Bernard Product — AI Diligence for Private-Asset Decisions

Q: What does Bernard produce?

Bernard writes a Knowledge Pack: a source-linked decision record that preserves what was claimed, what evidence supports it, where sources disagree, what remains unproven, how confidence changed, and what action is justified next.

Decision pressure

Concrete questions. Short windows.

Bernard is built for the professional buyer. The principal benefits.

Postwar work on paper

Approve before the hold expires?

Provenance is thin. Condition is pending. Export and dispute checks have to land before Friday.

€180K · Friday

Lisbon hotel stake

Test before exclusivity closes?

Title, beneficial ownership, debt, licensing posture, revenue assumptions, and regulatory exposure sit in different places.

€6M · 7 days

Mayfair flat

Move on the offer today?

Ownership history, title notices, planning issues, price comparables, and counterparty risk need one defensible view.

£4.2M · Today

Research loop

Each evidence pass replans the next.

Bernard runs diligence as a working record. Each pass checks claims, adds proof, preserves conflicts, and keeps gaps visible until the next move is clear.

Knowledge Pack Claims, evidence, source lineage, conflicts, gaps, confidence, views, and history remain the working record.

01 Read the record
Subject, documents, current claims, evidence, gaps, and decision window.
02 Find gaps
Missing proof and weak claims are ranked by decision risk.
03 Choose checks
Targeted questions and primary-source paths are selected.
04 Test evidence
Sources are checked against the open claims and known conflicts.
05 Update confidence
New proof, contradictions, confidence changes, and citations enter the pack.
06 Name the next move
Proceed, pause, ask for proof, or keep an open gap visible.

Knowledge Pack

Confidence in the moment. Evidence for future inquiry.

A Knowledge Pack is the decision record Bernard produces. It preserves what was claimed, what evidence supports it, where sources disagree, what remains unproven, how confidence changed, and what action is justified next.

Decision context Subject, domain, sensitivity, deadline, and reviewer intent.

Claims Findings with confidence, status, evidence links, and relationships.

Evidence Every cited source remains traceable to its origin.

Contradictions Disagreements and missing proof stay visible instead of being smoothed over.

Confidence and gaps Confidence changes, unresolved questions, and open actions remain inspectable.

Readable outputs Summaries and views are generated from the record, not detached from it.

Example next action

Claim C014     Lender released artwork
Status         Unproven - release letter missing
Confidence     Low
Evidence       E021 (consignment file) · E022 (lender schedule)
Next move      Ask for proof before approval

The reviewer sees the recommendation and the reason for it. The next move can be challenged, updated, or cleared when proof arrives.

A Knowledge Pack is a folder of plain-text files (.kpack) in the KP1 format. It can be opened in Bernard, handed to a colleague, or read by any system that understands the format.

KP1

Inspectable by people. Parseable by systems.

KP1 is the plain-text format Bernard uses for Knowledge Packs. It is published as an open preview, with a public spec, examples, and a permanent identifier. It keeps the shape of a decision inspectable: claims, evidence, confidence, contradictions, relationships, and revision history in files people can read and systems can validate.

AI systems

Use structured claims, evidence, entities, conflicts, and history as reasoning context.

Professionals

Read views, risk summaries, source notes, and open actions while preserving the underlying record.

Organizations

Keep a reusable, inspectable record across handoffs, revisions, and later questions.

The public KP1 preview includes the file contract, examples, and permanent identifier for teams that need implementation details.

View the KP1 spec on GitHub Permanent identifier (Zenodo DOI)

Trust boundary

Local-first by design.

Local-first means the Knowledge Pack remains the durable record the user keeps. Each research pass works from a scoped slice of the pack and returns cited findings, not opaque summaries.

Local record .kpack files, source files, drafts, notes, and rebuildable indexes stay under the user's control.

Approved task The user chooses when research, parsing, sync, or assistant work should use context.

Scoped context Only the material needed for the active question is used.

Cited return New claims come back with evidence, conflicts, gaps, and next actions.

Source discipline

Three tools finding one article is still one source.

Bernard separates retrieval breadth from evidentiary strength. Repeated mentions of the same source do not raise confidence; independent source lineage can.

Shared origin

Repeated mentions of one underlying source are treated as one source.

Primary records

Filings, registries, records, databases, and source documents anchor claims where available.

Open gaps

Missing primary routes stay visible as gaps, open actions, or contradictions.

Markets

Art is the proving ground. Real estate is where scale begins.

Bernard starts where mistakes are public, evidence is fragmented, and tooling is decades behind. The research lens changes by asset class; the record stays one thing.

Art-market diligence

Provenance, condition, ownership, market context.

Target output: source-backed provenance view, condition gaps, comparable range, and questions before the hold expires.

Real-estate assets

Title, obligations, planning, comps, counterparty risk.

Target output: offer-readiness view, red and amber flags, comparable support, and next requests before the decision window closes.

Financing and counterparties

Authority, pledge status, identity, release proof.

Target output: proceed, pause, or ask for proof with every material claim tied to a source.

Infrastructure path

From research packs to deal rooms.

Knowledge Packs are the wedge: a decision record that can accumulate into signals, then become an exchangeable record between counterparties.

Current

Research Packs

Source-linked decision records for a specific consignment, acquisition, loan, or counterparty question.

Saved buyer requirements become a private market sensor for new evidence, changing risk, and better-timed action.

Future

Deal Room

A lockable, signable, exchangeable record that lets counterparties inspect the same claims, evidence, and open actions.

Plain answers

What Bernard is for.

Can someone read a Knowledge Pack without Bernard?

Yes. KP1 keeps the record in plain-text files so a colleague, reviewer, or system can inspect claims, evidence, contradictions, confidence, gaps, and next actions without treating the output as a black box.

What does Bernard produce?

Bernard produces Knowledge Packs: source-linked decision records containing claims, evidence, contradictions, confidence, gaps, and next actions.

Who makes the final call?

The responsible professional does. Bernard supports professional diligence by organizing evidence, claims, confidence, contradictions, and open actions. It does not replace legal, fiduciary, appraisal, authentication, compliance, or other professional judgment.

Professional judgment

Where Bernard ends, judgment begins.

Bernard runs the research. The professional makes the call.

Bernard finds facts, connects sources, tracks uncertainty, and preserves what still needs proof. Bernard supports professional diligence by organizing evidence, claims, confidence, contradictions, and open actions. It does not replace legal, fiduciary, appraisal, authentication, compliance, or other professional judgment.

If you want to try Bernard on a real consignment, transaction, or counterparty review, we are accepting a small number of early partners.

Request Early Access

Diligence with a defensible record.