We benchmarked coding agents on our own internal tasks at Databricks and learned a lot!
There are many surprising opportunities to lower cost and increase quality, and many models including open source ones are truly competitive now. 🧵
@jem_arnold Is this right then?
1. **Proportion of Time Completed in Final Stage:**
Time proportion = **129 / 300 ≈ 0.43**.
2. **Contribution of Final Stage to WPeak:**
Contribution = **0.43 × 45W ≈ 19.35W**.
3. **WPeak:**
WPeak = **355W + 19.35W ≈ 374.35W**.
@jem_arnold That’s my legs dying and me turning the ERG down to finish the interval. I notice actually I consistently bonk in that 90-180 second range.
@jem_arnold I had a go at doing the test today. So I got through the 4.0w/kg step but failed halfway through the 4.5w/kg.
So my WPeak is 355W? If you could help me interpret what I got that’ll be great.
@JeroenSwart@ToutpourmaSante@peter__leo@science2sport@WendzH@Pelotrain I recently switched from riding on 170mm to 175mm long cranks as I wanted to lean more into torque.
Since changing, my legs are getting nuked way sooner now, and my power in general has taken a dive. Is this to be expected?
@Spragg_Perform@JoeLavo@Pelotrain@Alex_Welburn @AndyTurner132 Nice paragraph 👍 the only bit missing IMO is why CP can predict over a range of power values in the severe domain while FTP can’t