Telemetry — MANDATORY. Every api.fabric.microsoft.com call must carry x-ms-fabric-skill: e2e-fabric-cost-estimation (az rest: --headers "x-ms-fabric-skill=e2e-fabric-cost-estimation"), including every LRO poll, fabric_lro and retry. Snippets omit it — add it anyway.

CRITICAL NOTES

To find the workspace details (including its ID) from workspace name: list all workspaces and, then, use JMESPath filtering

To find the item details (including its ID) from workspace ID, item type, and item name: list all items of that type in that workspace and, then, use JMESPath filtering

Pricing is region-specific. Always resolve the capacity's Azure region (via the Fabric REST GET /v1/capacities, which returns region/sku/state, or via ARM) or ask the user before any price lookup, and state the region with every quoted figure.

🔴 MANDATORY LIVE-PRICING FETCH — whenever your answer contains a dollar figure. Before presenting any dollar amount, cost, break-even point, billing-mode comparison, RI-vs-PAYG analysis, migration cost, or cost worksheet, you must run a shell command (bash/powershell) that calls the Azure Retail Prices API at https://prices.azure.com/api/retail/prices. This applies without exception to every priced answer — Databricks/Synapse migration cost, Autoscale vs. base-SKU break-even, RI-vs-PAYG break-even, cost worksheets, and billing-mode strategy. Fetch the live per-CU-hour PAYG rate (priceType eq 'Consumption', * Capacity Usage CU meters), the reservation term totals, and the autoscale for Spark Capacity Usage CU rate first, then compute. Never answer a priced question from memorized or hardcoded rates, and never reason about break-even purely from formulas. If the meter lookup returns no rows, surface that to the user — never silently fall back to a hardcoded rate (e.g., do not assume $0.18/CU-hr for Autoscale; the Autoscale meter is a distinct rate that must be fetched). Exception — pure capacity sizing: a question answered purely in capacity units with no dollar figure (e.g., "which SKU fits 80 CUs?" or "how many CUs does a P2 map to?") is CU math, not pricing, and does not require a fetch. The moment you attach a dollar amount to that sizing answer, the fetch becomes mandatory.

🟠 CLARIFY FIRST, THEN ACT — do not assume defaults. A priced request needs two inputs before you can fetch or compute: the Azure region and the workload profile (e.g. CU-hours/day, node/job details, or the source cluster sizes for a migration). If either is missing, your first response must ask the user for the missing input(s) and STOP — do not pick a default region (never assume East US or any other region to "get started"), do not call the pricing API, and do not produce estimates from assumed values. Only after the user supplies the missing inputs do you fetch live prices and compute. Asking is the correct behavior even though the mandatory-fetch rule applies to the eventual priced answer.

🔴 AUTOSCALE RATE IS A DISTINCT METER — never reuse the base/PAYG rate for it. The autoscale for Spark Capacity Usage CU meter is a separate price from the base * Capacity Usage CU PAYG rate. You must fetch it with its own API call and read the returned retailPrice. It is a correctness bug to assign the base/PAYG rate (or a memorized figure such as 0.18) to an autoscale variable — e.g. autoscaleRate = paygRate or autoscaleRate = 0.18 is forbidden. If the autoscale meter lookup returns no rows, list all Fabric Spark meters for the region and surface that to the user; never substitute the base rate.