[AI Test Toolkit] Export full eval run results (cost, time, dataset)

[onbuyuka]

Summary

The Export Results / Download Test Summary action on the AI Eval log pages (report 149030 "AIT Test Summary") did not export the full run results. It only emitted: codeunit/test name, status, accuracy, turns, input, output, and error. The reporter needs the cost and timing data to analyse a run.

Root cause

The export passes the complete filtered AIT Log Entry view to the report (no row/marked-row limiting) — so the limitation is purely the report's column set. The dataset omitted fields that already exist on the table and are shown on screen:

• Cost/time: Tokens Consumed, Duration (ms), Start Time, End Time
• Identity: Version, Tag, Test Input Group Code, Test Input Code, Test Input Description, Run ID, Entry No.

Fix

Add those columns to the dataitem(Results; "AIT Log Entry") dataset. A modern Excel-layout report writes every dataset column into the workbook's Data sheet, so the new fields are exported in the raw data without changing the formatted .xlsx layout.

Follow-up (optional, not in this PR)

If the formatted/pretty sheet of AITestSummary.xlsx should also display the new columns, the layout file can be regenerated from the updated report (Report → Run → Edit layout). The raw Data sheet already carries them.

Impact / risk

Additive only — no codeunit/page changes; the export already passed the full filtered view. Field names verified against AITLogEntry.Table.al.

Verification

• Verified each added field exists on AIT Log Entry (fields 1, 4, 5, 9, 13, 15, 18, 24, 25, 26, 50).
• ⚠️ Not runtime-verified on an NST (AL does not run locally here); relying on PR CI for compile. A manual export check is recommended to confirm the Data sheet content.

Linked work item: AB#633442

🤖 Generated with Claude Code

[onbuyuka]

CI note: the Clean build failure is a known BCApps incremental-build infra issue — not this change

The ... (Clean) build failure is not caused by this PR. The incremental Clean build compiles the modified AI Test Toolkit app against the System Application bundled in the BC platform artifact (sandbox/29.0.51094.0) instead of the freshly-built System Application. That artifact predates #8381, which added Agent Task.OpenAgentTaskLogEntries / GetStepsDoneCount (called by AITAgentLogEntryPart.Page.al), so the compiler reports AL0132 ... does not contain a definition for ....

Evidence it is infrastructure, not this change:

• The Default build of this same app passes (compiles against fresh symbols).
• main`s full build passes (System Application compiled from source).
• The two methods exist, unwrapped (present under CLEAN29), in current main source and in the last-green baseline a97db24da.
• PRs that modify System Application (e.g. [Azure AD Plan] Update access for GDAP users who were previously guests #8441, [Retention Policy] Keep record marks for indirect-permission subset deletes #8469) pass Clean because they rebuild it from source; only AI-Test-Toolkit-only PRs hit this.
• Build log shows the Clean compile resolving Microsoft_System Application.app from the BcContainerHelper compiler-folder cache (the platform artifact), not the freshly-built 29.0.74476.0.

No code change is appropriate (a #if not CLEAN29 guard on non-obsolete methods would be wrong). This goes green once the platform-artifact pin advances past #8381. Holding for that.