Control activity separates
controlThe pH-control trajectory exceeds the configured within-phase threshold while the two references remain aligned.
Fermentation run qualification
Can these replicate runs be trusted together?
A local-first evidence review for fungal stirred-tank experiments—built to separate biological interpretation from mechanical, environmental, sensor, and execution uncertainty.
0.25–1.25 L fungal STRs · CSV, PI-style historian, samples and event records
One synthetic experiment
Two comparable references. One run that needs a closer look.
The prototype synchronizes the evidence, locates divergence, and shows what changed nearby without turning association into causality.
Clock-aligned trajectory
Agitation across replicate runs
R01R02R03
Deviation timeline
The evidence converges near the same window.
Each signal remains traceable to its source record. Timing establishes context, not causality.
Agitation drops
equipmentR03 falls from roughly 690 to 455 rpm and stays separated. The observation is material; the cause remains unresolved.
Phase timing shifts
biology or executionRecorded glucose depletion occurs about 5.9 hours later than the reference median, changing how clock-time differences should be read.
R03 begins separating from R01 and R02 around hour 17. The split follows abnormal control activity and coincides with agitation and later glucose-depletion differences. Available evidence cannot determine whether the explanation is biological, control-related, mixing-related, or execution-related.
Before the DOE conclusion
Qualify the evidence before asking it to explain the biology.
RunQual is designed to sit upstream of JMP—not replace it. Its job is to make the experimental comparison more trustworthy and the unresolved confounders more visible.
Reconstruct
Combine process trajectories, samples, recorded events, alarms, notes, metadata, and provenance.
Align
Compare by clock time, elapsed time inside recorded phases, and normalized phase progress.
Qualify
Surface divergence, missing evidence, sensor barriers, and nearby contextual differences for review.
Hand off
Export a normalized, annotated evidence package for scientist review and downstream JMP analysis.
Human-owned boundary
A recommendation is a review state—not a verdict.
The prototype can organize and rank evidence. It cannot infer ground truth from observational data, authenticate what happened in the physical reactor, or replace a scientist’s exclusion decision.
QUALIFIED
Sufficiently comparable under the configured evidence gates.
REVIEW
A meaningful deviation requires scientist assessment.
EXCLUDE CANDIDATE
Material evidence suggests an execution, sensor, or equipment problem—but no run is removed automatically.
INSUFFICIENT EVIDENCE
Required data, units, phase anchors, or coverage are missing.
The next experiment
Would this have made one of your past reviews faster or more defensible?
The prototype is ready for expert criticism. The next step is one permitted, deidentified historical experiment: ideally two acceptable replicates and one run suspected of mechanical or environmental trouble.
01 Which flags are useful, obvious-but-time-saving, false alarms, or scientifically invalid?
02 Which signal or context is missing from this review?
03 What final outcome defines success—biomass, enzyme, protein, metabolite, or another response?
Safe data boundary
Use only deidentified historical exports with explicit company permission. The prototype runs locally; no company data is used in this public demonstration.