Compare the human concierge decision with the local AI Brain recommendation before any automation is trusted.
Awaiting human decision
Human decisionNo human decision saved yet
Shadow recommendationRun local triage first
Confidence-
Alignment-
Evidence:Local scores and status decide the recommendation.
Unlock rule:Repeated aligned decisions unlock assisted queue routing only, not live sends.
Case
Human
Shadow
Alignment
Anonymized import preview
Paste anonymized CSV rows, run aggregate shadow alignment, and create a founder review packet without mutating the live queue.
No import run yet
Local-only rule
Use anonymized rows only. No file upload, no provider call, no queue mutation, no member message, and no checkout.
Rows analyzed0
Average alignment-
Review gaps0
Safety gated0
Alignment by case type
Case type
Rows
Avg
Main shadow action
Disagreement themes
Recommendation:Run an anonymized import to generate a founder review packet.
Case
Human
Shadow
Alignment
Next review
Founder Review Digest
Turn shadow runs and aggregate reports into one local decision packet for script changes, autonomy stage, and approval-gated next actions.
No digest yet
Evidence score-
Autonomy stageManual concierge
Single-case evidence0 runs
Aggregate evidenceNo aggregate report
Recommendation:Generate a digest after at least one shadow run or aggregate import report.
Script / rule changes
Next local actions
Approval gates
Script / Rule Revision Queue
Convert digest recommendations into local draft updates for intake, safety, verification, clarity, and follow-up routing.
No revisions yet
Open drafts0
Local reviewed0
Safety changes0
Categories0
Local review rule: marking a revision reviewed only approves local rehearsal. Real scripts, sends, provider logic, or autonomy changes still require explicit approval.
Category
Proposed local revision
Status
Gate
Script Outcome Learning Loop
Measure whether locally reviewed revisions improve later alignment, safety clarity, retention signals, or evidence quality before automation graduates.
No outcome checks yet
Outcome checks0
Local checked0
Candidate wins0
Manual holds0
Outcome rule: local outcome checks are rehearsal evidence only. Real retention, safety, conversion, or automation outcomes require approved beta data handling.
Revision
Outcome signal
Evidence
Status
Next action
Autonomy ladder
1. Manual conciergeHuman reviews every case.Current2. Assisted queueRules rank cases. Human approves.Local next3. Shadow triageAI predicts routing beside humans.Gated4. Supervised remindersLow-risk reminders only after approval.Gated