Decide what this roast is for
A sample roast and a production roast have different jobs. A sample roast may aim to reveal characteristics of incoming green coffee under a defined evaluation method. A production roast is designed for a particular product and brewing context. Neither purpose justifies ignoring consistency, but each calls for different acceptance criteria. Write the purpose before choosing a profile, then keep the sensory question attached to the batch record.
Record green-coffee identity, batch mass, equipment, operator and relevant environmental conditions. Note any unusual storage history or physical appearance. Coffees from Brazil's Cerrado Mineiro, Guatemala's Antigua and Indonesia's Sumatra can differ in structure and processing history, but origin is not a complete operating instruction. Respond to the actual material and your validated process rather than assigning an automatic profile to a place name.
Understand what the trace measures
A temperature probe reports conditions at its position, filtered through its construction, response time and surrounding environment. Probe placement, batch size and airflow affect the signal. A bean-temperature label in software does not mean the reading is the temperature at the centre of every bean. When comparing runs, first confirm that the measurement arrangement is comparable. A similar curve from a different machine is not proof of an equivalent roast.
Software such as Artisan or Cropster can organise roast data and calculate a rate of rise. That calculation describes change in the recorded signal over time. Smoothing settings and sampling intervals influence how the displayed line looks. A sudden feature deserves investigation, but it should not automatically be translated into a named sensory defect. Check the underlying data, operator events and measurement conditions before building a story around the graph.
| Evidence | Useful for | Does not establish alone |
|---|---|---|
| Time and temperature record | Comparing process behaviour on a defined system | Flavour quality or internal bean chemistry |
| Colour measurement | Checking appearance under a controlled method | An equivalent sensory profile |
| Prepared-cup observations | Describing the result in specified conditions | The unique cause of every difference |
Use colour as a controlled measurement
Roast colour can help a team detect drift, but preparation matters. Whole-bean and ground measurements answer related, not identical, questions. Particle size, sample presentation, surface condition and time after roasting may influence results. Follow the instrument's instructions and document the preparation. An Agtron instrument and a different manufacturer's colour meter should not be assumed interchangeable merely because both display a number.
Keep the instrument's check and calibration records where operators can find them. Use the appropriate reference procedure and distinguish routine verification from adjustment. If a result looks surprising, inspect sample preparation before repeating measurements until a preferred number appears. Report the scale or instrument system with the result; an unexplained colour value loses much of its meaning when it travels to another laboratory.
Bring the cup back into the investigation
After roasting, control storage, resting conditions and preparation for evaluation. Different rest intervals can alter the experience of the same batch. Compare like with like wherever possible and record exceptions. Use blind codes when the tasting team might otherwise recognise the supposedly successful profile. An attractive curve or an operator's confidence should not become a sensory cue before the first independent note.
Distinguish descriptive observations from diagnostic claims. “Persistent smoky aroma” is an observation. “Airflow caused this” is a hypothesis requiring further evidence. If several variables changed at once, you cannot confidently assign the sensory difference to one of them. Design a follow-up that changes a manageable factor while holding other relevant conditions steady, and repeat enough to see whether the pattern is dependable.
Compare within a sensible boundary
Machines from PROBAT, Loring and IKAWA use different designs and operating contexts. A useful internal benchmark does not automatically transfer between them. Even on one machine, maintenance, sensor replacement or a different batch size can change the relationship between settings and results. Treat such changes as reasons to re-establish the comparison, not as minor details to omit from the record.
Agree what action a roast review can support. A production team might repeat a batch check, investigate a deviation or revise an internal control limit after collecting suitable evidence. Avoid inventing universal limits from a few favourable cups. Maintain the distinction between a process target, a measured result and an acceptance decision. That separation makes the record useful when the next batch does not behave as expected.