How many
~30 minimum, 100–300 is the sweet spot for a trial.
Evaluation
Preparing a sample set for a BDS evaluation.
A short guide to assembling documents for a trial run. A well-chosen sample shows BDS at its best — sharp search, a clean entity graph, and useful entity profiles. Two things matter most: keep the set to one subject, and send enough documents to be representative.
One matter per set
~100–300 documents
Scanned PDFs welcome
No prep required
At a Glance
How many, what kind, what formats.
What kind
All about one case, matter, or topic — not a mix.
What formats
PDF (including scanned), Word, Excel, text, HTML.
Cohesion
Keep the set to one subject.
BDS learns the entities and relationships within your documents — which works best when the documents belong together. Before indexing, BDS automatically measures how cohesive your set is and reports it as Tight, Medium, or Diffuse; a diffuse set is flagged with a recommendation to split it.
Cohesive — ideal
Everything centers on a single subject:
- All filings and exhibits from one litigation or matter
- All reports and records on one incident or investigation
- A single custodian's email production
- Records for one company, deal, or program
Mixed — will be flagged
Several unrelated subjects in one set:
- Two unrelated cases combined into one folder
- Legal files + HR docs + product manuals together
- A general "everything we have" document dump
- Multiple clients or matters mingled
Have more than one subject? Send them as separate sets — BDS keeps each as its own project, which is exactly how it performs best.
Sizing
How many documents.
More is not always better for a trial. Search quality plateaus early; the entity graph keeps enriching up to a few hundred documents. This range shows the full picture without a long run.
0~30~300500+
Under ~30
Works, but the entity graph and search have little to draw on — fine for a quick look.
~100–300
The sweet spot. Search sharpens by ~60 documents; relationships keep building to ~250.
500+
Handled fine, but for a trial it mostly adds processing time, not new insight.
Formats & Sending
Send it as-is — we handle the rest.
Mixed formats and mixed document lengths in one set are fine. You do not need to convert or OCR anything — BDS reads scanned pages for you.
PDF · native or scanned
Word · .docx
Excel · .xlsx
Text · .txt
HTML
A Note on Sensitive Material
The trial runs on our hardware — send samples, not secrets.
For the evaluation, send non-privileged or de-identified documents.The trial runs on Venice Inference's hardware, so a representative, non-sensitive sample is enough to judge quality. Each evaluation runs in a dedicated encrypted workspace; when we deliver your results, that workspace and its encryption key are destroyed — your documents and every derived artifact (OCR text, search index, entity graph) — and we provide a signed certificate of destruction.
Privileged, regulated, or otherwise sensitive material belongs in production, not the trial. BDS is on-premise software: in a real deployment it runs on your own hardware, so your live corpus is processed in place and never leaves your control. The evaluation is a capability-and-fit check on a sample; production is where your confidential documents stay home.
Deliverables
What you get back.
From your sample set, a BDS evaluation produces:
Search you can question
Hybrid semantic + keyword search across every document, with grounded answers that cite their sources.
An entity knowledge graph
The people, organizations, and other entities in your set, with how they connect and co-occur across documents.
Entity profiles
A concise, sourced summary card for each significant entity, drawn from your own documents.
A cohesion & coverage read
The Tight / Medium / Diffuse cohesion result and a sense of how well the set size covers the subject.
In Short
Send a good sample; see what BDS can do.
One subject per set, roughly 100–300 documents, any mix of common formats, sent as-is — scanned PDFs included. Keep privileged material for an on-premise production deployment, and give us a representative sample for the trial.