§ settings / ai quality
AI quality
LIVE QUALITY SIGNALS
daily rollup over the last 30 days
Edit distance
n/a
average word-level delta
No rollup data
Citation validity
n/a
drafts with valid refs
No rollup data
Sent unchanged
n/a
accepted without edits
No rollup data
Refund quarantine
0
pending human approval
No rollup data
COVE INTERVENTIONS
high-risk reply verification over the last 7 days
Interventions / week
0
sensitive replies revised before review
Latency added
n/a
p95 n/a / target <2s
Removed claim patterns
LEARNING
human review signals used to improve future drafts
Learning
0
recent flagged drafts queued for review
Accepted unchanged
n/a
latest daily acceptance signal
Edit feedback
n/a
average word-level delta from agent edits
SELF-IMPROVEMENT RECOMMENDATIONS
high-edit intents converted into reviewable AI tuning actions
RERANKER
Voyage rerank controls for knowledge retrieval
ENABLE VOYAGE RERANK
Applies a cross-encoder pass after hybrid retrieval.
Enabled
MODEL
Use lite for lower latency; full rerank-2 for stricter ordering.
TOP K
Number of reranked candidates to keep.
SCORE THRESHOLD
Drop reranked results below this relevance score.
ADVERSARIAL TESTING
stress-test AI guardrails against out-of-kb, policy invention, and prompt injection
Adversarial eval dashboard
Run the full adversarial suite and export results.
AUTONOMY
end-to-end resolution loop, safety gates, and audit feed
Autonomous resolution controls
Configure workspace defaults, persona overrides, red flags, and pause controls.
ROLLUP TABLE
draft actions, citation validity, refunds, and escalations
| Date | Drafts | Edited | Discarded | Edit dist. | Citations | Refunds | Escalations |
|---|
RECENT FLAGGED DRAFTS
discarded drafts and edits with 50% or higher distance
| Ticket | Action | Edit dist. | Confidence | Model | Created |
|---|