The purpose of this pair was simple: stop drifting beyond target and generate a cleaner post-repair comparison.
Roasts 72 and 73 were useful, but they were not strict enough to fully trust as calibration anchors because the roasts were allowed to continue a little. This time the intent was different. Hold the target TRT exactly. No helping the roast. No emotional extension. Just observe what the repaired machine actually does.
That matters, because after a heater replacement the machine may be functional without yet being behaviourally understood. The goal here was not to make the “best” coffee. It was to create a cleaner cold and warm pair that could be trusted.
Roast details
Roast 74, Cold Calibration
Coffee: Honduras (Wanted), Coffea arabica
Batch size: 250 g
Start temperature: 21°C
Predicted TRT: 13:26
200°C: 4:54
230°C: 8:03
245°C: 10:08
Actual TRT: 13:27
Weight out: 213.5 g
Weight loss: 14.6%
Roast 75, Warm Calibration
Coffee: Honduras (Wanted), Coffea arabica
Batch size: 250 g
Start temperature: 69°C
Predicted TRT: 12:47
200°C: 3:36
230°C: 6:17
245°C: 8:25
Actual TRT: 12:47
Weight out: 210.7 g
Weight loss: 15.7%
Observation
The pair behaved clearly and cleanly.
Roast 75, the warm roast, accelerated strongly versus Roast 74 from the start and stayed ahead throughout the roast. It reached 200°C 1 minute 18 seconds earlier, 230°C 1 minute 46 seconds earlier, and 245°C 1 minute 43 seconds earlier. That is not a marginal effect. Start state remains a major variable on the repaired machine.
At the same time, Roast 74 cold was noticeably slower than expected for a 13:26 to 13:27 cold target. The slowdown was visible at every milestone, not just at the end. Despite finishing at effectively the same TRT as Roast 72 cold, it landed at 14.6% weight loss rather than 15.5%.
That makes Roast 74 especially important. It was not a bad roast. But it was a slower roast, and the lower weight loss reflects that clearly.
Interpretation
This pair is stronger than Roasts 72 and 73 because the targets were actually followed. That makes the result easier to trust.
The warm side looks much more stable now.
Roast 73 warm finished at 13:01 and landed 16.5% weight loss. That suggested the warm target was too long if the aim was to land closer to the cold reference zone. Roast 75 reduced the warm TRT to 12:47 and landed 15.7%. That is a much more believable and usable outcome. The correction worked.
So the repaired machine still behaves like a warm-start machine should: it accelerates significantly, and it requires a shorter TRT to stay in the same broad development band. On Honduras, 12:47 now looks like a credible working warm target.
The cold side is different.
Roast 72 cold at 13:26 landed 15.5%. Roast 74 cold at 13:27 landed 14.6%. Because the TRT was effectively identical, the difference has to be understood through roast pace rather than endpoint alone. Roast 74 was slower to 200°C, slower to 230°C, and much slower to 245°C. In other words, the same nominal TRT did not represent the same thermal work.
That does not mean TRT stops mattering. It means TRT still needs to be read in context.
What this pair confirms
First, strict adherence to target TRT produces much cleaner learning than intentionally extending a roast.
Second, warm start remains a major post-repair variable.
Third, the shortened warm target was the right move.
Fourth, the cold side is not yet stable enough to treat 13:26 as a locked truth. It may still be directionally right, but it is not yet repeatable enough to be considered a fixed anchor without another validation roast.
Practical takeaway
For Honduras on the repaired Gene Café CBR-301, the working position now looks like this:
Warm start is becoming clearer.
A TRT of around 12:47 is now the better warm anchor.
Cold start is still provisional.
A TRT of 13:26 to 13:27 can land differently depending on roast pace, so another cold validation roast is needed before treating that number as reliable on its own.
Broader lesson
This pair supports the current project approach rather than undermining it.
TRT remains the intentional control variable.
Weight loss remains the main validation output.
But observation still matters, because equal TRT does not guarantee equal roast progression.
Roast 74 and 75 are a good example of that. The warm correction behaved as hoped. The cold result stayed honest and showed that calibration is still in progress.
Next step
The next logical move is not to change everything. It is to validate the cold side once more.
If the next cold Honduras roast again runs slow through the milestones, the cold TRT likely needs to move slightly longer than 13:26. If it behaves more like Roast 72, then the issue may be variability in machine state rather than a fully shifted baseline.
Either way, the next job is clear: keep the system simple, repeat the coffee, and let the machine tell the truth.