Commissioning Docs That Track Themselves

Test results get logged the day the work happens. The agent keeps the tested-vs-outstanding count per string and block current, so closeout gaps surface while the crew is still on site — not during turnover.

The problem

Commissioning documentation is where solar closeouts go to die. Every string needs insulation-resistance (megger) results, every combiner and inverter has a checklist, and the results live wherever the tester left them: meter photos on someone's phone, a paper log in the job trailer, a spreadsheet last updated three weeks ago. Nobody knows the true tested-vs-outstanding count until the owner or the commissioning agent asks for the package.

Then the scramble starts. Strings that were tested but never recorded get re-tested. Strings that were never tested get found at the worst possible time — after demobilization, when sending a crew back costs real money and the punch list clock is already running.

How the agent workflow works

1. The job's testing scope is loaded once: every string, combiner, and inverter that requires a test or checklist, organized by block or roof section.
2. As testing happens, the technician sends the evidence to one place — a photo of the meter reading, a completed checklist form, or a short message with the string ID and result.
3. The agent reads each submission, matches it to the scope list, records the value, date, and tester, and marks that item complete. Readings outside the acceptable range are flagged the same day, while re-testing is still cheap.
4. A running status — tested, outstanding, flagged — stays current per string and per block, and a weekly summary goes to the superintendent and PM showing exactly what's left.
5. As closeout approaches, the agent assembles the documentation package from the logged records: results by string, checklists by equipment, exceptions listed rather than buried.

What you need to set it up

• The full testing scope in a structured list — string and equipment IDs exactly as they appear on the drawings and labels.
• Acceptance criteria in writing: minimum insulation-resistance values, required test voltage, what counts as a complete checklist.
• One submission channel testers will use (email, chat, or a simple form) and a naming habit for string IDs.
• A human QA owner who reviews flags and signs the final package — the agent assembles, a person certifies.

Where it breaks

The agent tracks records, not reality. If a string gets marked tested from a photo of the wrong meter screen, or a reading is transcribed from a similar string ID — B-114 versus B-141 is a real and common confusion — the tracker will show green on work that wasn't done. Meter photos in sunlight glare misread, and a checklist submitted complete is taken at face value; the agent cannot tell whether the torque was actually checked or the box was pencil-whipped. It also can't judge borderline results against conditions — a low reading on a wet morning means something different than the same number on a dry one. The tracker removes the paperwork gap. Field verification, spot re-tests, and the QA walk remain human work.

The bottom line

Worth running on any job with more strings than one person can hold in their head — which is most commercial work. The real return isn't the clerical time; it's learning about a missing test in week six instead of week sixteen, while the tester and the lift are still on site.