The case for mainline Linux 

A vendor kernel ships fast. Then it stops. The CVEs pile up. The security team calls. The board respins.

We have shipped enough programs to know that the difference between a product that lands on time and one that slips by a quarter is rarely a single heroic decision. It is the accumulation of small ones, made early, defended quietly, and revisited only when the evidence demands it.

What the work actually looks like

Mainline is harder on day one and easier for the next five years. We upstream what we can, carry small patches when we must, and document the diff in a single file the next engineer can read.

In practice the schematic, the layout, the firmware skeleton and the test plan start in the same week. Not the same quarter. The same week. The team is small enough that one engineer can hold the whole picture in their head, and disciplined enough that nobody pretends to when they cannot.

Vendor kernels are convenient on day one and expensive on day five hundred.

Where it usually goes wrong

The failure mode we see most often is not technical. It is a handoff. A spec leaves one team without the context a second team needs to act on it. Two weeks later, something obvious has been missed, and the cost of fixing it has multiplied by ten.

• One owner per decision, named in writing.
• One review per gate, scheduled before the gate opens.
• One source of truth for the design, the BOM and the test plan.
• One number that says whether the program is on track this week.

It is the boring choice that ages well.

The quiet result

A program run this way feels boring from the inside. There are no rescue missions. There is no hero week. The board enters the lab, passes, and goes into production. The line calls once, to say the first thousand units shipped clean.

That is the work. It is not glamorous. It is the only way we know to build hardware that survives the year after launch, and the four years after that.