Skip to contentWitnora
Menu
Sign inStart free
Sign in
Start free
Method & transparency

See how Witnora measures Agent behavior and evidence.

This public lab exposes sample evaluation runs and failure analysis. Customer workspaces use Nora-guided run summaries instead of raw corpus files.

These are public evaluation examples, not customer evidence and not Witnora service-health incidents.
Evidence snapshotExport reviewed dataset Open published evidence bundleOpen workspaceOpen components on GitHub
Privacy boundaryRaw prompts, credentials, arguments, and target responses stay in the customer environment by default.

The transparency lab contains declared claims, commitments, provenance, artifacts, and limitations. It does not expose private customer workflows or prove undeclared paths.

  1. ReportedSubmitted by the source
  2. RecordedSequence and integrity retained
  3. EnforcedControlled boundary established
  4. Outcome verifiedObserved through a declared read path
  5. Independently reviewedSeparate reviewer attested
Strength is dimensional. Legacy and partial paths are never silently upgraded.
Before release
Agent capability evaluationpassing

MCP servers, exposed tools, policy behavior, and runtime traces.

1 records100%
Before release
Release regression checksfailing

Browser and computer-use agents under adversarial UI and network faults.

9 records44%
After release
Runtime control and outcome verificationpassing

High-risk live actions requiring approval, verification, and audit.

1 records100%
11matching runs
Corpus records11Accumulated evidence rows
Pass rate55%5 failed records
Failure patterns10Top grouped failures
Taxonomy reviewed2/141 corrected labels
Review coverage14%50% reviewed-label precision
Last generatedJul 15, 04:11 AMMonitor snapshot timestamp

Evaluation coverage

Lifecycle evidence
Release regression checks4/9 passed
Agent capability evaluation1/1 passed
Runtime control and outcome verification1/1 passed

Top Failure Pattern

Corpus
Refund page reported a server errorNo console errors should be recorded1 occurrence

Recent Runs

11 matching records
ProductFaultStatusEvidenceTime

Run Inspector

Jun 25, 09:16 AM
PassedRuntime control and outcome verification
Agent
witnora-public-demo
Version
unversioned
Product
Runtime control and outcome verification
Scenario
-
Score
100
Evidence
12
Primary failure

No failure recorded for this run.

Evidence detail

Evidence Console

Published sample files
1 original files · 9 recorded events · 1 checks

Open Runs to read the event timeline, compare expected and observed results, and download available source files in this browser.

Local demo only. No real payments, emails, vendor portals, credentials, or production systems are used.

Failure Taxonomy Review

Copy command for local corpus
This run has no failure taxonomy labels to review.