Start with the completed run
Record the app, selected routes, input identity, source dimensions, chosen channels or region, settings and route status. Keep unavailable or refused results with their reasons. A successful export does not turn an unavailable analysis into a measured zero.
Read the figure as a display
Check axes, units, legends, color scales and the scope of the plotted observations. For images, retain orientation and spatial correspondence. A plot may display selected points, bins, rounded labels or a bounded preview. Its appearance alone cannot establish that every numerical row is visible or exported.
Count the content that was actually saved
Identify each named table and array before comparing row counts. A complete summary table may contain only a few fields even when the input has thousands of observations. Some CSV files contain several tables or JSON row arrays inside cells. Follow the selected app’s export contract; a file extension or an Export All label is not a universal schema.
Preserve precision and missing-result meanings
Compare numeric fields at their exported precision. A complete export can faithfully retain an already rounded text value; it cannot recover digits the app did not save. Keep units and field definitions with each value. A blank CSV cell or JSON null can mean that a comparison was unavailable. Do not replace it with zero or interpret a formatted percentage as an uncertainty estimate.
Distinguish reopening a file from restoring a session
Open a saved figure and inspect its numerical companions outside the viewer. A PNG or result ZIP can be an inspectable snapshot without containing the source or a restorable session. Use a Save Session/Load Session workflow only where that selected viewer explicitly supports it and its recorded evidence establishes the relevant content. A restored result view is still separate from reproducing an analysis from the original input.
Check identity separately from scientific agreement
Where a manifest is supplied, compare filenames, sizes and hashes, then inspect the contents and their dimensions. Keep the final package identity separately when the manifest does not include itself. Identify whether the manifest came from the app or an external reviewer. Hash equality establishes file identity; row and value agreement establishes a particular representation comparison. Neither proves physical truth, clinical validity or accuracy of every app route.
Keep a reviewable handoff
Retain the original input and attribution, preparation record, run settings, selected routes, figure/table/array files, and any native or external manifest. Include uncertainty only where the workflow supplies and defines it. Review paths and identifying scientific content before sharing; withholding a filename does not remove the data themselves. State the example and recorded version to which a comparison applies.
Choose a representation using a named example
These examples retain the source, version and acceptance limits in their linked guides. They do not imply identical exports across a product or new scientific or recording acceptance.
| What you need | Saved evidence and instructions |
|---|---|
| Compare every spectral row | The recorded CWRU workflow saves 60,996 FFT rows and 513 Welch rows. The numerical tables preserve values that cannot be recovered accurately from a spectrum image. |
| Retain every sounding | The Baja map displays 60,000 points while the private NPZ retains all 82,970 rows. Its viewer ZIP contains summary tables; it is a different representation from the sounding arrays. |
| Recover complete sequences | The ACTB workflow preserves fifteen complete translated ORF sequences in its FASTA exports. A text preview or a report saying that translation succeeded is not a substitute for those sequences. |
| Understand preserved rounding | The saved protein-composition CSV preserves exact residue counts alongside percentages formatted as text. Export completeness does not turn those formatted percentages into full-precision numeric fractions. |
| Review saved results in a viewer | The historical EEG session restored nine plots and sixteen tables, including 4,332 mean-spectrum rows. Its all-tables CSV is a Table/Metric/Value container; its walkthrough hash manifest was external, not a native Save Session feature. |