The check set
118 checks, and every one of
them names something
A check is not a dashboard tile. It is a question with a threshold, an object it resolves to, an action, and a risk of taking that action — written down once so the answer does not depend on which practitioner is looking.
In one paragraph
What the check set is
Zephto runs 118 versioned checks against Microsoft Foundry, Databricks, Fabric and the Azure estate underneath them. Each check resolves to named objects — a deployment, a job, a capacity, a resource — and prices its finding against the rates on the customer’s own bill rather than list price.
The output is a workbook and a three-page executive summary, produced under a stated check-set version so a later run can separate what changed in the estate from what changed in the platform.
- FinOps practitioners
- Platform and landing-zone teams
- Data platform owners
- AI engineering leads
- Cloud financial analysts
- Finance business partners
52 checks
Where the growth is
Tokens, DBUs and Capacity Units are three abstractions over one question: is this thing doing work? Every check in this block resolves back to that question and then to a dollar figure.
Foundry and Azure OpenAI
PTU sizing against P95 demand, idle provisioned deployments, max_tokens inflating reserved capacity, prompt-prefix cache alignment, batch-eligible traffic on Global Standard, frontier models doing mini-model work, Data Zone premiums with no residency requirement behind them.
Databricks
All-Purpose clusters running scheduled jobs, auto-termination windows measured in hours, warehouse class against real query pattern, Photon where it does not pay, repeatedly failing jobs that bill anyway, DBCU commitment coverage, Standard workspaces facing the Premium upgrade.
Fabric
F-SKU sized against a peak that no longer happens, capacities never paused, reserved capacity paused in the belief it stops billing, throttling absorbed silently, background versus interactive placement against the smoothing window, OneLake growth, dev sharing a production capacity.
66 checks
The estate underneath
Classic infrastructure drifts too, and it is where most estates still hold the larger absolute number. Grouped the way your subscriptions actually are.
Boundaries
What it does, and what it will not do
What it does
- Resolves every finding to a named object your team can open
- Prices against the rates on your bill, and says so when a rate had to come from the public price list instead
- Reports theoretical and realistic separately, with the realization assumption stated
- Counts an opportunity once, even when three checks flag the same object
- Carries a version number and a changelog, so two runs are comparable
What it will not do
- Change anything in your estate — every grant is read-only
- Read prompt or completion content, notebooks, data tables or report contents
- Claim a saving it cannot trace to billed spend
- Sum a ceiling and a counted figure into one headline
- Charge a percentage of what it finds
The current set is v2026.08. What changed in each of the last four versions is on the home page changelog, and every report states the version it was produced under.
See it against your own estate
One subscription or workspace is enough to find out whether the check set has anything on you.