Administration and FinOps program (FINOPS-001 to FINOPS-005)

Fabric Desktop by KratosBI helps developers understand, predict and avoid Fabric capacity use without becoming a copy of the Fabric admin portal or the Capacity Metrics app. This program adds evidence to what the app already shows: where each CU and cost number comes from, what capacity a workload would run on, what is risky before a cloud run, and when this PC is the better place to run.

Rules for every PR in this program:

No invented CU, savings, billing or performance numbers. Each figure is measured, estimated, scenario or unknown (cost-evidence.md).
Ordinary developers never need Fabric administrator rights. No capacity changes and no cloud runs happen on their own.
Only documented Microsoft APIs. The Capacity Metrics semantic model is not a production dependency.
No embedded AI; external tools reach these services through the AI integration thread's MCP and local API.
No Git changes (the Git-native development program owns Git, deployment and validation).

Audit (2026-10-17, main at 4079a41)#

PRs #5, #32, #58 and #94 are merged and present on main.

Feature	Partial	Missing	Evidence	Proposed action
FINOPS-001 Measurement and cost evidence	✓		`costs.py` estimates CU from run time x configured CUs and prices it; Monitor > Cost (`app/ui/estate.js` `costBody`) labels the whole panel "estimates" in a footnote. `scorecard.py` (Prove) already labels its figures measured, estimated or not available; side by side shows "(estimate)". Gaps: no shared schema; a metadata run without a run time counted as 0 CU (`seconds or 0`); Fabric jobs still running were dropped from the panel; the "saved" figure is a counterfactual but read like a measurement; no way to see how a number was worked out.	Shared `measurement.py`; `costs.py` emits classified measurements; Monitor > Cost shows a badge per figure and a "how this was worked out" dialog. PR 1 (this PR).
FINOPS-002 Capacity inventory and health	✓		`fabric_api.list_capacities` (paged; optional Capacity.Read.All, never prompted from Monitor); `readiness._capacity_check` (Active or not); `estate_service._env_capacities` (name and SKU per environment, cached without a timestamp); `FabricError` explains CapacityNotActive (#5); environments have a "Runs on" choice (#25, #32); run location shown (#58). Missing: region, a per-environment status with its observation time, stale marking, "no access" vs "not found", capacity context in the environment selector.	Extend `_env_capacities` into a capacity observation service with timestamps and access states; read-only summary in the environment selector and Monitor.
FINOPS-003 Measured CU import and reconciliation		✓	Nothing imports observed usage. Workloads plan item 62 records that there is no public per-capacity metrics API for non-admins.	Source-compatibility decision first (which Microsoft sources are supported and for whom), then a provider interface with a user-supplied export as the first provider. Likely a stop-and-ask point for Chris.
FINOPS-004 Predeployment risk analyzer	✓		`workload_service.workload_preflight` (Workload program), `meta_preflight` and the memory guard (#69), readiness checks, the `allow_fabric_runs` policy (#90), CapacityNotActive handling. Deployment validation (FD-113) is the Git program's.	A risk service that composes these checks into PASS, WARNING, BLOCKED or UNKNOWN with evidence, called from the existing preflight rather than a new one.
FINOPS-005 Local vs Fabric advisor	✓		Side by side history (`side_by_side.py`), Prove scorecards and receipts, notebook compatibility analysis, "Runs on" per notebook (#29). Prove's run planner (item 31) and placement recommendation are not built yet.	Build the advisor on Prove's receipts and side-by-side history, near the existing "Runs on" control; agree the split with the Prove program first.

Overlap with other workstreams#

Workstream (thread)	Owns	How FinOps uses it
Workload management workspace	Workload identity (`workloads/model.py` ids), execution history, operations view, Monitor and Home, cost allocation (items 66 to 68)	Measurements carry its workload ids. Its cost allocation builds on the classifications here. Monitor and Home edits are announced to it first.
Prove this notebook	Receipts, scorecards, side by side, profiler and cost model (FD-119)	Scorecards already use measured, estimated and not available. Receipts are Prove's schema: a classification field is proposed to Prove, not added here.
Git-native development (DevOps)	Deployment, promotion, validation (FD-113, FD-120)	FINOPS-004 reads its validation results; nothing here changes Git.
SQL development	SQL query performance	Not used by FINOPS-001 to 005.
Enterprise governance and readiness	Policies (#90 policy keys), audit log (#89)	FINOPS-004 reads workspace policies; overrides are recorded through the audit log.
Product management	Feature inventory in `docs/product/`	FINOPS ids are registered there by that thread.

PRs#

PR	Feature	Status
1	FINOPS-001 Unified CU measurement and cost evidence framework	Merged (#117)
2	FINOPS-002 Capacity inventory and health diagnostics (capacity.md)	Merged (#123)
3	FINOPS-003 Measured CU import and reconciliation foundation (usage-import.md)	Open for review (#153)
3b	FINOPS-003 second provider: Azure Cost Management exports, billed per capacity (usage-import.md)	Open for review
4	FINOPS-004 Predeployment capacity and CU risk analyzer, part 1: the risk service (risk-check.md)	Open for review
4b	FINOPS-004 part 2: findings in the run dialogs, acknowledgement, settings (risk-check.md)	Open for review
5	FINOPS-005 Local vs Fabric execution advisor	Not started

FINOPS-003 usage source decision#

Researched on Microsoft Learn on 2026-10-17; Chris chose "both, Capacity Metrics app first" on 2026-10-18. Nothing was called on a tenant.

Source	Supported for this use?	Who can get it	Granularity	Decision
Capacity Metrics app semantic model (DAX or XMLA)	No: "only supported for use by the reports provided in the app"	Capacity admins	Item and operation	Never used
Capacity Metrics app, exported data (CSV)	Yes, as a file; exports can be sampled	Capacity admins, who can hand the file over	Per operation with start, end and Total CU (s) (Timepoint page), or per item over a period (Compute page)	First provider (FINOPS-003)
Azure Cost Management exports	Yes	Cost Management Reader on the capacity's subscription	Per capacity, meter and day, in CU hours and money	Second provider (FINOPS-003b), for billed totals; never matched to runs
Chargeback app export	Yes, as a file; its semantic model has the same rule	Capacity admins	Per workspace, item, user and day	Not planned
Real-Time hub capacity overview events	Documented, but needs a capacity admin or contributor and an Eventstream that uses capacity	Capacity admins or contributors	Whole capacity every 30 seconds	Not used
Fabric REST job API	Yes	Every developer	Job times and status, no CU	Already used for estimates

So no supported live API gives a developer measured CU per run: measured CU comes from files a capacity admin or finance exports. The import is provider independent (sidecar/fabric_desktop/usage/), keeps everything on the PC, and never needs admin rights from the developer.