How do you set success and stop criteria for a chemical trial?

This method covers plant or line trials that compare a new industrial chemical with the product it would replace. It does not set the trial's sample handling or full plan, and it does not replace the site's own change, safety and environmental approvals for bringing a new chemical onto the line.

What do you need before setting success criteria for a chemical plant trial?

Before setting success criteria for a chemical plant trial, you need the trial's purpose in one sentence, the open questions the lab work left, the test method for each measured property with its known precision, and the names of the people who will decide. The Australian competency unit for managing trials lists the same starting work: define the purpose of the trial, confirm it with all stakeholders, identify health, safety and environment risks, and determine the information required from the trial (training.gov.au PMC557001, 2016).

Collect the precision of each test method before you write a threshold. A standardized method usually states its repeatability and reproducibility; ISO 5725-2:2025 describes the basic interlaboratory method that produces those figures (ISO, 2025). ASTM E691-23 notes that knowledge of test method precision is useful "when comparing test results against standard values (such as specification limits) or between data sources" (ASTM E691-23). For an in-house method, run repeat measurements of one retained sample to estimate it.

Collect the new chemical's safety data sheet as well. At a site covered by US OSHA process safety management, a new process chemical goes through management of change before the trial, unless it is a replacement in kind (US OSHA, 29 CFR 1910.119(l)).

How do you set success and stop criteria for a chemical plant trial, step by step?

You set success and stop criteria for a chemical plant trial in five steps, each written down before the trial starts.

1. Baseline. Record the current product on the same line, at the same settings and over the same kind of period as the trial, for example several shifts or days, so day-to-day variation is in the baseline.

2. Metric and method. Name one primary metric that answers the trial purpose, such as a defect rate, a consumption rate or a measured property, plus at most a few secondary ones. Name the test method, sampling point and number of measurements for each.

3. Pass threshold. State the difference from baseline that counts as a pass, in the metric's unit, and the statistical test that will judge it. Set the difference together with the method's precision and the number of measurements; the next sections explain how.

4. Stop limits. State each reading that ends the trial early, its limit and who stops the line. Cover product quality, equipment condition and safety.

5. Sign-off. Have production, quality and the customer's or supplier's technical owner sign the criteria and the analysis method, with the date, before the first trial sample is taken.

A criterion that cannot be written as a number, a unit, a method and an owner is not ready. Treat it as an observation to record, not a pass or fail rule.

Why must chemical plant trial criteria be fixed before the results?

Chemical plant trial criteria must be fixed before the results because a criterion chosen after the data are seen can always be chosen to fit them, and then the trial no longer tests anything. Clinical trial guidance states the principle plainly: ICH E9 says the primary variable "should be specified in the protocol" and that its redefinition after unblinding "will almost always be unacceptable, since the biases this introduces are difficult to assess" (ICH E9, 1998). ICH E9 adds that the principal features of the eventual statistical analysis should be described in the protocol (ICH E9, 1998).

A plant trial has no blinding, but the bias works the same way. Once production sees that scrap dropped on one shift, the temptation is to make scrap the main metric; once quality sees that a property drifted, the temptation is to widen its tolerance. Either change turns the trial result into an argument between the supplier and the customer.

Pharmaceutical development uses the same idea in go/no-go decisions: the criteria a compound must meet at each stage "need to be agreed upon by the medicines developers", and failure to meet them leads to a no-go decision (EUPATI Toolbox, accessed 2026). If a criterion turns out to be wrong during the trial, record the reason, keep the original criterion for this trial's verdict, and use the new one in the next trial.

What goes wrong when a chemical plant trial has no control or baseline?

When a chemical plant trial has no control or baseline, any change in the result can come from the season, the raw material lot, the operator, the equipment condition or the measurement itself, and nobody can show that the new chemical caused it. The NIST/SEMATECH e-Handbook's chapter on product and process comparisons covers the techniques that compare the performance of one or more processes against known standards or one another (NIST/SEMATECH e-Handbook, chapter 7, accessed 2026). Comparing the new chemical with the current one means having data on both.

Day-to-day variation is the usual trap. The NIST/SEMATECH e-Handbook separates repeatability or short-term precision from reproducibility or day-to-day variation, and shows two processes with the same short-term variability where one has large between-day variability (NIST/SEMATECH e-Handbook, section 2.4.4, accessed 2026). A baseline taken on one good day and a trial run over a week of normal days will make the new chemical look worse than it is; the reverse makes it look better.

Take the baseline on the current product immediately before the trial, over a comparable period, on the same line and with the same test method and laboratory. If the line must keep running the current product elsewhere, a parallel line can serve as a control, provided it is shown to run alike.

Which stop criteria should end a chemical plant trial early?

The stop criteria that should end a chemical plant trial early are readings that show product going out of specification, equipment being harmed, or a safety or environmental limit being approached. Write each one as a reading, a limit and the person who stops the trial, so nobody has to debate it on the shift.

• Product quality: a measured property outside the customer specification on a set number of consecutive samples.

• Equipment: a reading such as temperature, pressure drop, deposit build-up or wear outside the range the equipment owner sets.

• Safety and environment: any limit in the new chemical's safety data sheet or the site's permit, such as an exposure, emission or effluent limit, and any incident.

• Process: loss of control of a key process variable that the operators cannot correct within the operating procedure.

The Australian competency unit for managing trials includes recognising indications of developing problems, taking appropriate action and concluding the trial when appropriate (training.gov.au PMC557001, 2016). A stop is not a fail verdict on its own. Record the reading, return to the current product under the operating procedure and the safety data sheet, and decide afterwards whether the trial is invalid, failed or needs a new run.

How large must a difference be to count as a pass in a chemical plant trial?

A difference counts as a pass in a chemical plant trial when it is large enough to matter to the customer and the trial has enough measurements to show it against the test method's precision. A difference smaller than the method's scatter on a single result can still be shown, but only with enough measurements; set the threshold, the method precision and the number of measurements together, using the within-lab precision when baseline and trial are measured by the same laboratory.

Set the pass threshold in three moves. First, state the smallest improvement that would justify the change, in the metric's unit, from the customer's cost or quality target. Second, check it against the method's precision from its standard or your repeat measurements; if the method scatters more than the improvement you need, choose a better method or more measurements. Third, set the number of measurements so the trial can detect that difference.

The NIST/SEMATECH e-Handbook gives the sample size for a one-sided test when the standard deviation is known: to detect an increase of one standard deviation with a significance level of 0.05 and a 0.10 probability of missing it, the minimum is 9 measurements (NIST/SEMATECH e-Handbook, section 7.2.2.2, accessed 2026). Smaller differences need more: the required number grows with the square of the standard deviation divided by the difference (NIST/SEMATECH e-Handbook, section 7.2.2.2, accessed 2026).

How do you check that chemical plant trial criteria will hold up?

You check that chemical plant trial criteria will hold up by reading them against six tests before sign-off; criteria that pass all six can produce a pass or fail verdict without argument after the results. The table below decides whether a set of chemical plant trial criteria is ready.

The pass threshold row fails most often: a threshold set near the method's scatter without enough planned measurements cannot give a clear verdict. The sign-off row is the one that protects the verdict: criteria signed and dated before the first sample can be shown to the customer, the supplier and an auditor as the rule the trial was judged by. Keep the signed page with the trial record, and file any later note about a criterion that looked wrong beside it, not in place of it.

Lubinpla's trial preparation fixes the pass and stop criteria and the analysis method before results come in, and does not change them after the results are seen.