Process-input campaign inquiry

What caused the leading treatment, what the current evidence cannot decide, and which experiments should run next.

Client: Withheld
Program: Confidential process-development program
Evidence reviewed: Bench, pilot, and production campaigns
Decision required: Next controlled screen and process-input choice
Basis date: Withheld
Classification: Client confidential
Publication: Names, dates, condition labels, and raw values withheld

Conclusion

Do not treat the leading condition as a single-factor result.

Treatment A changes the primary input and a cofactor. Treatment B changes only the primary input. The current screen cannot identify which difference caused the higher outcome.

The reported outcome also includes a contribution introduced by the treatment itself. After correction, Treatment A has a local maximum and declines at the highest tested level.

Risk if the current plan proceeds

The program can select a process input for an effect caused by the unisolated cofactor, then carry an uncontrolled thermal variable into scale-up.

Recommended decision Do not select a process input from the current screen.

Next action Run the replicated bench-scale screen specified in Section 4.

Inspect the basis for this conclusion

Basis of conclusion

Halmos corrected the comparison basis, compared the available scales, tested a reduced process model, and simulated the power of the next screen.

Evidence basis fixed before publication redaction

Required actions before protocol release

Priority	Action	Owner	Completion evidence
Blocking	Freeze the decision rule	Analytics lead	Primary endpoint, minimum useful effect, and exclusions approved before randomization
Blocking	Run the replicated bench-scale screen	Experimental lead	All primary-input, matched-cofactor, and baseline conditions complete with predeclared replication
Control	Instrument the scale-up	Pilot lead	Continuous temperature by position and a defined deviation response

Campaign evidence

The reported gain includes material introduced by the treatment.

Figure 2.1. Corrected outcome in the initial bench screen. Values are normalized and treatment names are withheld. The curves show observations, not fitted responses.

Scale	Evidence reviewed	What it establishes
Bench	Initial screen · no replication	Treatment A leads; the cause is unresolved
Pilot	Limited observational record	Shows a similar central result; not a causal comparison
Production	High-variance observational record	Large variation prevents a clean treatment comparison

Decision impact · blocking The comparison basis was wrong

The reported outcome counts material introduced by the treatment.

Decision impact · blocking The leading effect is confounded

Treatment A changes a cofactor that the current controls do not isolate.

Scale-up risk · control required Scale introduces another variable

Production traces show uncontrolled thermal and spatial variation.

Independent analysis

The cofactor hypothesis fits the evidence. It is not proven.

Halmos first removed the treatment-derived contribution from the reported outcome. The inquiry then compared the corrected result across coded input level, scale, and available process conditions.

Y corrected equals Y reported minus k times X treatment

A reduced process model predicts that the cofactor can be limiting. Treatment A changes both factors. The model is directional; only a matched control can test this explanation.

Question	Independent result	Decision use
Does Treatment A lead after basis correction?	Observed	Requires replication
Is the primary input the cause?	Not established	Do not select an input yet
Is the cofactor a plausible cause?	Supported by model	Test with a matched control
Can the result transfer to production scale?	Not established	Control the thermal profile first

Inspect computation and verification

Analysis completed

The work included response analysis, cross-scale summaries, power simulations, a reduced process model, literature review, and treatment costing.

Proposed next campaign

Run a replicated screen before scaling up.

The next screen must separate the cofactor from the primary input and locate the useful operating range. Use one material lot, one process batch, and randomized unit positions.

—
Primary-input screen
Test three coded inputs across the predeclared operating range with independent replication.
—
Matched-cofactor controls
Hold the primary input constant and vary the cofactor independently.
—
Baseline controls
Measure the material-lot baseline in the same process batch.

Purpose Determine whether the lead follows the primary input, the cofactor, or neither. Measure the corrected primary outcome and the predeclared secondary responses.

Protocol and release criteria

Experimental unit: One bench-scale unit. The issued client protocol fixes the number of independent replicates for each condition; the count is withheld here.
Randomization: Use one material lot and one process batch. Randomize unit position before the run.
Primary decision: Determine whether a process input improves the corrected outcome after the matched cofactor control is considered.
Release rule: Do not select an input until the predeclared contrast and the minimum useful effect are met. Set that effect before the run.

Limits

Limits of this inquiry

The initial bench conditions have no replicates. The process model is directional. Several secondary responses were not measured. The production-scale record is incomplete.

What would change the conclusion

A replicated Treatment A advantage not reproduced by the matched control would support a primary-input effect.
The matched control reproducing Treatment A would support the cofactor explanation.
No material difference among controls would require a new operating-range or process hypothesis.

Book a campaign review Return to Halmos Labs