Policy Lifecycle

A common question: "I created a policy from today's findings — what happens to next week's?" The short answer: a policy keeps working on every future scan, automatically, for everything it covers. This page explains exactly how.

A policy is a standing rule, not a one-time cleanup#

When you create a policy — including one Aevi suggests — CostLens stores a set of rules: waste type → action + thresholds + guardrails + scope. Those rules are not tied to the specific resources that existed when you created them. From then on, they're re-applied to whatever findings exist on each scan.

That's what makes it a policy rather than a one-off bulk fix: it governs the type of waste, so any future resource of that type is covered automatically — no need to re-create the policy or re-run Aevi.

When policies run#

Auto-fix runs as part of analysis, not the instant a finding is detected. CostLens re-analyses your accounts:

• On a schedule — roughly every 12 hours
• When you first connect an account
• Whenever you manually re-sync / re-analyse

So a newly detected finding is acted on at the next run after it appears — bounded by your cloud's billing lag plus the scan cadence, not instantly.

What happens to a new finding — by action#

For a finding that matches one of your policy rules:

Action	What happens to each new matching finding	Do you review it?
Off	Suppressed — not surfaced	—
Recommend	Surfaced as a recommendation	Manual apply
Approval	A fix is created and queued for your approval every time (per resource)	Yes, each one
Auto-fix	Applied automatically — if reversible, within guardrails, not excluded	No

Approval is always per resource — there's no "approve once, applies forever". Each new matching resource is its own decision (until a ramp graduates the check).

Guardrails are re-checked every run#

Even at Auto-fix, every candidate passes the same guardrails on each run: the max-$/fix cap, daily limits, maintenance window, minimum observation window, exclusions, protect-production, and the dependency check. And irreversible actions can never auto-apply — the strongest they reach is Approval, no matter how a policy is configured.

When a new type of waste appears#

This is the important one. Your policies cover the waste types you've given rules to. If a brand-new type of waste appears — say you start using a service CostLens hasn't been governing — its findings are surfaced as Recommendations (nothing automatic) until you add a rule for it.

Stale findings clean themselves up#

When the underlying problem goes away — a resource is deleted, or is no longer idle — the recommendation resolves automatically. Policies never act on outdated findings, and your impact numbers stay honest.

The recommended way to run policies#

• Start safe. Put new checks on Approval (or turn on Dry-run), and graduate to Auto-fix via the ramp once you trust them.
• Review periodically — monthly, or whenever your estate changes. Re-run Suggest with Aevi, or act on the coverage banner, so newly-emerged waste types get rules. Treat Aevi as a periodic advisor, not a set-and-forget autopilot.
• Watch the Activity tab. Planned on the next run shows exactly which resources will auto-fix or queue before it happens; the Recent Activity log and approvals show what happened.
• Use scope. Be aggressive in dev/test and conservative in production with separate, appropriately-scoped policies.

A month in the life#

• Day 0 — Aevi suggests a policy from 12 findings: Idle compute → Auto-fix, Unattached storage → Approval. You create it.
• Day 0, next run — the 12 findings are handled: idle compute auto-fixed (reversible), unattached storage queued for your approval.
• Day 9 — a new idle Kubernetes node group is detected. Next run: auto-fixed automatically under your standing Idle compute rule. No Aevi, no new policy.
• Day 20 — you adopt a new service that produces a new type of waste with no rule. It appears as a Recommendation, and the Configure tab flags it as not covered. One click on Update with Aevi adds a rule.
• Day 30 — after 5 clean auto-applies, an Approval-tier check with a ramp of 5 graduates to hands-off.