Roast 91 and Roast 92 were supposed to be simple.

That was the point.

Same coffee.
Same batch size.
Same target temperature.
Same cold-start intent.
Same total roast time.

In other words, this was meant to be a straightforward repeatability check.

And in a way, it still was.

Just not in the way I would have preferred.

The setup

Both roasts used Colombia Supremo (Washed).

Both were 250g batches.
Both were run at 250°C.
Both were cold-start roasts.
Both were stopped at almost exactly the same total roast time.

Roast 91 started at 20°C and finished at about 13:21, giving 210.7g out, which is 15.7% weight loss.

Roast 92 started at 22°C and finished at about 13:22, giving 212.4g out, which is about 15.0% weight loss.

That is not a tiny difference.

That is enough to matter.

Why this was unsettling

If I am being honest, Roast 92 put a spanner in the cleanest version of the consistency story.

Because on paper, these two roasts should have been very close.

The start temperatures were close enough not to explain the result.
The coffee was the same.
The target temperature was the same.
The total roast time was effectively the same.

And yet Roast 92 landed 0.7% lighter.

That is not nothing.

It is exactly the kind of result that forces a more careful reading.

What the data shows

The important thing is that Roast 92 did not land lighter for no reason.

It heated more slowly.

That slower pace showed up throughout the roast, not just at one moment.

Roast 91 reached 200°C at about 4:07, while Roast 92 reached it at about 4:50.
Roast 91 reached 230°C at about 6:13, while Roast 92 reached it at about 7:27.
Roast 91 reached 245°C at about 7:56, while Roast 92 reached it at about 9:14.
Roast 91 first touched 250°C at about 8:28, while Roast 92 did so much later, at about 10:14.

That matters because even though the two roasts ended at almost the same total time, they did not spend the same amount of time in the hotter upper part of the roast.

Roast 92 arrived there much later.

So the lighter result makes sense.

The problem is not that the result is unexplained.

The problem is that the reason for the slower heat-up is still not clear.

The identified challenge

That is really the challenge this pair has exposed.

At the moment, I can say this with confidence:

• the drift was real
• the slower pace was visible in the app overlay
• the slower pace showed up in the milestone timings
• the final weight loss confirmed that the slower pace changed the result

What I cannot yet say with confidence is why Roast 92 heated more slowly.

The two JSON logs show the effect.

They do not fully explain the cause.

That matters, because it puts a limit on how strongly I can talk about repeatability right now.

What this changes in my thinking

Roast 91 and Roast 92 do not destroy the project method.

But they do force a correction.

They make it harder to say:

same TRT gives the same result

That is too strong.

What feels more honest now is this:

TRT is still useful as the planned input, but it is not enough on its own.
Roast pace through the roast, milestone timings, and final weight loss still matter.

That is a less tidy conclusion, but probably a more accurate one.

What still holds

Even with this wobble, I do not think the machine becomes meaningless.

Roast 91 still stands as a strong cold-start Colombian reference on this current heater state.

Roast 92 still makes sense as a lighter result once the slower progression is acknowledged.

So the machine is not acting randomly.

But this pair is a reminder that a repeat roast is only really a repeat if the roast pace stays close enough as well.

That is an important distinction.

The current dilemma

The dilemma is fairly simple.

If Roast 92 had landed right on top of Roast 91, I could say with more confidence that the current cold-start Colombian baseline was holding cleanly.

Instead, I now have:

• one stronger cold-start Colombian espresso roast
• one lighter repeat with the same plan
• and no fully satisfying explanation yet for why the second roast heated more slowly

That means caution is needed.

Not panic.

Not overreaction.

Just caution.

What happens next

The obvious answer is another controlled repeat.

That is what Roast 93 is for.

If Roast 93 lands close to Roast 91, then Roast 92 begins to look more like an outlier.

If Roast 93 lands close to Roast 92, then the baseline may have shifted.

If Roast 93 lands somewhere in between, then that tells me something useful as well.

That is the next sensible step.

For now, the practical lesson from Roasts 91 and 92 is this:

same plan does not guarantee same result
and TRT alone is not enough to carry the consistency claim

The method still works.

But it needs to stay honest.

One roast at a time.