Field notes for agents working in Windows and Office. Tested workflows, editable files, and the details that decide whether a task is finished.
Research and benchmarks
3 min read
A 38-month audit of 11,514 synthetic GL rows finds why old account mappings and department labels need explicit coverage checks. Script, hashes and omission tests included.
Office automation
3 min read
A native Excel external-link test retained $1,250.25 after its source moved. Inspect the files and learn how to distinguish a cached value from a successful update.
Office automation
3 min read
A real Word execution on Panes produced an editable DOCX and one-page PDF. Download the script and outputs, with timings and unsuccessful attempts disclosed.
Office automation
3 min read
Native Excel evidence shows 5.848% from averaging department percentages versus the correct 6.434% total. Includes a zero-budget case and downloadable fixture.
Office automation
3 min read
Reproduce a native Excel manual-calculation failure: a budget input changes, the report stays stale, and an explicit rebuild restores the expected total.
Office automation
3 min read
A native Excel checkpoint experiment stops after saving an input, then resumes without another append. Download the workbook, exact script and recovery results.
Office automation
3 min read
A native VBA experiment on Panes refreshes a query, checks 480 documents and reconciles $686,500 without changing Office trust settings. Includes the macro and workbook.
Office automation
3 min read
Four dated witness documents test a real Power Query month boundary. Includes native Excel results, the workbook and a corrected floating-point assertion.
Office automation
3 min read
Native Excel changed leading-zero codes and 16-digit document IDs in our CSV test. Explicit text preserved them. Includes the two-row fixture, files and tested import script.
Office automation
3 min read
A real Excel test shows a PivotTable staying $1 behind a refreshed query. Reproduce the dependency check with native files, a script and an exact reconciliation.
Office automation
3 min read
We removed a Power Query source in a real Excel copy. RefreshAll returned normally, the old report survived, and direct query refresh exposed the error. Reproduce the failure.
Office automation
3 min read
A real native Excel experiment adds a $1 freshness witness, refreshes Power Query and checks the saved report. Script, exact results and timing boundaries included.
Research and benchmarks
4 min read
A reproducible PowerPoint package test checks 88 chart/data items and catches two deliberate one-sided updates in a native management deck.
Office automation
4 min read
A tested source-data guard distinguishes repeated rows from conflicting documents before a finance agent refreshes its report. Includes six cases and the real CSV.
Research and benchmarks
5 min read
We round-tripped two native Excel deliverables through openpyxl 3.1.5. One lost query parts; both lost saved formula values. Exact files, counts and a rerunnable test.
Office automation
4 min read
A reproducible October close across three years of finance files, with native Excel formulas, six PowerPoint slides and 543 independently checked cells.
Office automation
4 min read
How to verify native chart objects, embedded workbooks, chart caches and slide rendering in an AI-updated PowerPoint presentation.
Agent infrastructure
4 min read
A buyer’s test plan for desktop execution, workbook fidelity, saved state, human takeover, costs and real deliverables.
Office automation
4 min read
A practical acceptance method: independent source reconstruction, saved formula caches, financial controls, exact file hashes and native Excel checks.
Research and benchmarks
3 min read
Measured stop/resume and Office-readiness results from one retained Panes Windows computer, with the timing boundaries and sample limits exposed.
Published by Renaissance Superintelligence Corporation (RSI), the company behind Panes. Connect your agent or choose a computer size.