What makes CloudScout different
A research-grade agent loop, not a dashboard with alerts.
Three primitives no other cost tool ships. Each one closes a specific failure mode of dashboard-only cost tools, and each one has a test that breaks if the safeguard regresses.
time-machine
Onboard against your last 90 days, not against an empty dashboard.
Day 0: paste your CUR. CloudScout replays Scout + Advisor + Verifier across the last 90 days and shows what the ledger would have looked like. The first sale no longer requires faith.
Why it matters: removes the “trust me, savings will show up” gap that kills cost-tool deals.
counterfactual lab
Every advisory carries the “what if we hadn’t shipped this” delta.
For each advisory, the lab runs the same workload against the un-modified IaC for 30 days and compares CUR slices. Counterfactual cost is what AWS would have charged. Realized cost is what they did. The gap is what the ledger pays out on.
Why it matters: separates “notional savings” from “actually cheaper bill”. Auditors love it. Finance teams pay only on the gap.
self-eval ledger
The agents grade themselves in public, every cycle.
Scout calibration (30d Brier), Advisor expected-value vs realized, Verifier dispute-overturn rate — all written to agent_eval_events with a replay hash. When an agent regresses, the ledger says so before a customer notices.
Why it matters: an agent product that can’t prove it’s improving is just a chatbot with extra steps.