Skip to contentChangelog
- Added optional blind human review so qualification can be assessed before model verdicts are shown.
- Review items can now be snoozed with an expiry time while retaining their queue status.
- Reply notifications now cover all actionable inbound messages and retry failed deliveries.
- Added mobile-first creator applications with field validation, referral codes, and social sharing previews.
- Inbox polling, processing, and sending now emit monitoring heartbeats after successful cycles.
- Added inbox-health checks for classification backlogs, review delays, send failures, and duplicate sends.
- Campaign reporting now uses isolated workspace access instead of a shared account context.
- A failed client reporting sync no longer prevents collection for other workspaces.
- Discovery runs now retain and report partial progress after an external-service interruption.
- Added transient-response retries and download-size checks for cached profile images.
- Added multi-model qualification review with structured decisions and comparison scorecards.
- Added a human review workspace that prioritizes model disagreements and saves adjudications.
- Qualification review now runs in parallel, saves each completed decision, and resumes interrupted batches.
- Personalization runs now enforce a rolling daily spend limit and cancel pending work when the limit is reached.
- Creator names, biographies, and captions remain isolated from prompt instructions.
- Discovery restarts now count only newly saved profiles and stop when progress can no longer continue.
- Malformed verification responses now retry instead of terminating a worker.
- Qualification scorecards now separate random evaluation samples from targeted rare-segment samples.
- Added confidence intervals and an explicit insufficient-sample result for thin segments.
- Personalization checks now flag copied caption text and normalize generated punctuation before saving.
- Added disposable-email and mail-host checks before full contact verification.
- Referral reporting now records qualifying outreach automatically and deduplicates creator handles.
- Reply classification skips threads with no genuine inbound message.
- Added request and response validation around email operations.
- Added creator-qualification scorecards covering precision, calibration, and agreement with human reviewers.
- The reply sender now rechecks quality gates and returns failed, unedited drafts to review.
- Added persistent campaign scan checkpoints to detect unresolved fields across scheduled sends.
