2.35.0

A total of 168 new calibration models are integrated into the risk engine. These models significantly enhance fraud detection configurability. Customers can now select their preferred accuracy range for fraud scoring. This enables on-the-fly swapping of risk engines tailored to individual needs.

This flexibility empowers customers to fine-tune fraud detection thresholds independently, improving responsiveness and reducing operational overhead.

A new UI feature Correlated Engagement Counter is added that displays the number of engagements linked by the same correlation identifier.

A tooltip is added to the risk engine calibration model dropdown to improve usability and clarity for users interacting with the configuration page.

Issue Description	Issue ID
In web agent console, the OPT_OUT action isn’t showing correct date during transfer.	5532033

When using multiple tabs in the Gatekeeper web admin portal, users may be prompted to re-authenticate periodically, typically every few minutes.

If the fraudster was deleted but was still in the cache, it’s possible that the export services might still show it up as empty string. However, when the context-manager tries to get gk_fraudster_id by gk_voiceprint_profile_id, it gets NOT_FOUND as the voiceprint was deleted.

In the Gatekeeper web admin portal, when you add comma to the autocomplete in the search box, it’s used as a separator (like space), and search of only a comma returns no results.

When Contact Center Flow version is set to v2 and Detect Fraudsters is enabled on enroll, the Detect Fraudster widget doesn’t show a pass label.

WAV header duration value is unreliable when fetching media through GetMedia API, GetMediaUnmasked API, or through /media URL obtained from the PublishMedia or PublishUnmaskedMedia API.

Use duration provided by the GetMediaInfo API instead of relying on the duration of the WAV header.

In the Web Agent Console miniature view, language prompts aren’t localized to match the miniature view.

In Web Agent Console, some Dutch prompts aren’t localized, and they default to the English locale.

Any NVSL frameworks at 16 kHz don’t work, as Gatekeeper doesn’t support 16 kHz audio yet:

• FACTORY_TI_16kHz_10
• FACTORY_TI_16kHz_9
• FACTORY_TI_16kHz_8_2
• FACTORY_TI_16kHz_8
• FACTORY_TI_16kHz_7_2

Rarely it may happen that a post call session is isolated in a separate engagement and not aggregated with the other live sessions.

DEMO and Bootstrap risk engine calibration models might not work properly and they show poor accuracy. Use these models only for non-production use.

Gatekeeper Web administrator applications don’t officially support Private Browsing Mode. Without them, the Console may display an authorization error. This isn’t required for agents authorized for a single scope.

The autocomplete for first name and last name doesn’t work when using special character, such as ‘@’ or ‘_’.

Web Admin Portal has a time zone selection issue related to browser updates.

Kibana Dashboard – The Generate PDF feature doesn’t work when Optimize for Printing is enabled.

A Clustering or Backward Search action may time out due to the memory required for media collection exceeding configured limits. This is likely due to a collected engagement containing more than an expected number of sessions but could have another cause.

Clustering and Backward Search have the following limitations:

• No estimations available for media collection.
• Clustering and Backward Search require manual scaling.

Audio streaming for transcription is limited to one realtime (this value isn’t configurable).

For Stereo audio, the redacted overlay interval applies on both channels in the audio viewer.

Realtime redaction isn’t supported when Transcribe API is used.