More precise handling of agent actions
We updated how the audit handles scrolling, unsupported actions and response limits. An unsupported action produces a technical failure, without a score for that agent.
Implemented · release not confirmed
We have tested the changes offline. A live release and its effect on earlier audits are not confirmed, so we cannot yet recommend repeating a scenario.
Which audits may be affected
Earlier recorded implementations using the listed agents may be affected. Missing or unrecognized versions remain unknown. No model was changed.
Version and agent details
- Known earlier implementations
- None: earlier implementations carried no version label, so older reports cannot be matched to this change.
- Implemented from
- Included in the first versioned implementation
- Agents
- ChatGPT, Gemini
Dates and status
- Update recorded
- Provider announcement
- Our implementation change; no provider announcement date claimed.
- Our verification
- · Offline tests
- Implementation
- Live release
- Not confirmed
- Effective for customer audits
- Not confirmed
Reference documentation
Current first-party documents reviewed on the dates below. They describe API or file-handling behavior; they do not confirm our live release.
If an older report has no version details, we cannot reliably match it to a change. These updates do not start an audit, use your balance or send you an email.