Insights
Insights reads the Ledger and tells you where your policies and your agents disagree with what you expected. A person decides every change.
What it shows
Section titled “What it shows”- Stats: actions, denials, approvals and verdicts by rule, tool or agent, over 7 or 30 days.
- Shadow: where a shadow policy would have decided differently from the enforced one, action by action.
- Anomalies: six detectors (
new_tool,volume_spike,off_hours,deny_spike,mismatch_spike,user_ref_burst) compare each hour with the agent’s last 14 days and open an episode with its evidence. Policies can readanomaly.activewhen an owner turns that on. - Replay: run any policy version over last week’s real actions and see what it would have decided. See Test a policy against last week.
Suggestions
Section titled “Suggestions”Insights proposes changes: loosen a rule whose approvals people always grant, tighten a rule that let a MISMATCH through, raise a rule’s verify level, add a budget. Each suggestion shows its evidence and a replay of the change over recent actions. Accepting one pushes a new policy version and runs it in shadow, never enforced. A person promotes it to enforce later.
$ deedglass insights suggestions list…Optional phrasing by a language model rewrites a suggestion’s sentence from aggregates only, under a monthly cap. Off by default.
What it doesn’t do
Section titled “What it doesn’t do”- Insights never activates or enforces a policy. Its code can create a version and run it in shadow; nothing else. A test checks that.
- It doesn’t learn from your data in a model. The detectors are fixed rules over your Ledger.
- Suggestions are only as good as the history they read. A week with no traffic suggests nothing.