Coffee | One Roast at a Time

Date: 2026-02-09
Roaster: Gene Café CBR-301
Batch: Sample Roast (low mass)

In: 52.5 g
Out: 44.3 g
Weight loss: 15.6 %

1. Purpose & framing (important)

This roast sits squarely inside the agreed Sample Roast guardrails:

  • It is not trying to optimise flavour
  • It is not intended to be repeatable
  • It exists to answer one question only:

Can the Gene Café apply and control heat coherently at very low bean mass, and do the usual decision anchors still hold?

This roast answers that question better than Sample Roast 1.

2. System context: warm start (deliberate variable)

Unlike Sample Roast 1, Sample Roast 2 was started on a warm machine.

Key implications:

  • Starting temp ≈ 52 °C at t=0
  • No true “cold-start drying phase”
  • Early moisture removal and momentum are compressed
  • Audible cues become even less reliable than usual

This was intentional, and it matters.

The roast behaved exactly as a warm-start sample should.

3. Early phase behaviour (0–3 min): compressed but stable

Despite the warm start:

  • Temperature rise is smooth and linear
  • No stalling, no runaway acceleration
  • Fan and stirring stayed at baseline (2 / 2)

This confirms something important:

At ~50 g, the Gene Café is not struggling for thermal coupling early on — it is actually very stable.

That’s a good sign for system control.

4. Mid-roast (yellow → light brown): visual > auditory

From ~150 °C through ~190 °C:

  • The curve is extremely smooth
  • No sudden RoR spikes
  • No evidence of scorching or tipping risk
  • Colour change would have been fast but even

Your memo notes:

“thought I heard pops but was not a crescendo”

That matches expectations for:

  • Red Honey processing
  • Very low bean mass
  • Warm start
  • Active exhaust + Gene Café drum acoustics

This is not a failure to detect crack — it’s crack being a weak signal here.

5. Fan behaviour & system intervention

You observed:

“Same heat behavior with fan when kicked in to 3 and recovered back to 2”

The JSON confirms:

  • The familiar Gene Café airflow intervention pattern still appears
  • Even at ~50 g, the machine:
    • Protects itself
    • Increases airflow
    • Recovers back to baseline cleanly

This is a key result:

Low batch size does not break the machine’s thermal safety logic.

That’s a strong green tick for sample-scale viability.

6. End of roast decision: colour + structure led

Cooling was triggered based on:

  • Visual colour
  • Overall roast “feel”
  • Not on crack sound or timers

With:

  • 15.6% weight loss
  • Peak temps hovering ~250–252 °C
  • No visible runaway

This places the roast:

  • Structurally in medium territory
  • Slightly more developed than Sample Roast 1
  • Entirely reasonable for a red honey coffee

Importantly:

The roast was ended intentionally, not reactively.

That’s exactly how sample roasts should end.

7. Weight loss: the most important anchor here

At 15.6% WL, this roast tells us several things:

  • Heat application was real and effective, despite low mass
  • The warm start did not cause underdevelopment
  • The Gene Café can still “do work” on the beans at ~50 g

Compare that to full-batch behaviour:

  • This WL would normally land you in a safe, drinkable zone
  • Here, it mainly confirms energy delivery, not flavour quality

Weight loss did its job as a diagnostic metric.

8. What this roast proves (system-level)

Sample Roast 2 clearly demonstrates that:

  1. The Gene Café can roast very small batches coherently
  2. Airflow logic still functions correctly at low mass
  3. Crack audibility becomes almost meaningless under:
    • warm starts
    • honey processing
    • low charge weight
  4. Colour + time + weight loss remain valid anchors
  5. Warm-start sample roasts are viable — but must be treated as a different class of data

This is exactly the learning Sample Roasts exist for.

9. What this roast does not do (by design)

To be explicit and aligned with the project:

  • It does not define a profile
  • It does not predict cup quality
  • It does not justify changing full-batch strategy
  • It should not be compared directly to Roast # references

This stays firmly in Notes-only territory.

10. Forward learning (calm, non-prescriptive)

Two useful takeaways to carry forward — quietly:

  • Warm-start sample roasts are valid, but should be mentally tagged as such
  • On honey / natural coffees at low mass:
    • Crack sound ≈ noise
    • Weight loss + colour are king

No action required yet. Just awareness.