Aule AI Metric Contract Lab

Step one

Name the decision before the chart.

A useful metric changes a real decision. Start with the question, owner, action, and harm if the number is wrong.

CompanyMetric nameMetric owner Business questionDecision owner Decision this metric supportsReview rhythm Action thresholdMaximum decision impact What goes wrong if the number is wrong?

Saved on this device

Step two

Write the rule a machine can follow.

Use exact fields, time rules, filters, and exclusions. If a data analyst cannot reproduce the number from this page, the definition is not finished.

Plain-language definition Exact calculationUnit GrainTime windowTime zone Authoritative source tables or reportsRefresh promise Included recordsData steward ExclusionsVersion Known edge cases

Why this matters: OpenAI says its Data agent can use metric definitions, custom calculations, and data relationships from semantic layers. Its setup guidance strongly recommends an authoritative semantic layer and tells users to check the source, period, filters, and metric definition before relying on a result.

Step three

Decide how the answer earns trust.

Reconcile the metric against a known result, define the acceptable difference, and set a sharing boundary before publishing.

Known comparison report or queryComparison period Expected valueAllowed differenceReconciliation owner Missing or stale data ruleAllowed audience Required caveat in the dashboardData class

Publishing check: OpenAI’s help article says data used in an analysis is copied into a published ChatGPT Site. Review the copied data and the audience before sharing. Existing source permissions do not replace a publishing decision.

Step four

Your AI metric contract.

This is a working control record. Approve it with the business owner and data steward, then test it against the actual connected systems.

0%ready

Contract laneDefinition not ready

Named fields0/15

Proof controls0/6

Metric version1.0

Release gaps

What must close first

Owner plan

Next seven days

Authoritative metric record

Owner

Decision owner

Grain + window

Audience

Need to turn the contract into a measured workflow?Bring Aule the definition, source, proof result, and the decision it must support.

Review the metric workflow ↗

Sources, evidence limits, and scoring method

OpenAI: Now everyone can put data to work, September 10, 2026. OpenAI says the Data agent in ChatGPT Work can connect to approved data, use business definitions and semantic layers, investigate questions, and create interactive dashboards.
OpenAI Help: Using the Data plugin in ChatGPT Work and Codex, updated September 10, 2026. This is the source for setup guidance, result-verification checks, connected-account permissions, and the note that analysis data is copied into a published ChatGPT Site.
ChatGPT Business release notes, September 10, 2026.

Evidence limit: these sources describe product capabilities and recommended practices. They do not publish an independent accuracy rate or prove that any local metric is correct. This tool does not connect to ChatGPT, a warehouse, or a BI system. It creates a buyer-controlled contract and test plan.

Scoring: 15 required definition fields count equally for 60% of the score. Six proof and sharing controls count equally for 40%, with Partial worth half credit. The lane is blocked when the formula, source, owner, baseline, or sharing review is missing. User-entered values and scores are planning evidence, not audited results.