Before comparing two facer trials, agree on the decision the comparison is meant to support, identify the construction and process used for each run, and keep the measurement conditions consistent. Record unavoidable changes separately. A difference between two finished boards is not automatically a difference caused by the facer.
A trial can produce good boards and still produce a weak comparison. The problem usually becomes visible at the review table: one sample has a clear production history; the other has a result, a photograph and the description “same as before”. There is no reliable way to reconstruct what “same” included.
For a GRECHO sample-review discussion, we would start with that missing comparison record, not ask for another batch of measurements immediately. The useful question is not simply which board performed better. It is whether the evidence answers the question that justified making the boards.
Two legitimate questions that need different trial plans
Consider an illustrative planning situation, not a reported GRECHO customer trial. A manufacturer is comparing its reference facer with a candidate. The candidate is run with a different adhesive setting because that produces a more acceptable surface.
There are now two possible decisions:
| Decision being investigated | A defensible comparison | What the result would not establish by itself |
|---|---|---|
| Can the candidate replace the reference under the existing process? | Reference and candidate under a documented common process, where safe and technically appropriate | Performance under every other process setting |
| Can a candidate-plus-process package meet the board requirement? | Each identified construction with its own documented, approved process | That the facer alone caused the difference |
Neither question is wrong. The mistake is running the second comparison and reporting the answer as though the first had been tested.
If safe operation requires a process change, make it and record it. Do not keep an unsuitable setting merely to preserve the appearance of a controlled experiment. Instead, describe the trial honestly as a construction-and-process comparison.
This distinction also helps separate an initial sample suitability review from a comparative trial. The former asks whether a direction deserves evaluation. The latter asks what the observed difference can be attributed to.
“Freeze” means define and record, not pretend nothing varies
Some inputs can be held constant. Others can only be kept within an agreed range. Still others, such as the production day or a raw-material batch, may change despite careful planning.
Use three labels in the trial record: deliberately changed, controlled, and recorded but not controlled. A blank cell is not a fourth category meaning “probably unchanged”. Mark it unknown.
The following is a proposed review structure, not a universal test protocol:
| Record group | What to identify before the run | What to preserve afterwards |
|---|---|---|
| Facer | Material reference, revision, lot, roll, contact face and unwind direction | Retained sample and the actual roll used |
| Board construction | Core, thickness, relevant surface state, adhesive, slurry or foam system | The as-made construction, including deviations |
| Process | Approved recipe, operating window and intended sequence | Actual recorded conditions and interventions |
| Sampling | Collection positions, timing and the production units represented | Sample map linking each specimen to its run |
| Conditioning and test | Specimen age, environment, method, direction and geometry where relevant | Test record, not just the result copied into a summary |
| Decision | Primary requirement, secondary requirements and the release authority | Accepted, rejected or unresolved scope with reasons |
Lot identification belongs in this record, but it does not replace control of the experiment. Our lot and roll traceability guide explains how to maintain identity; the question here is what those identified samples can legitimately be compared against.
Keep time from becoming the hidden explanation
Running all reference material on one day and all candidate material on another can entangle the material change with a day change. NIST's experimental-design guidance treats nuisance factors separately from the factor being investigated and describes blocking as a way to organise comparisons around them.[1]
The practical lesson is modest: when planning allows, place meaningful reference and candidate comparisons within comparable production periods rather than letting material identity and operating period become inseparable.
A returning reference can be useful. If the reference looks different at the end of the session, the candidate should not be judged against the start-of-session reference as though the baseline were unquestionably stable. This is an indication to investigate drift, not a mathematical correction that automatically rescues the trial.
Randomisation, repeated runs and blocking should be designed around the equipment and the safety constraints. Where changes are difficult or costly, restricted randomisation needs to be recognised rather than hidden; NIST discusses such arrangements separately.[2] Do not repeatedly switch a production line in an unsafe or technically inappropriate way just to imitate a laboratory schedule.
Five pieces from one board do not represent five production runs
Suppose five test pieces are cut from one board. They can reveal differences across that board and variability in specimen preparation. They do not, simply by being five pieces, demonstrate reproducibility across five independent production runs.
Keep both levels visible: how many boards or runs were made, and how many specimens were cut from each. Combining the results into one average can conceal the distinction. NIST's discussion of nested variation is relevant here: observations can be grouped within a larger experimental unit, and their structure matters to interpretation.[2]
For a first practical review, retain the individual results with their sample positions. A mean without the underlying pattern may disguise a good centre and weak edges, or one favourable board among several inconsistent ones. Deciding the required sample size and statistical analysis belongs with the competent trial and testing team; this article does not prescribe a minimum number.
A short change ledger is more useful than a long explanation afterwards
Before examining the final result, compare the planned and actual records. This illustrative ledger shows how to keep observations and conclusions separate:
| Recorded event | What can be said | What should remain unresolved |
|---|---|---|
| Candidate used with a different adhesive setting | The evaluated package included both changes | Facer-only effect |
| Reference and candidate tested at different ages | Conditioning was not matched | Whether age contributed to the result |
| Both builds show the same unexpected change | The reference also moved | Whether the candidate explains that movement |
| Only one labelled position differs | A local difference was observed | Its prevalence across production |
| A required process field is missing | The comparison has an information gap | Any conclusion that depends on that field |
Do not exclude an inconvenient specimen after seeing its result merely because it weakens the preferred conclusion. Record the reason for any exclusion and distinguish an identified testing error from unexplained variation.
Agree on the release decision before seeing the best sample
A useful trial should not leave the team with only “better” or “worse”. It should support a defined next step: reject the direction, repeat a limited comparison, extend a promising result, or release a specifically described construction within an agreed scope.
Set the primary requirement before running. Keep secondary requirements visible, too. A stronger interface may not compensate for an unacceptable surface; a cleaner surface may not answer a board-performance requirement. Which trade-offs are acceptable is a project decision, not something a supplier can infer from the most favourable measurement.
A good closing note might read: “The candidate-plus-process construction met the agreed screening checks. A facer-only comparison was not performed. Further production verification is required before extending the result to other conditions.” That is a useful conclusion, even without a dramatic improvement percentage.
For GRECHO's part in application-matched facer selection, the most helpful request includes the decision to be made, the two construction records, the change ledger and the results still needing explanation. Sample and documentation coordination can then follow the actual uncertainty.
Questions to settle before comparing results
Must every variable be identical?
No. The factor under investigation must differ, and some process changes may be necessary. The important step is to identify them and limit the conclusion accordingly. A changed facer plus a changed process is a package comparison unless a suitable design separates their effects.
Can the previous month's accepted board be the reference?
It can provide context, but it is not automatically a matched experimental control. Differences in raw materials, conditioning, production period and test preparation need to be considered before attributing a result to the new facer.
Does one successful board justify a production release?
It can justify a next step, but it does not by itself define the breadth of release. The trial owner should specify which construction, process range and verification evidence the decision covers.
Discuss the comparison record with GRECHO.
References
[1] NIST/SEMATECH. Randomized block designs. Public experimental-design guidance; not a board-manufacturing standard. Source
[2] NIST/SEMATECH. Nested variation and restricted randomization. Guidance on experimental structure. Source
