# Legalnaut — full content A case intelligence workspace for litigation teams and their customers. Every fact and contradiction anchored to the document it came from. Operated by A2Z WEB PTE. LTD., Singapore. --- ## Features URL: https://legalnaut.com/features Advanced language models read every page once: OCR, transcription, dates, people, claims, and the comparison of every claim against the rest of the file. The reading a paralegal would bill for happens in the background, so the hours go on judgement rather than on looking things up. ### Read the file once Every page goes in once. Scans, mail and recordings too. An archive, a folder or a forwarded mailbox travels one road. What arrives as an image is read by OCR, what arrives as a recording is transcribed, and nothing is analysed twice. - **Archives and mailboxes, not one file at a time** Drop in a ZIP of the whole file, or forward mail to the address the workspace gives you. Formats are recognised by content, so a badly named file still lands where it belongs, and an attachment is filed as a document of its own. - **OCR on every page, with no ceiling** A scanned bundle is read to its last page. Where the text comes out below 70% confidence the document says so, and you know which pages to check against the original. - **Chat screenshots read as conversations** A WhatsApp or Messenger screenshot becomes a transcript of who wrote what and when, shown as a thread rather than as a wall of OCR. - **Recordings become searchable text** Hearings and voice messages are transcribed in the background. A minute of audio counts as one page, and the transcript costs nothing more. - **Duplicates never cost a page** Every file is fingerprinted with sha256 before anything runs. A file already in the case is recognised and skipped, whatever it is called. - **Authenticity checks on e-mail and photographs** An e-mail is checked against its DKIM signature and its received chain, a photograph against its EXIF data. Each check reports verified, failed or unavailable, and unavailable is shown as exactly that. - **The cost before the import** The import screen counts the pages before it starts, so you see what a folder will take from the monthly pool before you commit it. ### Know what happened The chronology the documents wrote, not the one you remember Dates, people and events come out of every document and onto one timeline. Each entry keeps the document, the page and the sentence it came from. - **Many events from one document** A forty-page pleading that describes twelve dated events puts twelve points on the timeline, each with its quotation. - **Your reading, kept with the record** Mark an entry key, relevant or background; disputed or not; favourable, adverse or neutral. Those marks are yours, never a model's, and they drive the exports. - **Gaps in the material** The chronology export prints every stretch of ninety days or more with no dated item, so a silence in the file is visible before the other side points it out. - **Who said what about whom** A profile for every person and organisation: their statements, what others said about them, the documents they wrote, the events they took part in. Two spellings of one name merge as an overlay, and the source index stays as it was. - **Correspondence, measured** Who wrote to whom, how often, how long a reply took, and every silence of thirty days or more. Counted from the headers, not interpreted. - **What relies on what** When a letter refers to an earlier document, the two are linked. The document page shows what it relies on and what relies on it, which is how a missing attachment gets noticed. - **Quotations in the original language** Summaries and titles are written in your working language. A quotation is never translated, because a translated quotation stops being a quotation. ### Find where the accounts disagree Every claim, checked against the rest of the file Statements are broken into claims, and each claim is compared with the whole file in both directions. Where two documents disagree, both are shown with their quotations. - **Two quotations, side by side** A contradiction is two quotations side by side, each with its document and page, and a sentence on why they conflict. Confirmations and context are kept as well, because a claim with independent support is evidence in your favour. - **The software suggests, a lawyer confirms** Each pair is a suggestion until you confirm or dismiss it. A dismissed pair never comes back, and only a confirmed one reaches a draft, a report or a chat answer as a fact. - **A party cannot corroborate itself** A later letter from the same author does not count as confirmation of their earlier claim. Those pairs are marked and kept out of the lists. - **You can see how far the check went** The document page shows how many candidate pairs were examined and offers to examine more. A pair already judged is never sent to a model again. - **Near-duplicates flagged, not merged** A later version of the same letter, or the same letter filed twice, is badged on the document and left for you to decide. - **The case against you, in writing** Devil's advocate writes the strongest opposing argument from your own matter: the weakest exhibits, the confirmed contradictions that hurt, the documents the other side will ask for. ### Prove it From issue to exhibit to gap, on one chart A proof chart puts each issue you have to establish next to the facts and exhibits that carry it, and shows where nothing does yet. Everything drafted from it is checked against the file before you see it. - **The proof chart** Define the issues and who bears the burden. Attach a document as supporting or undermining. An issue with nothing behind it is marked as a gap. Suggestions come from your own documents, and nothing attaches without a click. - **An evidentiary motion from the chart** Each exhibit with the fact it is offered to prove, in the form the court expects. Each gap becomes a request for production. - **A citation check on every AI text** Every draft and every chat answer is checked against the records: the document exists and you may see it, the page exists, a quotation of four words or more appears in a document cited in the same paragraph, a case number is one the file knows. Anything else is reported, never quietly repaired. - **Pin-cites to the page** A claim, a contradiction or a chat answer cites the page. One click opens the document on that page with the passage highlighted. - **Exhibit numbers that stay put** A matter numbers its exhibits once, with your prefix, and every export uses those numbers. - **Bundles with an integrity annex** The evidence bundle is one PDF: an index, every exhibit converted and merged, a bookmark per exhibit, and an annex listing each file's sha256 and authenticity result. - **Chronology, statement of facts, case report** The working chronology as DOCX, a statement of facts written from your key and relevant entries with pin-cites, a case report as PDF. Built from the records without a model, so they cost no pages. - **A redacted copy for a third party** A pseudonym registry per matter turns names into roles in every export, so a bundle can go to an expert or a mediator without the identities. The records underneath are unchanged. ### Prepare the hearing Walk in knowing the witness and the deadline Everything the file holds about one person, gathered for the examination, and every procedural time limit computed from the provision and the date of service. - **The witness dossier** For one person: their statements, what was said about them, the documents they wrote, the events they took part in, and the statements the records contradict. Every line with document, page and quotation, assembled without a model. - **An examination outline from the dossier alone** Background to lock in, facts that help, facts that hurt, an impeachment sequence for each inconsistency, and the questions the file cannot answer. Every question points at a line in the dossier, and an unverified inconsistency is listed as a risk. - **Reminders and a calendar feed** Each deadline sends a reminder in the recipient's language at the workspace's own hour, and the whole list exports as a calendar feed. - **A matter briefing on demand** Posture, what changed, next steps, gaps and risks, written from the matter's own material and marked as a working draft. ### Keep the client informed The client reads the same file, in plain language The people paying for the case can see where it stands without a phone call: an update written for them, a viewer role that cannot change anything, and a log of who looked at what. - **An update in plain language** One click drafts a status update for the client from the matter's material: what happened, what is next, what is waiting on them. Factual, not predictive, and marked as a draft for you to send. - **A seat for the client, at no cost** Add the client as a viewer. They read and export, and cannot edit or delete. Sensitive material stays out of their view. - **Every look is logged** Views, downloads, exports and deletions are written to an audit log that the people it records cannot edit. - **Nine interface languages** The interface runs in English, Polish, German, French, Spanish, Italian, Portuguese, Dutch and Ukrainian, so a client abroad works the same file. The evidence stays in its original language. ### Ask the file Ask a question. It searches, reads and cites before it answers. The assistant does not answer from a summary. It has tools on the records, uses them in front of you, and every sentence it writes names the document and the page it came from. - **Four ways to search** Full-text over the documents, with filters for type, category, jurisdiction and date. Search by meaning, so a paraphrase in another language still finds the document. Search in the extracted text of attachments, scans and recordings. Search in the documents the team added after the file arrived. - **It reads the whole document, not a snippet** A long document is read to the last page, window by window. A mail thread is read in order, message by message. An attachment is opened in full, whether its text came from OCR or from a transcript. - **It knows the shape of the file** Events on the timeline by date, actor or jurisdiction. Claims by speaker and by kind: accusation, admission, denial, threat, promise. A profile of any person by name or alias. Contradictions, each with a note on whether a lawyer has confirmed it. The documents tagged with a given provision, marked as a hypothesis for a lawyer to check. The counts for the whole case. - **A count comes from the database, not from memory** When a question needs a number, the assistant runs one read-only query on the case tables. A document the team has hidden, or one the user may not see, is absent from every table the query can reach. - **Answers that name their sources** Every document it relies on appears as a citation with the page, and the citation is a link that opens the document on that page. The line under the message says how many of the citations were verified against the file. - **Ten questions to start from** Summarise the case. What did a person say about a topic. Who is this name. What happened in a period. A chronology of a topic, the contradictions about it, the evidence for an assertion, the correspondence with a person, a document in full, a search of the attachments. A short form fills in the name or the period, and the question lands in the box for you to sharpen before it is sent. - **It files and takes notes on request** It can add a document, claim, event or contradiction to a matter with a note on why it matters, and leave a note for the team on any of them. A note it writes is marked as the assistant's and is visible to everyone on the file. Those are the only two things it can write. - **The rules it answers under** It answers in the language of the file unless you ask for another. It calls the parties by their neutral procedural designations. A legal qualification is marked as needing a lawyer. Example questions: Summarise the case; What did [person] say about [topic]?; Who is [name]?; What happened in [period]?; Timeline of [topic]; Contradictions about [topic]; Evidence for [assertion]; Correspondence with [person]; Document [ID] in full; Search the attachments. ### Where we do it differently - Instead of: A price for every seat, so the team shares one login — Unlimited users on every plan. Add the client, the paralegal and the expert. - Instead of: Storage billed by the gigabyte — Storage is not metered. Keep the scans and the recordings in the file. - Instead of: Credits, tokens or a monthly AI allowance to run out of — One unit: a page of analysis, counted once when a file enters the case. Chat, search, exports, drafts and dossiers are not on the meter. - Instead of: A price on a quote, after a sales call — Published prices, a 14-day trial with no card, and a card only when you decide to pay. - Instead of: A surcharge for OCR and transcription — A scanned page and a minute of recording cost the same page as a typed one. - Instead of: AI output that rewrites the record — The text, the path and the checksum of a document are fixed on arrival. What a model writes lives in separate columns and can be regenerated without touching the evidence. - Instead of: AI findings presented as facts — A finding is a suggestion until a lawyer confirms it, and every AI text is checked against the file before you see it. - Instead of: Files deleted on a retention schedule — Nothing is deleted on a schedule. If a plan lapses the workspace turns read-only, and reading, searching and exporting stay open. - Instead of: An import refused when the quota is hit — A file the pool cannot cover waits, stored and unbilled, and starts on its own when pages arrive. Running out mid-case costs one pack of 5,000 pages, on any plan. ### What it will not do - It does not give legal advice, and it does not decide what a document means. That stays with the lawyer. - It does not file anything with a court. - It is not an e-discovery platform. There is no privilege review and no review batching. - An AI finding is a suggestion until a lawyer confirms it, and every generated document opens with the words "working draft". - Statutory holidays are not applied to deadlines. The provision and the arithmetic are shown, so a holiday end is yours to correct. --- ## Pricing URL: https://legalnaut.com/pricing Free trial: 14 days, no card. Annual billing saves 20%. Prices in USD. Money-back guarantee: 30 days from the first payment on a first subscription, refunded in full on request to contact@legalnaut.com. The unit is a page of analysis: one page of a PDF or scan, about 400 words of e-mail or Word text, or one minute of a recording — counted once, when a file first enters the case. Users, storage, chat, search, exports and retention are never metered. Extra pages: a pack of 5,000 pages for $129, valid 12 months, on any plan. ### Solo — $79/month ($63/month billed annually) One case file, read to the last page. - 1 case - 2,500 pages of analysis a month - Unlimited users - Extra pages any time: 5,000-page pack for $129 - Chronology, entities, contradictions, deadlines - Chat, search and exports without a meter ### Practice — $299/month ($239/month billed annually) For a practice running several client files at once. - 5 cases - 10,000 pages of analysis a month, shared across cases - Unlimited users - Extra pages any time: 5,000-page pack for $129 - Everything in Solo - Nothing crosses between two client files ### Firm — $799/month ($639/month billed annually) For a firm that opens a new file every week. - Unlimited cases - 30,000 pages of analysis a month, shared across cases - Unlimited users - Extra pages any time: 5,000-page pack for $129 - Everything in Practice - Archive import and onboarding done with you ### Enterprise — Contact us Your own installation, your own AI keys, your own rules. - Unlimited cases - Pages and cases set with you - Unlimited users - Runs on your servers with your model provider keys - Every AI cost visible, down to the document - Security review and DPA negotiation - Named contact for onboarding --- ## Legalnaut vs uploading your case file to ChatGPT URL: https://legalnaut.com/compare/general-ai-assistants General assistants are genuinely good at reading documents. If you have thirty pages and a question, pasting them into ChatGPT or Claude will often get you a useful answer in a minute, and there is no reason to buy software for that. The difficulty starts at the scale a real case file reaches, and at the point where the answer has to survive being checked. ### When Legalnaut is the better fit - The file is larger than one conversation, and you will ask about it more than once. - An answer has to name the document it came from, because someone will ask. - Scans and photographs make up a real share of the file. - More than one person works the same case file. ### When a general AI assistant is the better fit - You have a handful of documents and a one-off question. - You want drafting help rather than a searchable record. - The work is research about law rather than about a specific file. --- ## Legalnaut vs an e-discovery platform URL: https://legalnaut.com/compare/ediscovery-platforms E-discovery platforms are serious engineering, built for disputes where the document population runs into the millions and the review cost dominates the case budget. On that ground they have no substitute, and this page is not going to pretend otherwise. They are built around a different question. Review platforms sort a large pile into responsive and privileged. They are not built to tell you what happened in what order, or to put a witness statement next to the invoice that contradicts it. ### When Legalnaut is the better fit - The file is large enough to lose things in, and small enough that per-gigabyte pricing makes no sense. - The work is understanding the dispute rather than producing disclosure. - You want the chronology, the entity profiles and the contradictions, not a review queue. ### When an e-discovery platform is the better fit - You are producing disclosure at scale and need a defensible review workflow. - Privilege review across a large team is the main task. - The matter requires processing formats and volumes that only a dedicated platform handles. --- ## Legalnaut vs a document management system URL: https://legalnaut.com/compare/document-management-systems A document management system is the filing cabinet a firm runs on. Version control, matter-centric filing, retention, integration with the mail client. If your firm has one, it is not going anywhere, nor should it. A DMS knows where a document is filed. It does not know what the document says, which is a different job and a harder one. ### When Legalnaut is the better fit - You can find the document but not the sentence. - Preparing a hearing means rereading a file you have already read. - The question is what happened, not where the file is. ### When a document management system is the better fit - You need firm-wide filing, versioning and retention policy. - Compliance requires records management the DMS already provides. - The problem is filing discipline rather than case analysis. --- ## Legalnaut vs reading the file yourself URL: https://legalnaut.com/compare/manual-review The lawyer who has read every page of a file is better than any software, and this page is not going to argue with that. Judgement about which fact matters is not a thing to hand over. The difficulty is not the first reading. It is the second, when someone asks a question about a document you read three weeks ago and you have to find it again in nine hundred emails. ### When Legalnaut is the better fit - The file has grown past the point where one person holds it in their head. - More than one person needs the same view of it. - Disclosure keeps arriving and the chronology keeps going stale. ### When reading the file yourself is the better fit - The file is small enough to reread in an afternoon. - The matter is closing and nothing new is coming in. --- ## Glossary ### Chronology URL: https://legalnaut.com/glossary/chronology A chronology is a dated list of the events in a dispute, each one traceable to the document that evidences it. A chronology is the backbone of a case. Counsel builds one to see the sequence of events plainly, to spot the gap where nothing happened, and to hand the court a single ordered account instead of a bundle. What separates a working chronology from a list of dates is provenance. Every entry names the document it came from, so an opponent challenging an entry can be answered by pointing at the source rather than by memory. Built by hand, a chronology of a thousand-document file takes days and goes stale the moment new disclosure arrives. Built from the documents themselves, it updates when the file does. ### Contradiction URL: https://legalnaut.com/glossary/contradiction A contradiction is a pair of statements in the case file that cannot both be true, each one anchored to the document it appears in. A witness statement puts a meeting in June. An invoice for the same work is dated March. One of them is wrong, and which one is wrong matters. Contradictions are findings, not errors. They are the raw material of cross-examination, and they are the reason a case file is worth reading in full rather than in summary. Finding them by hand means holding a thousand documents in your head at once. Finding them mechanically means comparing every claim against every other claim and putting the surviving pairs in front of a lawyer to judge. ### Exhibit URL: https://legalnaut.com/glossary/exhibit An exhibit is a document formally put before a court, carrying a reference number that stays fixed for the life of the proceedings. The number is the point. Once a document is EX-041, every skeleton argument, witness statement and transcript refers to it that way, and renumbering mid-case creates the kind of confusion that costs a hearing. Exhibit numbering is why case software has to treat numbering as data rather than presentation. An export that renumbers exhibits to suit its own layout is worse than useless. ### Bundle URL: https://legalnaut.com/glossary/bundle A bundle is the paginated set of documents filed for a hearing, indexed so any page can be found from the index in one step. Courts have rules about bundles, and they are not decorative. Pagination has to be continuous, the index has to match, and the copy in front of the judge has to be the copy in front of counsel. Assembling one means converting every source format to PDF, keeping the exhibit order, generating the index from the actual contents, and doing it again when a document is added the night before. ### Provenance URL: https://legalnaut.com/glossary/provenance Provenance is the record of where a piece of evidence came from and what has happened to it since. For a document in a case file, provenance answers: which file did this text come from, who supplied it, when did it enter the file, and has anything changed since. A checksum makes the last question answerable rather than assertable. If the sha256 of the file today matches the sha256 recorded when it arrived, the file has not been altered, and that is demonstrable rather than promised. ### Checksum URL: https://legalnaut.com/glossary/checksum A checksum is a short fingerprint computed from a file, which changes if even one byte of the file changes. sha256 is the one most commonly used for evidence. Two files with the same sha256 are the same file; two files with different sha256 values are not, however similar they look. In practice it does two jobs. It proves a document has not been altered since it entered the file, and it stops the same document being imported twice under two different names. ### Chain of custody URL: https://legalnaut.com/glossary/chain-of-custody Chain of custody is the documented sequence of hands a piece of evidence has passed through, from collection to court. A break in the chain is an opening for the other side. If nobody can say who held a document between collection and disclosure, its integrity is arguable. Software cannot supply a chain of custody on its own, but it can record the part that happens inside it: when a document arrived, by which route, who has opened it, and whether its content has changed since. ### Audit log URL: https://legalnaut.com/glossary/audit-log An audit log is an append-only record of who did what in a system and when. For a case file the log matters most when something goes wrong: a document is missing, an export went to the wrong recipient, a deletion is disputed. The property that makes a log worth having is that it cannot be edited by the people it records. A log a user can rewrite answers no question at all. ### E-discovery URL: https://legalnaut.com/glossary/e-discovery E-discovery is the process of identifying, collecting and producing electronic documents for litigation. The discipline grew out of large commercial disputes where the document population runs to millions and the cost of review is the dominant cost of the case. Most e-discovery platforms are built for that scale and priced for it. They answer "which of these million documents are responsive and not privileged". They are not built to answer "what happened, in what order, and where do the accounts disagree". ### Privilege review URL: https://legalnaut.com/glossary/privilege-review Privilege review is the check that no legally privileged document is handed to the other side during disclosure. A privileged document produced by mistake can be unrecoverable. That risk is why review is slow and why it stays a human judgement rather than an automated one. Tools help by narrowing the pile: flagging correspondence with counsel, grouping by sender, surfacing near-duplicates so one decision covers a family of documents. The decision itself stays with a lawyer. ### Redaction URL: https://legalnaut.com/glossary/redaction Redaction is the permanent removal of protected text from a document before it is disclosed or filed. The failure mode is famous: a black rectangle drawn over text that is still selectable underneath. A redaction that can be undone by copy and paste is not a redaction. Anything that removes text has to remove it from the file itself, not from the way the file is displayed. ### Anonymisation URL: https://legalnaut.com/glossary/anonymisation Anonymisation replaces identifying details in a document with labels, so the substance can be read without exposing who it concerns. It is what lets a case file be discussed in a training set, a precedent note or a screenshot without naming the parties. Doing it by pattern matching alone breaks on inflected languages and on names that are also ordinary words. Doing it from the known party names, in every spelling and case form, is what makes it reliable. ### OCR URL: https://legalnaut.com/glossary/ocr OCR turns an image of text, such as a scan or a photograph of a page, into text a computer can search. Court files are full of documents that exist only as scans, often photocopied several times over. Without OCR those documents are invisible to search, which means they are invisible to whoever is preparing the case. Quality varies with the source. A clean scan produces near-perfect text; a fax of a fax produces fragments. Knowing which documents came out unreadable is as useful as the text itself, because those are the ones that still need a human eye. ### Transcription URL: https://legalnaut.com/glossary/transcription Transcription turns recorded speech into text, so a recording can be searched, quoted and cited like any other document. Voice notes, call recordings and hearing audio carry evidence that never reaches the written file otherwise. For evidential use the transcript has to keep timestamps, so a quotation can be checked against the recording at the point it was said. ### Entity extraction URL: https://legalnaut.com/glossary/entity-extraction Entity extraction pulls the people, organisations, places and dates out of a document and records where each one appeared. Once extracted, the question "everything this witness said, and everything said about them" becomes answerable in one step rather than by rereading the file. The hard part is identity, not detection. The same person appears as a full name, a surname, an initial, an email address and a misspelling, and treating those as five people produces a profile that is worse than none. ### Claim URL: https://legalnaut.com/glossary/claim A claim, in case analysis, is a single assertion of fact made in a document, recorded together with who made it and where. Breaking documents into claims is what makes a file comparable with itself. Two claims about the same event, from two documents, can then be put side by side. A claim is not evidence on its own. It is a pointer to the sentence it came from, which is the thing that gets quoted. ### Deduplication URL: https://legalnaut.com/glossary/deduplication Deduplication removes repeated copies of the same document from a case file, keeping one. A disclosure set collected from several mailboxes contains the same email many times over. Reviewing it many times over costs money and produces inconsistent decisions. Matching on checksum catches exact duplicates cheaply and before any processing cost is incurred. Near-duplicates, such as the same letter with a different footer, need a different test and a human decision. ### Retrieval-augmented generation URL: https://legalnaut.com/glossary/rag Retrieval-augmented generation is a technique where a language model is given relevant documents to read before it answers, instead of answering from memory. It is the difference between asking a model what it recalls about a topic and asking it to answer from a specific file in front of it. For legal work the retrieval half is what matters. An answer is only as good as the documents that were fetched to support it, and an answer that cannot name those documents cannot be checked. ### Hallucination URL: https://legalnaut.com/glossary/hallucination A hallucination is a confident, fluent statement from a language model that is not supported by any source. Courts have sanctioned lawyers for filing submissions containing invented citations: the AI Hallucination Cases database kept by Damien Charlotin at damiencharlotin.com listed more than 2,000 such decisions worldwide as of September 2026. The risk is not theoretical and it is not rare. The defence is structural rather than aspirational. If every answer carries the document it came from, an unsupported claim has nowhere to hide, and the reviewer can see immediately that a citation is missing. ### Proof chart URL: https://legalnaut.com/glossary/proof-chart A proof chart is a table that sets each issue a party has to establish against the facts and exhibits that prove it, and shows where nothing does yet. Before a pleading and before a hearing, counsel asks the same two questions: what do I have, and what am I missing. A proof chart answers both on one page, issue by issue, with the burden of proof marked and every exhibit either supporting or undermining. Built by hand it lives in a spreadsheet and goes stale as disclosure arrives. Kept with the case file, it is the source for the evidentiary motion: each exhibit with the fact it is offered to prove, and each gap as a request for production. In Legalnaut the chart belongs to a matter. Suggestions come from the matter's own documents, nothing attaches without a click, and an issue with nothing behind it is flagged as a gap. ### Pin-cite URL: https://legalnaut.com/glossary/pin-cite A pin-cite is a citation that points to the exact page of a document, not only to the document. A citation to a two-hundred-page bundle is an invitation to read the bundle. A pin-cite names the page, so the reader, the opponent and the court can check the sentence in seconds. Getting the page right depends on the text carrying its page from the moment it is extracted. Legalnaut writes a page marker into the extracted text of every scanned or PDF document, so a claim, a contradiction, a chat answer and a draft can all cite the page, and one click opens the document there with the passage highlighted. ### Cite-check URL: https://legalnaut.com/glossary/cite-check A cite-check is the verification that every citation in a document points at a real source that says what the document claims it says. Cite-checking used to be a junior's afternoon with the bundle. With language models drafting, it is the step that separates a usable draft from a sanction, because a fluent draft can cite a document that does not exist. Legalnaut runs a deterministic check on every AI text, with no model involved: the cited document exists and the reader may see it, the cited page exists, a quotation of four words or more appears in a document cited in the same paragraph, and a case number is one the file knows. Anything that fails is reported under the draft or the chat message, never repaired quietly. ### Witness dossier URL: https://legalnaut.com/glossary/witness-dossier A witness dossier is everything a case file holds about one person, gathered for the examination: their statements, what was said about them, the documents they wrote, and the statements the file contradicts. Preparing to examine a witness means finding every line they wrote or said, every line said about them, and every place their account disagrees with the documents. In a large file that is days of rereading. Legalnaut assembles the dossier from the records without a model, each line with its document, page and quotation, and exports it as a document. The examination outline is then written from the dossier alone, so every proposed question has a line to point at. ### Limitation period URL: https://legalnaut.com/glossary/limitation-period A limitation period is the deadline after which a claim can no longer be brought. Miss it and the claim is gone regardless of its merits, which is why limitation is the deadline lawyers track most carefully. Deadlines in a case file arrive buried in correspondence and orders rather than in a calendar. Getting them out of the documents and into a calendar, with the document that set them attached, is the whole job. --- ## About Legalnaut URL: https://legalnaut.com/about Legalnaut exists because of a case file that got too big to hold in one head. Nine hundred emails, several years, two jurisdictions, and a recurring problem: the document you needed was one you had already read, and finding it again took an afternoon. Search was not the answer, because you rarely remember the filename. You remember what the document said. ## What we build A workspace that reads a case file once and then keeps it navigable. A chronology assembled from the documents. Profiles of the people and organisations in the file. Contradictions surfaced as pairs, each side attached to the document that evidences it. Every finding names its source. That is not a feature, it is the condition on which any of this is usable in legal work. ## What we do not build We do not give legal advice, decide what a document means, or produce anything that should go into a filing without a lawyer reading it first. Drafts are labelled as drafts. The judgement stays where it belongs. ## Who we are Legalnaut is operated by **A2Z WEB PTE. LTD.**, registered in Singapore under 202614429R, at 7 Temasek Boulevard, #12-07 Suntec Tower One, Singapore 038987. Our infrastructure is SOC 2 Type II certified. Write to us at [contact@legalnaut.com](mailto:contact@legalnaut.com). --- ## Privacy policy URL: https://legalnaut.com/privacy This policy covers legalnaut.com and the Legalnaut application. The controller is **A2Z WEB PTE. LTD.**, 7 Temasek Boulevard, #12-07 Suntec Tower One, Singapore 038987, registration 202614429R. ## 1. What we collect **Account data.** Name, email address, password hash, workspace membership and role. **Case material.** The documents you import, the text extracted from them, and the analysis derived from that text. We treat this as confidential material belonging to you, not to us. **Usage data.** Server logs, error reports, and a record of consequential actions in the audit log. **Billing data.** Handled by Stripe. We never see or store card numbers. ## 2. Why we process it To provide the service you have subscribed to, to bill you for it, to keep the service secure, and to meet our legal obligations. Case material is processed only to provide the service. ## 3. Sub-processors | Processor | Purpose | Location | | --- | --- | --- | | Language model providers | Document analysis and chat answers | EU / US | | Cloud hosting and object storage | Application hosting, file storage | EU | | Amazon Web Services (Textract) | OCR when local extraction fails | EU | | Stripe | Subscription billing | US / EU | | Zoho Mail | Inbound document forwarding | EU | Our agreements with every model provider prohibit training on customer content. ## 4. Retention Case material is kept for as long as your workspace exists. Nothing deletes it on a schedule. If a subscription lapses, the workspace becomes read-only; clearing it is a deliberate action taken on request or after a documented period, never automatically. Invoicing records are kept for five years under Singapore tax law. Server logs are kept for 90 days. ## 5. Your rights If the GDPR applies to you, you have the right to access, rectification, erasure, restriction, portability and objection, and the right to complain to a supervisory authority. Write to [contact@legalnaut.com](mailto:contact@legalnaut.com) and we respond to verified requests within 30 days. ## 6. Security TLS 1.2 or better in transit. Encryption at rest for databases and backups. Hashed passwords. Optional two-factor authentication. Separate database and file storage per workspace. Access to production restricted on the principle of least privilege. SOC 2 Type II certified infrastructure. No transmission over the internet is completely secure. If you believe an account has been compromised, write to us immediately. ## 7. Cookies We set a session cookie needed to keep you signed in, a cookie recording your theme choice, and one recording whether the sidebar is collapsed. None of them track you across sites. Analytics, if enabled, runs only with your consent. ## 8. Changes Material changes are announced in the application before they take effect. --- ## Terms of service URL: https://legalnaut.com/terms These terms govern your use of Legalnaut, provided by **A2Z WEB PTE. LTD.**, Singapore, registration 202614429R. ## 1. The service Legalnaut is software for organising and analysing case documents. It is a tool used by legal professionals. **It does not provide legal advice, does not create a lawyer-client relationship, and its output is not a substitute for professional judgement.** ## 2. Your material You keep all rights in the documents you import and the analysis derived from them. We process that material only to provide the service. We do not use it to train models and we do not disclose it, except where compelled by law. ## 3. Your obligations You confirm you are entitled to process the material you import, including any personal data in it. You are responsible for the accounts of people you invite to your workspace, and for what they do with them. ## 4. AI-generated output Summaries, chronologies, extracted claims and drafts are produced with the help of language models. They can be wrong. Every one of them shows the source it was drawn from so it can be checked, and checking it before use is your responsibility. ## 5. Fees and refunds Plans are billed in advance, monthly or annually, through Stripe. Trials require no card. You can cancel at any time from the billing screen; the plan runs to the end of the paid period and does not renew. **Money-back guarantee.** If you tell us within 30 days of the first payment on your first subscription that Legalnaut is not right for you, we refund that payment in full. Write to contact@legalnaut.com. The guarantee covers the first payment on a first subscription and does not apply to later renewals, which you can avoid by cancelling before they fall due. Nothing here limits any statutory right of withdrawal you may have. **Pages of analysis.** A plan includes a monthly number of pages of analysis. A page is one page of a PDF or scanned document, approximately 400 words of an e-mail or word-processed document, or one minute of an audio or video recording. Pages are counted once, when a file first enters the workspace; a duplicate of a file already in the workspace is not counted. Unused pages from the monthly allowance do not carry over. Page packs bought separately are valid for 12 months from purchase and are used after the monthly allowance. A file the allowance cannot cover is stored and analysed once pages become available; it is not analysed or charged before then. **Users and storage.** Plans do not limit the number of users. Storage is included for the ordinary needs of the case files in the workspace; we may contact you about a workspace whose storage is far out of proportion to its pages of analysis. ## 6. Availability We aim for continuous availability but do not guarantee it. Maintenance is announced where practical. ## 7. Liability To the extent permitted by law, our aggregate liability is limited to the fees you paid in the twelve months before the claim. Nothing here excludes liability that cannot lawfully be excluded. ## 8. Termination Either party may terminate at any time. On termination your workspace becomes read-only and you may export your material. ## 9. Governing law Singapore law, with the courts of Singapore having exclusive jurisdiction. --- ## Data processing agreement URL: https://legalnaut.com/dpa This agreement forms part of the [terms of service](/terms) where the GDPR or a comparable law applies to your use of Legalnaut. You are the controller; **A2Z WEB PTE. LTD.** is the processor. ## 1. Subject matter Processing of personal data contained in documents you import into a workspace, for the purpose of providing the service, for as long as your subscription runs. ## 2. Categories of data and data subjects Whatever your case material contains. In litigation that routinely includes names, contact details, employment details, financial information and correspondence, and may include special categories of data. Data subjects are the parties, witnesses and third parties appearing in your documents. ## 3. Our obligations We process personal data only on your documented instructions, keep it confidential, apply the security measures described on the [security page](/security), assist you with data subject requests, and notify you without undue delay of any personal data breach affecting your material. ## 4. Sub-processors You authorise the sub-processors listed in the [privacy policy](/privacy). We give notice before adding a new one, and you may object. ## 5. International transfers Transfers outside the EEA are made under Standard Contractual Clauses. Executed processing agreements are in place with every model provider, each prohibiting the use of customer content for training. ## 6. Deletion and return On termination you may export your material. We delete it on your written request. Nothing is deleted automatically. ## 7. Audit We make available the information needed to demonstrate compliance, including our SOC 2 Type II report under NDA on request. --- ## How to find contradictions in witness statements URL: https://legalnaut.com/blog/find-contradictions-between-statements-and-documents Two witnesses describing the same meeting differently is normal and rarely worth much. A witness describing a meeting one way while an email sent that afternoon describes it another way is a different thing, and it is the thing worth the search time. ## Compare claims, not documents Comparing documents is the wrong unit of work. Documents are long, they cover several subjects, and two documents that overlap on one point will differ on twenty others for no interesting reason. Break the file into claims instead. One claim is one assertion of fact, recorded with who made it and which sentence it came from. Now comparison is tractable, because you are asking whether two short statements about the same thing can both be true. ## Contemporaneous material carries more weight A witness statement is written for the proceedings, by someone who knows what the case is about, often years after the events. An invoice, a calendar entry or an internal email was written by someone with no reason to think about any of this. When those two disagree, the disagreement is worth something. When two statements disagree, it is usually memory. So rank your search: statement against contemporaneous document first, document against document second, statement against statement last. ## What a usable contradiction record looks like Both sides, in full, with their sources. Not a note saying the accounts differ. Six months later that note is unusable, because rebuilding what it referred to takes as long as finding it did the first time. Record: - the two claims, quoted; - the document each came from, with its exhibit reference if it has one; - the date of each document, because which came first often decides what the conflict means; - one line on why they cannot both be true. That last line is the one people skip. Write it. If you cannot, you may be looking at two statements that are merely different rather than inconsistent. ## Watch for the false positive Most apparent contradictions dissolve on inspection. Two documents use the same word for different things. A date refers to when something was agreed rather than when it happened. A figure is net in one place and gross in another. This is why the output of any automated pass is a queue for review rather than a list of findings. The machine can put every candidate pair in front of you with both sources attached. Deciding which ones are real is legal judgement and stays that way. ## Doing this at the scale of a real file Comparing every claim against every other claim by hand is not possible past a few hundred documents, which is why it usually does not happen and why inconsistencies surface during cross-examination instead of during preparation. Legalnaut runs that comparison and surfaces the surviving pairs with both documents, both dates and both checksums attached. Contradictions are treated as findings rather than errors, which is also why they are marked in the brass accent rather than in red. A finding is something you use. --- ## Check these 5 things before AI text enters a filing URL: https://legalnaut.com/blog/check-these-things-before-ai-text-enters-a-filing Language models produce fluent, confident, well-structured text that is sometimes about a world that does not exist. In most work that is an annoyance. In a filing it is a professional problem, and it has already ended badly for lawyers in several jurisdictions. ## The risk is specific, so the check can be too The failure mode is not that a model is bad at law. It is that a model has no way to distinguish between recalling something and constructing something that fits the shape of the answer. Both come out in the same confident register. So the check is not "is this good". It is "can each factual claim here be traced to a document I can open". Five passes, in order, and they get faster with practice. ## 1. Every claim names a source Take each sentence that asserts a fact and ask what it points at. A passage that cites nothing is not a draft, it is a hypothesis. Send it back or delete it. This is why it matters that the tool you use attaches sources as it works rather than offering to find them afterwards. Retrofitted citations get attached to whatever is closest, which is a different thing from where the claim came from. ## 2. Open the source, do not trust the label A reference that looks right is not the same as a reference that is right. Open the document. Confirm it exists, confirm it says what the passage says it says, and confirm it is the document you think it is rather than a similar one. For file material a checksum makes this cheap: the same document, verifiably unaltered since it entered the file. For cited authorities there is no shortcut. Look it up. ## 3. Quotations are verbatim, or they are not quotations Check quoted text character by character against the original. Models paraphrase while keeping quotation marks, which is the most dangerous error on this list because it survives casual review. A related trap in multilingual files: a quotation translated into the language of the filing has stopped being a quotation. Quote the original and translate alongside it. ## 4. Dates and numbers, individually Dates drift. Figures get rounded, transposed, or attached to the wrong party. These are exactly the errors an opponent enjoys finding, and they are quick to check against the source you already have open from step two. ## 5. Cut what nobody asked for Models fill space. Ask for a summary of three documents and you may get a fourth point that reads well and rests on nothing. Anything that does not trace back to a document in step one comes out, however good it sounds. ## Build the check into the tool, not into your discipline A process that depends on remembering to be careful at eleven at night is a process that fails at eleven at night. Legalnaut attaches the source to every generated statement and marks AI drafts as drafts requiring review, so an unsupported claim has nowhere to hide and nothing leaves the system labelled as finished work. The checking still belongs to a lawyer. What changes is that the material needed to check is already attached. --- ## Why your served bundle is invisible to search URL: https://legalnaut.com/blog/why-your-served-bundle-is-invisible-to-search You search the file for a word you are certain appears in it, and get nothing back. The document is there. The word is on the page. What is missing is any text for the search to match, because the page is a photograph of a page. ## A scan is a picture, not a document A PDF can hold two very different things. One is text with layout information around it. The other is an image of a piece of paper, wrapped in a PDF container so it can be emailed. Both open in the same viewer and look identical to a reader. Only one can be searched. Court files are full of the second kind: material photocopied, faxed, scanned by an office machine set to whatever it was set to, and served as a single large file. ## The failure is silent, which is the problem Nothing warns you. Search returns no results, which looks exactly like search returning no results because the word is not there. The document sits in the file, indexed by filename only, and whoever is preparing the case works around a hole they do not know exists. This is the part worth internalising: an unsearchable document in a large file is functionally a missing document, and it goes missing without any error message. ## What OCR actually gives you Optical character recognition reads the image and produces text. On a clean scan of a typed page the result is close to perfect. Quality falls off with generation loss, so a fax of a photocopy of a fax produces fragments, and handwriting in the margin usually produces nothing. Two practical consequences follow. First, run it on import rather than on demand. OCR at the point a document enters the file means the file is searchable from the start. OCR when someone notices a gap means the gap has already cost something. Second, keep the original. The OCR text is for finding the document. The document is what gets quoted, exhibited and put in front of the court. ## Knowing what failed is half the value A page of ligature soup is not text, and treating it as text is worse than treating it as nothing. It pollutes search results and it will eventually be quoted by something that cannot tell the difference. What you want from the process is a clear answer in two parts: here is the text we could recover, and here is the list of documents where recovery failed. That second list is a work queue. Those are the pages a person has to open, and there are usually far fewer of them than the size of the bundle suggests. ## Where this sits in the workflow In Legalnaut, extraction runs by content rather than by file extension, because served material arrives named badly or not named at all. A local parser tries first. When what comes back is too short or too broken to be real text, the document goes to OCR, and audio goes to transcription with timestamps kept so a quotation can be checked against the recording. Documents where nothing usable came back are listed as such rather than left to look like empty documents. That list is short and it is worth an hour. --- ## How to build a case chronology from 900 emails URL: https://legalnaut.com/blog/build-a-case-chronology-from-900-emails A commercial dispute is usually an argument about what was agreed and when. The answer sits in correspondence written years ago by people who have since left both companies, and the side that can lay out the sequence cleanly is the side that looks credible. ## Start from the documents, not from memory The chronology most teams build first is a list of what everyone remembers, checked against documents afterwards. That order is backwards and it shows up under pressure. An entry you cannot trace to a document is an entry you cannot defend when the other side asks where it came from. Work the other way. Take the documents you have, pull the dated assertions out of them, and let the sequence assemble itself. Gaps in the resulting timeline are findings in their own right: a month where nothing was written down is often the month worth asking about. ## What belongs in one entry Four things, and nothing else: - **The date.** From the document, not from context. If the document is undated, the entry says so rather than guessing. - **What happened**, in one sentence, in neutral language. - **The source**: which document, and where in it. - **Who says so**: the author of the document, because a claim by your own client and a claim by the other side carry different weight in the same list. Anything else belongs in a note attached to the entry, not in the entry. A chronology that carries argument stops being usable as a reference and turns into a draft submission. ## Three passes, in this order **Extract.** Every dated assertion in every document, including the ones that look irrelevant. This is the pass to be generous on. Filtering during extraction means deciding what matters before you know what the sequence looks like. **Deduplicate.** The same event will appear in twelve documents. Collapse them into one entry with twelve sources rather than twelve entries. This is where a hand-built chronology usually breaks, because merging by eye across hundreds of rows is not something a person does reliably at four in the afternoon. **Rank.** Now filter. Which entries carry the dispute, which are context, which can be dropped. Do this last, in one sitting, with the whole sequence in front of you. ## Where chronologies go stale Disclosure arrives in tranches. A chronology finished in March is wrong by May, and the version circulating in counsel's inbox is a copy of a copy with two people's edits in it. The fix is structural rather than disciplinary. Keep one chronology, keep it attached to the document set it was built from, and regenerate rather than patch. If rebuilding is expensive, people will patch, and the patched version is the one that goes to the hearing. ## When two documents disagree They will. An invoice dated March describes work a witness statement puts in June. Do not resolve it in the chronology. Record both entries, mark the pair, and deal with it as a contradiction rather than as a date problem. A chronology that quietly picks the more convenient of two dates is worse than one that shows both, because the inconsistency is still in the file and the other side will find it. ## What this looks like when it is automated Legalnaut does the extract and deduplicate passes on import. Dates, parties and events come out of the text with a pointer back to the sentence they came from, near-identical entries are merged, and the ranking pass stays where it belongs, with the lawyer. New disclosure folds into the same timeline instead of starting a new document. If two entries conflict, the pair is flagged rather than reconciled.