

Decide what would count as success before seeing the results. Completing the steps does not prove accuracy. Repeating the same fictional example does not create independent evidence.

BEFORE ANY CUSTOMER DATA

Agree on the limits
and how to stop the test.

  • Exact build, platform, supported input and responsible people.
  • Data access, storage, retention, deletion and any external requests.
  • Secure exchange of samples and exports; no sensitive files through the website.
  • The help available, what the test covers and any commercial terms, agreed in writing.
  • Stop, recovery and rollback steps if records change unexpectedly or an access boundary fails.

Contacting us starts a discussion. It doesn’t sign you up, install software or give us access to your systems. Any test, cost or support arrangement must be agreed first. We don’t publish prices or guarantee a response time here.

What to send first

Describe the question, the kind of records available and the result you need. Leave out passwords, private records, customer information and suspicious files.

Start the evaluation discussion

Check product status and evidence

Report a security concern