Research cases

This archive contains research cases and a report-update scenario with retained drafts, reviews and revisions. The dedicated seven-task Toolkit evaluation set and other test materials are listed in the evaluation inventory. Existing cases can be reused for tests and repairs without outside task contributions or an additional outside reviewer. Astra can review in a context separate from the author.

APIs or browsers for AI agents

Original English research on choosing APIs, DOM automation, or visual interaction by action coverage, verified completion, recovery, permissions and cost.

Full report · English case archive · Sources · Reviews and revision

The article draws on 11 primary English sources read on 10 October 2026. The archive retains the genuine first draft, a separate-context review, the accepted migration-cost correction and the full current-version review. The hypothetical workflow and unmeasured cost model do not establish comparative performance or a general Toolkit quality advantage.

Office agent research in China

The historical Chinese case examines office agents, with information through September 9, 2026 and some dynamic documents checked on September 10.

Chinese report · Chinese case archive · English translation · English notes on the historical revisions

Updating an existing report

Coding-agent procurement report update: previous report, source checks, concrete revisions and preserved limits.