What I found
The line was not falling apart. Every spike in hours per unit lined up with a week we barely built anything. When a week ships 2 units, a normal load of fixed hours divides into a huge number. The metric was measuring volume, not how hard the crew worked.
What I built
A model that pairs weekly hours against units built for the full 2026 year to date, per assembly line and as a rollup, with a minimum-volume filter and a rolling-average option. The point was to stop anyone from quoting a three-unit week as if it were a trend.
How it works
Drag the minimum weekly volume. Weeks below the line drop out, and the headline range recomputes in place. Watch the band on the top chart collapse and the threshold line move across the units chart.
No minimum. Range about 39 to 463 hours per unit, roughly 12x, 30 of 30 weeks.
Raw, hours per unit runs from about 39 to 463 across the built weeks, a spread of roughly 12x. The spikes look alarming.
Hours per unit
Units built
Shown: total across all lines. The model also produces battery pack, EDU, general assembly, full vehicle, and HPC views, weekly, monthly, or quarterly.
View the weekly data
| Week ending | Hours per unit | Units built |
|---|---|---|
| Jan 17 | 45.2 | 30 |
| Jan 24 | 48.9 | 28 |
| Jan 31 | 38.6 | 35 |
| Feb 7 | 52.3 | 26 |
| Feb 14 | 41.7 | 33 |
| Feb 21 | 57.8 | 24 |
| Feb 28 | 63.1 | 22 |
| Mar 7 | 69.4 | 20 |
| Mar 14 | 74.6 | 18 |
| Mar 21 | 81.2 | 16 |
| Mar 28 | 79.0 | 14 |
| Apr 4 | 76.5 | 13 |
| Apr 11 | 60.3 | 21 |
| Apr 18 | 55.0 | 25 |
| Apr 25 | 78.9 | 17 |
| May 2 | 84.4 | 15 |
| May 9 | 82.0 | 12 |
| May 16 | 50.1 | 27 |
| May 23 | 142.5 | 8 |
| May 30 | 168.0 | 6 |
| Jun 6 | 158.7 | 6 |
| Jun 13 | 205.4 | 4 |
| Jun 20 | 223.9 | 4 |
| Jun 27 | n/a | 0 |
| Jul 4 | 271.6 | 3 |
| Jul 11 | 298.2 | 3 |
| Jul 18 | 187.3 | 5 |
| Jul 25 | 351.0 | 2 |
| Aug 1 | 462.8 | 2 |
| Aug 8 | 133.9 | 9 |
| Aug 15 | 96.5 | 11 |
Raw, hours per unit runs from about 39 to 463, a spread of roughly 12x. Set the minimum to 12 units and the range tightens to about 39 to 84, roughly 2x. Same crew, same plant. The swing was the denominator.
Impact
- The metric became something a director could act on, instead of a number that set off an alarm every low-volume week.
- It shipped with a rule attached: read it on a minimum volume or a rolling average before it goes in a deck.
- The same structure supports per-line views for battery pack, EDU, general assembly, full vehicle, and HPC, weekly, monthly, or quarterly.
The transferable lesson
A ratio is only as stable as its denominator. Before a per-unit metric goes near a meeting during a ramp, decide the minimum volume it is allowed to speak at. That one decision is the difference between a number people argue about and a number they can steer by.