The badge on each card is the
literal scope of the credentials we hold in that phase.
READ-ONLY
Days 1–10 · Assessment
Connect read-only
credentials. Ten analysis agents scan utilization, idle time, sizing,
and orphaned resources against your live metrics.
You get:
a quantified waste report — every finding backed by your own
timestamps and utilization samples, ranked by monthly dollars.
Write credentials are rejected at intake. Mutations are blocked at the
connector layer — not by policy, by code.
SHADOW
Days 10–20 · Shadow mode
The full decision
loop runs — proposals, deduplication, all ten safety checks — but
nothing executes. Every would-be action is logged at the exact moment
it would have fired.
You get:
a replay of your own ten days with our actions overlaid:
"23 actions, est. $X saved, 2 blocked by your
guardrails."
Watch decisions stream live in your dashboard before any of them are real.
SUPERVISED
Days 20–30 · Supervised actions
You enable two or
three reversible, low-blast-radius action types — auto-suspend tuning,
zombie termination. You approve each action individually.
You get:
verified savings — each action's cluster cost before and after, from
your own billing data, at list price.
These are measured numbers. Not projections.