Proof of concept · synthetic estate
Cloud cost optimization that knows how your estate is connected
A cloud bill says what everything costs. It can't say whether an "unused" address still has a DNS record pointing at it, who owns an untagged database, or which backups production would restore from. Put AWS, Azure and GCP into one graph and those become questions you can ask.
- Resources
- …
- Monthly spend
- …
- Findings
- …
Why there are two numbers
Gross is what every finding would save if every one were acted on. It is an upper bound, and it is the right number for sizing the opportunity.
Not every finding should be acted on. Some snapshots are kept on purpose. Some development teams work late. Some workloads end before a commitment would. So every kind of finding carries a confidence: how often a finding of that kind turns out to be genuinely worth acting on — not how sure the query is that it matched.
Confidence-weighted multiplies each finding's saving by its kind's confidence and adds them up. It is the number to put in a forecast.
The confidences are judgements set for each kind of finding, not measured on this estate. On a real estate they should be tuned against what teams actually go on to do. You can try your own on the What if tab, with the other thresholds.
Confidence by kind of finding
An unattached address is almost always waste; moving a workload to stop its data transfer often is not worth it.
Questions a bill can't answer
Each needs more than the bill, and in each the graph changes the answer: what depends on what, what sits behind one balancer, what a backup protects. Click through to the evidence.
Savings by the evidence they need
Every finding, grouped by the least an estate has to supply to find it. Billing alone gets you started; the rest needs more.
How it works
Each stage reads files the previous one wrote, so the synthetic estate here can be swapped for real exports without touching anything downstream.
Billing
FOCUS 1.4 cost data from all three clouds, plus inventory, flow logs and usage metrics where an estate has them. Every source but billing is optional.
Graph
Resources, what they attach to and depend on, and who owns them, in an embedded graph database. No server; it runs on a laptop, offline.
Detectors
One set for every cloud: waste, commitments, rightsizing, schedules, storage tiers, data transfer. Each checks what would break before it recommends.
One set of numbers
The explorer, the HTML report, the API and the command line all read the same engine, so every figure matches and traces to a billing column.