Define the question before choosing controls
Name the quantity you intend to measure and the input scope it applies to. Distinguish a descriptive plot, a candidate detector, a numerical estimate and a comparison against a reference. A familiar method name does not establish that two apps use the same estimator, preprocessing or normalization.
Select the routes that have evidence for the task
Inspect the selected route list rather than assuming a default or Select All describes the intended analysis. A saved example can cover a few routes in an app with many choices. Check masks, reference data, model fields and other prerequisites before running. Preserve unavailable and refused outcomes with their reasons; they are not zero measurements or successful calculations.
Separate input scope, method and display choices
A channel, image plane, volume or output-species selection defines the data included. A window, estimator, threshold or solver tolerance defines part of the calculation. Plot limits and table previews define what is shown. Record these roles separately: changing a preview does not necessarily change the calculation, while changing an input selection does.
Read values together with units and definitions
Retain exact values, units and control meanings. An overlap percentage and overlap sample count must describe the same segmentation. A list of requested percentiles is different from a bin count. Output points are different from a solver’s internal steps. Do not convert a setting into an accuracy claim or infer that a control affects every method in the app.
Keep preparation and comparison settings matched
When comparing existing results, first align input identity and order, preparation, selected region or channel, method and output definition. Preserve any stated reference scope or tolerance. A small numerical difference does not by itself identify a defect; a matched result does not establish physical truth or performance on other inputs.
Treat changed settings as a different run
Keep the configuration that produced the saved result, including explicit defaults and the release that defined them. Preserve the previous result when changing a setting. If a reproducibility seed is supplied, retain its exact value and purpose; it does not automatically establish uncertainty estimation. A new configuration needs its own run record and whatever checks the task requires.
Verify the recorded configuration after completion
Inspect the saved input contract, selected-method list, settings, run status and result units. Confirm that the calculation used the intended scope instead of relying only on controls currently visible in the interface. Keep warnings and method metadata with the exported numerical tables. Follow the linked guide’s availability gate and evidence limits before attempting any example.
Match the control to its actual effect
These examples explain settings from existing detailed guides and saved receipts. They are not new runs, universal defaults or approval to bypass an unavailable app.
| Setting to distinguish | Named example and instructions |
|---|---|
| Percentile levels, bins and selected domain | The reviewed Medical example requests levels 1, 5, 25, 50, 75, 95 and 99, uses 128 bins and explicitly selects Volume (3-D). Those settings have separate purposes. The current percentile contract requires finite, strictly increasing levels; duplicate levels are refused. |
| Window, detrending and overlap | The recorded CWRU configuration uses Hann, constant detrending, 1,024 samples per segment and 50% overlap, or 512 samples. Mean removal applies to the complete primary channel; Welch adds no per-segment detrending. Full-record FFT amplitude and segmented Welch density remain different outputs. |
| A threshold versus a requested fraction | The saved Z-score example uses an absolute threshold of 3. Its recorded expected-anomaly-fraction control is 0.01, but that control does not make this route flag exactly one percent of rows. Its initial reference segment is not independently verified healthy operation. |
| Estimator and low-b definition | The current DWI defaults are WLS and a low-b threshold of 50 s/mm². The saved comparisons cover OLS and WLS at 256 sampled locations. They do not establish NLLS, RESTORE, preprocessing or whole-volume accuracy merely because those choices appear in the interface. |
| Output samples versus integration history | The SBML example requests 51 output points from model time 5 to 10 while initial conditions remain at time zero. Points include both endpoints; they are not internal solver steps. Retain integrator, tolerances, step cap and exact event-order seed with the accepted engine revision. |
| Default route selection versus the saved workflow | The ACTB ORF app initially selects 20 routes. Its saved translation example selects ORF Finder (6 frames), Longest ORF, Translation and Coding Potential (unavailable). Preserve the unavailable predictor’s reason; that route does not produce a probability. |
| A model ranking within its candidate set | The locally reviewed Model Selection workflow holds a supplied topology fixed and compares explicitly selected models. A preferred AIC, BIC or AICc result is conditional on those candidates and assumptions. Public execution remains unavailable; this guide does not authorize bypassing that gate. Model Selection: candidate assumptions and availability boundary |