There’s a moment in this process where curiosity quietly turns into doubt.

After consolidating the roast data into a single CSV and looking at correlations between weight loss, peak temperature, and total roast time, something just didnt land wit me. The relationships weren’t clean. Peak temperature didn’t predict weight loss. Total time didn’t tightly predict weight loss. Even time to threshold temperatures only weakly aligned.

My initial reaction was thoughtful – if the available metrics don’t correlate neatly, how can repeatability be trusted or achieved even?

It’s easy at that point to drift toward the wider internet conversation, bean density measurements, charge temperatures, probe placement, rate of rise curves, more instrumentation (most of which I currently dont understand btw). The implication being that precision requires more variables.

But stepping back, and after a bit of brain struggling, a simpler pattern emerged.

The dataset that I have created so far is small. More importantly, it mixes different coffees, different intentions, and occasional experiments. Cross-coffee analysis introduces noise. I have noticed all be its just a slight signal at present that within coffee repetition seems to reduce it.

The realisation in the end was straightforward i.e. repeatability doesn’t come from correlating everything at once. It comes from holding variables steady and repeating within one coffee.

So the decision now is deliberate consolidation.

Brazil Santos will be roasted repeatedly:

• Fixed batch size.
• 250°C set.
• Fan shift at ~200°C.
• No mid-roast adjustments.
• End point defined by a narrow weight loss band.

No optimisation. No density measurements. No new variables.

Just repetition.

If weight loss clusters and flavour stabilises, that becomes the house style lane for my espresso. If it doesn’t, the deviation will be visible, and measurable.

This feels less exciting than experimentation, but more important. It’s a shift from exploration toward consolidation and ultimately a better understanding of the machine, the coffees and the overall roasting process.

Perhaps predictability doesn’t emerge from adding complexity.

Perhaps it emerges from removing it.

For now, the goal isn’t to prove a theory.

It’s to make one espresso taste like the last one, on purpose.