Naturally, I Called It the TPS Report

Spent the last day or so building a status page for the Lumbergh pipeline. Given that the whole system is themed after Office Space, there was exactly one thing it could be called.

The TPS Report.

What it shows

The main view is every intake in the pipeline and what’s happening to it:

• What stage it’s in. Standard intakes get the full planning treatment — decomposition, review, the works. Fast-lane intakes skip straight to Claude Code and QA. Seeing which lane something took, at a glance, turns out to matter a lot.
• Time spent. Per stage, not just overall.
• Rounds of QA. How many times the reviewer sent it back before it passed.
• Overall status.

Click any one of them and you get the detail for that stage — plus the original intake text, so I can see what we actually asked for. That last bit sounds minor and isn’t. Most of the times I’ve looked at a result and thought “that’s wrong,” the honest answer was that the request was wrong, and the fastest way to know is to have the ask sitting right there next to the outcome.

Feeding it by hand

I added a Submit Intake form as well. Normally Claude files intakes into the pipeline itself, which is the point. But being able to walk up and hand it a job directly is worth having — for testing, for the one-off, and for the times I want to watch a specific thing move through.

The gauges

Across the top: CPU, GPU, and memory use, each with a thirty-minute sparkline running behind the number, plus disk space and GPU temperature.

Sparklines behind live numbers is a small trick and I recommend it to anyone building an ops dashboard. A number tells you where you are. A number with half an hour of shape behind it tells you where you’re going, and that’s usually the question you actually had. “GPU at 80%” means nothing on its own. “GPU at 80% and climbing steadily for ten minutes” means something.

Everything’s running on Sparky — the DGX Spark sitting three feet from my coffee — so those aren’t cloud metrics from a vendor’s console. That’s the actual box, doing the actual work.

Why this was worth a day and a half

Because before it existed, the answer to “what is the pipeline doing right now” was read the logs and reconstruct it. Now it’s look.

That sounds like a convenience. It isn’t. When checking on something is expensive, you check less, and you find out about problems later and from worse evidence. When it’s free, you glance at it on the way past with a cup of coffee — and you catch things you weren’t looking for.

A stats page with better dashboard analytics is next.

The projects, experience and opinions here are mine. AI helped me turn my notes and build records into this piece and polished it for Cairoglyphics.ai.