Skip to content

Blinded analysis mode

Blinded analysis mode reduces experimenter bias during analysis by hiding file and condition identity in the CloudScope GUI. Combined with a randomized file subset, it lets you analyze a large dataset without bias in either the choice of files or the scoring of each file.

Turning it on

Open the Config panel from the left toolbar and toggle blinded mode. The setting is saved immediately and applied across the home page and pool plots.

What is masked

While blinded mode is on, CloudScope replaces identifying information with stable File N labels:

  • File names and file-list / acquisition-tree rows, including parent and grandparent folder names.
  • Experimental metadata columns (for example condition and genotype), which collapse to a fixed Blinded value.
  • Footer and analysis-panel selection labels.
  • Pool-plot grouping columns and point labels; pool selections still map back to the correct underlying rows.

The Experiment Metadata panel is disabled (and closed) while blinded so that condition and genotype cannot be read or edited.

What is not masked

Blinded mode is a bias-reduction display layer, not a security boundary. By design the following remain unblinded:

  • Load and save controls, recent paths, native file pickers, and save paths.
  • App Info and log output.
  • Real file paths and identifiers in backend state, saved files, and logs.

Anyone with disk or log access can still recover file identity. Use blinded mode to keep the person doing the analysis from being influenced by identity, not to enforce access control.

  1. Sample without bias. Use the randomized file subset notebook to draw a randomized, per-condition subset from a large folder and write it to a manifest CSV.
  2. Load the subset. In CloudScope, use Load CSV from the history menu in the load/save controls to open the sampled manifest.
  3. Blind the analyst. Turn on blinded mode in the Config panel before scoring or running analyses.
  4. Analyze and save. Run analyses as usual; results are saved next to each source file with true identities intact.
  5. Unblind for interpretation. Turn blinded mode off (or reload) once analysis is complete to group and interpret results by condition.