Coffee | One Roast at a Time

When the heater was replaced, the obvious question was whether the machine would behave the same as before.

The answer, after four post-repair calibration roasts, is no.

Not wildly different. Not broken. But different enough that old assumptions could not be trusted. That was the real purpose of this phase: not to prove that the machine was fixed, but to understand what “fixed” actually meant in practice.

These first four roasts have now given a much clearer picture.

What this phase is trying to answer

The goal of post-repair recalibration is simple:

  • to understand how the repaired machine now behaves
  • to test whether older TRT to weight-loss expectations still hold
  • to rebuild confidence using new evidence rather than inherited assumptions

The practical working model has stayed the same:

observe roast behaviour -> respond with TRT -> validate with weight loss

That matters, because TRT remains the main intentional control variable in this project. Fan, drum speed, and set temperature are not being actively manipulated during the roast. The question is whether the same TRT still produces the same kind of result.

So far, the answer is no.

The first signal, Roast 70

Roast 70 was the first cold-start post-repair calibration roast.

It landed at 13:05 TRT and returned only 12.5% weight loss.

That was an important result, not because it was a disaster, but because it showed immediately that the repaired machine was not behaving like the old one had. A roast time that would once have felt reasonable now landed much lighter than expected.

That single roast was enough to remove any confidence in simply reusing the old map.

The second signal, Roast 71

Roast 71 followed as a warm-start comparison.

The purpose here was not to chase a perfect roast, but to test whether warm-machine behaviour still meaningfully changed the outcome. It did.

The warm machine still accelerated the roast compared with a cold start, which means thermal state remained an important variable even after repair. But it did not restore the old logic. In other words, “warm” still mattered, but it did not mean the old timing assumptions were suddenly valid again.

That was the second important finding: the repaired machine still shows cold versus warm separation, but within a new calibration context.

The breakthrough, Roast 72

Roast 72 was the first genuinely strong post-repair anchor.

This was a cold Honduras calibration roast with the following result:

  • Start temp: 23°C
  • 200°C: 4:31
  • 230°C: 7:14
  • 245°C: 8:54
  • TRT: 13:26
  • Weight loss: 15.5%

This roast mattered because it showed something more useful than “the machine is different.”

It showed that the repaired machine could be brought back into a practical development range by extending TRT. That is a much more useful finding than simply saying the machine was lighter than before.

Roast 72 now stands as the first cold post-repair roast that landed where a usable reference roast should land. It was not just better than Roast 70. It was structurally convincing.

That makes Roast 72 the current best cold-start calibration anchor.

The confirmation, Roast 73

Roast 73 used the same Honduras coffee, but as a warm-start comparison.

The result was:

  • Start temp: 67°C
  • 200°C: 3:18
  • 230°C: 5:41
  • 245°C: 7:11
  • TRT: 13:01
  • Weight loss: 16.5%

This roast is extremely useful because it confirms that warm starts are still doing materially more work, much earlier in the roast.

Compared with Roast 72, the warm roast reached:

  • 200°C 1:13 faster
  • 230°C 1:33 faster
  • 245°C 1:43 faster

But the total roast time was only shortened by 25 seconds.

That mismatch explains the outcome. The roast was moving faster all the way through, but not shortened enough at the end to compensate. The result was a roast that landed about 1.0% higher in weight loss than Roast 72.

That does not make Roast 73 a failure. Quite the opposite. It clarifies the model.

It shows that warm starts still require a more meaningful TRT reduction than this test used.

What these four roasts now say together

Taken together, Roasts 70 to 73 establish five important points.

1. Pre-repair timing assumptions are no longer safe

This is the clearest conclusion.

The repaired machine is not behaving identically to the old machine, at least not in a way that allows old TRT to weight-loss mappings to be treated as current truth. Older data still has value as historical context, but not as a live predictive model.

That is exactly why recalibration is necessary.

2. The repaired machine is viable

The first cold roast after repair was light, but that did not mean the machine was incapable. Roast 72 proved otherwise.

The machine can still land in a useful development range. It simply needs to be understood again on its own terms.

That is an encouraging result. Recalibration is working.

3. Start state still matters a great deal

The cold versus warm difference remains real.

Roast 73 confirms that a warm machine still changes the pace of the roast substantially. The start temperature gap between Roasts 72 and 73 was reflected all the way through the roast milestones and into the final weight loss.

This means thermal state remains one of the most important practical variables in the system.

4. Time-to-200°C still matters

Time-to-200°C remains useful as a checkpoint.

Not because it magically explains everything, and not because it should be treated as a complete model, but because it still provides an early read on how the roast is moving relative to expectation.

At this stage, it remains one of the most useful live reference points available.

5. The warm-start correction must be larger than first assumed

Roast 73 is especially useful here.

A reduction of 25 seconds from the cold reference was not enough to achieve parity. The roast still landed higher than the cold roast by about 1.0% weight loss.

So the practical conclusion is that warm-start compensation on the repaired machine needs to be more aggressive than that.

For this Honduras coffee, the current working view is that if a cold roast around 13:26 lands in the right zone, then a warm-start parity roast probably belongs closer to 12:45 to 12:50 rather than 13:01.

That is not a fixed rule. It is a current evidence-based working position.

Where things stand now

The picture is much clearer than it was four roasts ago.

At the start of this phase, the only honest conclusion was uncertainty. The machine had been repaired, but the old map could no longer be trusted.

Now there is at least a basic post-repair structure:

  • Roast 70 showed that the repaired machine could land much lighter than expected
  • Roast 71 showed that warm starts still matter, but do not restore old calibration
  • Roast 72 established the first strong cold-start post-repair anchor
  • Roast 73 confirmed that warm starts still require a materially shorter TRT

That is real progress.

Current practical anchors

For now, the most useful live anchors from this phase are:

  • Cold Honduras reference: Roast 72, 13:26, 15.5% WL
  • Warm Honduras reference: Roast 73, 13:01, 16.5% WL

And the most useful current practical interpretation is this:

A warm roast aiming to land near the Roast 72 cold zone likely needs a TRT somewhere around 12:45 to 12:50, rather than just a small reduction from the cold roast.

What has not changed

It is worth saying clearly what has not changed.

The project still is not trying to force precision beyond what the machine and method can honestly support. This is not about pretending that one formula now explains everything. It is about rebuilding confidence through repeated observation.

That means continuing to separate:

  • what was observed
  • what seems likely
  • what still remains only a hypothesis

That standard matters even more after a repair, not less.

Summary

Four post-repair roasts have been enough to show that the machine is working, but not the same.

The repaired Gene Café is landing differently enough that pre-repair TRT assumptions must remain historical. At the same time, the new evidence shows that the machine is responsive and usable, and that a practical recalibration path is already emerging.

Roast 72 is now the best cold-start anchor. Roast 73 confirms that warm starts still need more compensation than first assumed. Time-to-200°C remains a useful checkpoint. And the current direction is no longer guesswork in the dark, but early structure built from fresh evidence.

A small technical note is worth adding here. The replacement heater specification may not be identical to the original part, particularly around 220V versus 230V rating. I am not treating that as a proven causal explanation on its own, but it is one reasonable reason to avoid assuming that pre-repair roast timing should map cleanly to post-repair results. The practical conclusion remains the same: the repaired machine had to be understood again through fresh roasts rather than old assumptions.

That is where things stand for now.