Why benchmark at all
Two platforms can quietly disagree about how many documents exist. You want to find that before a hearing, not during one.
The corpus
The public Enron email corpus
Real corporate mail, in the public domain, obtainable by anyone who wants to repeat the run.
Whole mailboxes, per custodian
As published, per custodian.
What we measure
The processing funnel
Every stage from container to reviewable document, compared stage by stage rather than only on the end total.
Deduplication behaviour
What collapses across custodians, and whether attachments are kept and flagged rather than removed.
Search semantics
Proximity, noise words, hyphenation — where tokenisation returns a different set for the same words.
Chat and mobile messages
Reviewable units, conversation continuity, and how deleted-message provenance is represented on each side.
How a run works
Same input
The identical published mailbox goes into both systems.
Item-level bridging
Each document is matched to its counterpart, so a difference traces to items.
Classified differences
Design difference, accepted divergence, or defect. Defects get fixed.
Repeatable
Scripted and re-run when the pipeline changes, so a regression shows up.
Ask us for the data
Want it run on your own file types and languages? Ask.
We will walk your team through a full run, stage by stage. Our security posture covers where it all runs.