The Verification Layer: Veracity-Engine
Whether your lawyers come to trust the AI you give them turns on one question: Can its work be checked?
VE (Veracity-Engine) checks every quotation, claim, characterization, and authority, in work written by a person or a model, against the document it rests on, and prints the record beside the work. VE verifies the answers your other systems produce as readily as the answers it writes itself.
A Blecher.ai project · Built by JDBlecher · Patents Pending
Lexis+ AI hallucinates 17%, Westlaw Precision AI 33%, GPT-4 43%, Ask Practical Law AI 17% (of what it dares to answer).
Zero hallucinations, 100 of 100 correct answers, (96/100 pre-PAD, 100/100 post-PAD).
refusals to answer. Ask Practical Law AI declined 62 of 100 questions to protect its score.
The benchmark ran with a deliberate handicap: a budget model we would never ship, reading from an overburdened context window. The verification layer (software) does the work. Veracity-Engine is domain and model-agnostic.
Every authority carries a chain of evidence, so the answer can be checked without taking anyone's word for it.
Each authority opens the same way: what was identified, where it was retrieved from, how the document's identity was confirmed, and what was checked against it. VE names the source and its tier, so a reader knows whether they are looking at the issuing body's text or somebody's copy of it.
VE catches the error a word-perfect memo still carries: the authority it relies on has been replaced
One API call can check a whole document, or an answer any other system produced, and returns the record the interface shows
Whatever your team builds, and whatever you already run, can call VE's API and get the record back. The responses carry the full verdicts, sources, audit ledger, and evidence chain.
It checks the answers your other systems produce as readily as the answers it writes itself.
A second endpoint verifies an answer produced anywhere else, by any AI, with seven components you can ask for individually.
The record arrives in Word as comments on the sentences, so nobody has to open anything new
The add-in puts Chat and Veracity in a task pane beside the document. The add-in writes each verdict back as a native Word comment, anchored to the sentence it checked, and a colleague without the add-in sees it in the ordinary review pane.
Nothing to learn: the review pane is Word's.
The prior art is an unattended black box: work nobody checked, from a process nobody saw.
VE is completely auditable: multi-step work is drafted as a plan, with human-in-the-loop checkpoints configured as desired.
Give it the directive. It explores, analyzes, executes, and reports back.
A step can be gated so a lawyer signs off before or after it.
1) Define the workflow, 2) Approve, 3) Come back to a finished product.
Proactive Authority Detection
Verification that runs without being asked.
PAD reads every answer the AI gives you, checking every quotation, claim, characterization, and authority before anyone thinks to ask. It also surfaces the authorities that bear on the question and were never cited: the gaps a reactive check never flags.
You will be able to trust the conversations you have with AI.
The wholesale replacement: it reads what Lexis and Westlaw read, checks what it cites, and drafts what you would sit and draft.
Pennies on the dollar, at the same depth, and faster than the manual read.
The whole of CourtListener, the CFR, advisory opinions, judges' individual rules, docket items, unreported decisions other services don't reach, and your own files on one connection, with no training required.
And if your firm keeps its Lexis or Westlaw subscription, VE reads through it.
Lexis suggests $60 a search. Westlaw, $99. A document outside your plan can run $150. And most firms run both: about $6,000 per lawyer per year.
Sources: LexisNexis Law Firm Per Search Pricing · Westlaw Predictable Pricing
The Sovereign Library: Every authority verified by any user becomes instantly available to every user, free, with full opinion text, treatment history, and the citing cases. The library is shared public law; your files stay yours.
Juxtaposition Analysis
Tactical deconstruction of the adversarial record.
Where two statements cannot both be true, they are put side by side and graded. Open one assertion and the record answers it, with the paragraph and page that does the answering. Where a letter contradicts both itself and the record, VE marks the row as a vulnerability and writes out the cost. The same grading runs wherever two documents disagree: a data room at odds with itself, a disclosure schedule that never qualified the rep, a guidance letter that moved, a filing that says it two ways, a contract whose sovereign changed the rules.
A whole case file, or a whole tranche, read at the depth of the one case worth all the marbles.
Pleadings, opinions, economic reports, hundreds of documents, from a zip file, a data room export, or inside Word: the whole file goes in at once. Every page is read at full depth. Nothing in the file goes unread.
The needle cuts both ways: the issue nobody spotted and the authority nobody cited, and the value nobody found.
Full Tabular Review: every document classified and read against the same questions, and the ones that matter lit
Ask it the way you would ask a colleague
The question is plain English, and the trail of sources is open before the answer finishes.
The privilege guarantee is enforced by hardware, and the platform operator has no path to access privileged material.
United States v. Heppner, No. 1:25-cr-00503-JSR (S.D.N.Y., Rakoff, J.; memorandum Feb. 17, 2026, reported at 820 F. Supp. 3d 292). Consumer-AI use defeated both privilege and work-product protection. The court turned on what the vendor's terms permit: collecting inputs, training on them, and disclosing them. And it expressly left open whether an enterprise-grade, zero-retention, no-training deployment produces a different result.
“It doesn't matter what your associates type. It matters where they type it.”
A case file uploaded to public AI is a case file placed in the public domain, privilege included: Veracity Engine runs inside infrastructure you control, so privilege is never questionable.
UK and R (Munir) v Secretary of State for the Home Department [2026] UKUT 81 (IAC); [2026] 4 WLR 37, the Upper Tribunal, verbatim:
“[T]o put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege…”
“Closed source AI tools which do not place information in the public domain … are available for tasks such as summarising without these risks.”
A document uploaded to a public AI tool transits a chain of third parties under contract. VE's answer under Sovereign Shield is that there is no chain: on hardware you control, there is no subprocessor list to object to because there are no subprocessors.
The whole file, crunched down to the work product.
The research conversation becomes a structured, verified record in one step, whatever the practice.
Connected once, the firm's material is there, and reach is chosen per question
Connected once, the firm's material is there. From the chat, a lawyer chooses how far to reach for any given question: this matter, their matters, or the whole library. Neither level waits on the other, and nothing here needs an administrator.
The organization's shelf: your precedent bank, policy library and closed matters, connected once and set by whoever owns it. The lawyer's desk: this matter's file, added by the person using it, no ticket required.
Nothing was trained: the material stays a document, so it can be cited, and removed.
You do not have to teach VE anything, and nothing has to happen to your data before it works
It shows you what it read; a model trained on your material cannot, and cannot forget a document when it turns out to be wrong.
Training is easy.
Citation format, fixed as you draft.
The day reconstructs itself.
Every entry drafted from the matter's activity. You review and approve.
The Complete Platform
Every capability your practice needs. One subscription.
Agentic Associate
Autonomous multi-agent execution
Shepardizing
Good law, bad law, citing cases
RAG
Your case, not generic law
Redline Review
Two-phase proofread + veracity
Deviation Detection
Flags non-standard contract terms
Intelligence Dividend
Verified authorities cost $0 forever
Document Editing
Real-time collaborative
Version Control
Word-level diff, auto-save
Citations Tab
Bluebook compliance, health score
Time & Billing
AI narratives, gap detection
Calendar & Deadlines
Nothing missed. AI augmented.
Knowledge Base
Firm-private, AI-searchable
RBAC & Sharing
Granular permissions
Matter Management
Entity graph, party index
Curation System
Conversation to brief
Dashboard Intelligence
Widgets, workflow tracking
Immediate Action Pad
Zero-friction AI chat
Export
Documents, conversations, PDF
