GLOSSARY

The terms that show up in the console.

Short, quotable definitions. Each term appears in at least one product or methodology section. When legal, sales, or a new onboard asks "what is X here?", the answer is on this page.

Realized Savings Ledger

Immutable record of every saving that cleared the four criteria in the method: approved CR, pre and post measurement, materially attributable delta, and recorded ledger row. The single source behind any public number.

Change Request (CR)

Artifact with why, savings projection, change scope, attached evidence, and assigned reviewers. Waits for two human approvers before any execution. No shortcut.

Confidence rung

Four levels adjustable per agent: suggest only, draft CR, auto-execute below $X, autonomous. No rung ever bypasses human dual-control approval.

Bidirectional anomaly

Outlier detection in both directions — spend rising or falling. A silent drop is often a dead workload, data gap, or tariff change no one noticed. Every anomaly is classified and stays in the inbox until validated.

Bidirectional rightsizing

A sizing recommendation in both directions: shrink an oversized resource and flag a starved one to grow before it degrades. Works across Azure, AWS and Google — compute and managed databases. The "up" direction raises spend, so it's framed as risk mitigation — it lands as review-only, carries a positive cost delta, and is never counted in the Realized Savings Ledger.

Variance PVM

Decomposition of the difference between two periods into seven buckets: Price, Quantity, Mix, Action, New, Retired, Data Quality. Each bucket comes with a written explanation.

Dual-control approval

Human approval by two distinct individuals before any execution. Non-negotiable, even at the autonomous rung. A principle inscribed in the manifesto.

Kill switch

Immediate-stop mechanism — global (takes down every agent) or per-agent (isolates one). The persistent variant also clears agent memory.

Read-only service principal

The Azure identity leancosts uses. No write permission on any resource. Every decision exits as a CR for your team to execute.