Who it’s for / General Counsel & Legal
Grotabyte for General Counsel and Legal Teams
The documents are rarely the hard part. What gets argued is your process: when preservation started, what it covered, how the review population was culled, and whether the method behind it survives being described out loud to the other side.
Rated 5 out of 5. “From 4–6 months to minutes.” — EMSL
Also for: eDiscovery Manager · Litigation Support
What you are accountable for
You are accountable for preserving everything a dispute touches from the moment it is reasonably anticipated, and for producing on a methodology you are willing to defend on the record.
What makes that hard
Preservation starts before the scope is known
A hold issued today has to cover material that arrives next month from custodians you have not thought of yet. A hold implemented as a snapshot of what exists at issue time is a spoliation exposure with a delay fuse.
The method is discoverable, not just the documents
Technology-assisted review gets challenged at meet-and-confer. If you cannot state the control set, the recall estimate and its interval, and what elusion testing found in the discard pile, you are defending a black box.
A redaction that is only a black rectangle
Draw a box over selectable text and the text is still in the file. Inadvertent production of privileged material is not a technical embarrassment; it is a waiver argument.
Budget is committed before the case is understood
Scope, custodian count and outside-counsel spend get decided early, on instinct, because the tooling cannot tell you anything until collection and indexing are already paid for.
What Grotabyte gives you
Holds that stay ahead of the matter
A legal hold is a standing predicate over custodians and dates, not a snapshot — material that arrives after issue and matches the predicate is covered by it. Custodians acknowledge their own notices, nobody attests on their behalf, and releasing a hold requires a second approver.
Early Case Assessment before any index is built
Custodian ranking, communication timelines, external exposure and off-hours anomalies come back before you commit to collecting and indexing — so scope, custodian list and budget are decisions rather than guesses.
TAR and CAL you can put on the record
Continuous active learning against a held-out control set, recall reported as a Wilson score interval rather than a bare percentage, and elusion testing run on the discard pile. The methodology is disclosable, which is the only kind worth using.
Redaction as removal, and a privilege log to match
A redaction removes text from the produced document rather than painting a box over words that remain selectable underneath. Every span carries one of seven stated bases — privilege, work product, PII, PHI, non-responsive, confidential, trade secret — feeding the privilege log, while the sealed native stays byte-identical.
Productions you can file, and a Term Hit Report for the negotiation
Concordance DAT and Opticon OPT load files, family-aware Bates numbering that continues across volumes, and a chain-of-custody report — with no document cap. The Term Hit Report gives you hit counts per proposed term before you agree to a search protocol.
One checkable thing
CCPOA cut eDiscovery time by 70% by running search, review and production against the archive itself rather than collecting into a separate review platform first.
Questions this role asks
If we issue a hold today, what happens to email that arrives next week?
It is preserved. The hold is stored as a predicate over custodians and date ranges, and it is evaluated against material as it enters custody, so anything matching it later is captured by the same hold without re-issuing anything. Retention cannot dispose of held material, and a release needs a second approver — one person cannot quietly lift a preservation obligation.
Opposing counsel challenges our TAR protocol. What can I actually put on the record?
The control set is held out from training, so recall is estimated against documents the classifier never learned from. Recall is reported as a Wilson score interval, which states the uncertainty instead of hiding it behind a single number, and elusion testing samples the discard pile to estimate what was left behind. Those are the figures that go into a protocol or an affidavit — the method is designed to be described, not to be defended by assertion.
How do you stop a redaction being lifted out of the produced file?
By not implementing it as an overlay. The produced text has the redacted span removed from it, so there is no hidden layer to extract, and each span carries one of seven stated bases that flows into the privilege log. The sealed native record is untouched and stays byte-identical, so the original is still available to the court if the withholding is contested.