Roast 13 — Ethiopia Guji Karume (Natural)

Date: 2026-02-01

Context

This roast was recorded in real time and narrated as it happened, not as an instructional video, but as a “fly on the wall” capture of decision-making under uncertainty.

The coffee is an Ethiopian Guji Karume natural. The roaster is a Gene Café CBR-301. The intent was not to chase a specific profile, but to continue testing how reliable first crack actually is as a signal on naturals, and what I do when it isn’t clear.

This roast is paired with a video where I talk to myself throughout the process.


What the data says

  • Batch size: 250 g
  • End weight: 213.2 g
  • Weight loss: 14.7%
  • Set temperature: 250 °C
  • Total roast time: 12:57
  • Fan behaviour: stepped from level 2 → 3 at ~4:04 (~202 °C)

The JSON log records a crack flag at 11:23, but that timestamp needs context.


What I thought I heard vs what actually happened

I pressed the crack button before crack was clearly established.

Crack did happen, it was real and audible, but the rolling, confident crack followed roughly 10 seconds after I pressed the button. In other words, the button marked awareness, not onset.

This wasn’t a mistake so much as a familiar pattern emerging:

  • crack recognition precedes crack confirmation
  • especially on naturals
  • especially on this system

The decision to cool was made based on time, aroma, colour progression, and restraint, not on a single sound cue.


Development (how I’m interpreting it)

Measured strictly from the crack flag, development looks longer than intended.
Measured from the point where crack was clearly rolling, development sits closer to ~1:20–1:25, which was the intended window.

This gap matters. It reinforces that:

  • first crack is not a moment
  • the crack button is not a ground truth
  • development decisions are bounded by intent, not timestamps

What I’m taking forward

This roast reinforces a working assumption I’ve been circling for a while but haven’t yet promoted to the Guide:

  • On naturals, first crack is a range, not an event
  • Clicking the crack button marks recognition, not certainty
  • Time, aroma, colour, and weight loss remain more reliable anchors

This isn’t a conclusion yet, just a pattern becoming harder to ignore.