Glass Box Attribution
concept
Core idea
A tool whose attributions you cannot explain is a liability in court. Glass box attribution gives full transparency into how an attribution is derived: it surfaces the underlying sources, confidence levels, and reasoning, rather than handing you a conclusion with no visible basis (a black box). Under a Daubert-style challenge, a black box you cannot independently explain is exactly where the case breaks.
Components
Three things to demand of any tool whose output may become evidence:
- Reproducibility. Given identical inputs and the same tool version, does it produce consistent outputs?
- Transparency. Does it surface the basis for an output (sources, confidence levels, reasoning), or only the conclusion? Glass box over black box.
- Version control. Are outputs date- and version-stamped, so a finding ties to a specific tool version and can be reproduced later?
When to use
Selecting, defending, or challenging any analytics tool whose output may be presented in court.
Example
The standard to aim for: another investigator, 18 months or more after the original analysis, could load the preserved inputs into the same tool version and reproduce the result, with the attribution showing why it was made, not just what it concluded.
Related
The Three Layers of Blockchain Evidence, Commercial Analytics & the Daubert Challenge, Five Evidence Preservation Principles, Court-Tested Attribution Evidence Types