D

DocJuris

Remove the models and there is nothing this vendor sells, which is the A band. The catalogue is organised as roughly two dozen applications whose unit of delivery is a finished AI work product: a tracked-changes redline, a deviation memo on the customer's template, a restated contract, an Excel lease abstract, a completed security questionnaire, a citation-backed research brief. Pricing follows the same logic and is stated to be sized to contract volume and the applications enabled rather than to headcount, so the buyer is charged for the work the models complete.

The custom route is the clearest tell: the vendor scopes a described workflow and ships a new application in weeks, meaning the product line itself is generated by model capability. The counter-argument is named rather than hidden: the catalogue does contain an operations layer, intake forms, matter boards, approvals, a repository and a counterparty deal room, and a workspace of that kind would still open without models, which is the B band's shape.

It does not carry the record, because each of those is described in the vendor's own words as model-run, the repository being searchable by meaning with obligations mined and invoices carrying verdicts computed rather than guessed, and because the vendor positions expressly against the workflow product, describing what it returns as a finished redline rather than a workflow ticket. Verified 12 September 2026.

Accuracy is asserted without measurement and grounding is claimed without a described method, which is the C band exactly. The limbs bite more here than on most contract products because the catalogue includes a Legal Research application that produces citation-backed briefs from primary law across eight jurisdictions, so authority is being generated and not merely clauses. What is published: grounding claims in the vendor's own idiom, that markups derive from the customer's own playbook and precedent, that generated documents put deal terms onto the customer's form with nothing silently invented, and that security questionnaires are answered from the customer's own policies with nothing invented.

What is absent is everything A and B ask for: no accuracy figure, no test set, no error rate, no described retrieval method, no statement of what the system does when it can find no support, and no way for a reader to open and verify the primary authority behind a research brief. R40 governs the floor and the D limb does not fire, because the accuracy content is not a bare no-hallucination claim: Subscription Terms section 5.03(b) states that given the probabilistic nature of machine learning the Services may produce incorrect Output that does not accurately reflect real people, places, facts or laws, and directs the Client to evaluate accuracy including by using human review.

A contractual limitation is more than most C records carry and it is recorded rather than credited to the band. Verified 12 September 2026.

A written commitment that the models work alongside a supervising lawyer, with real review surfaces, short of the threshold at which the system acts alone. The modes are published and named: Quick Pass where speed matters, and Guided Review where the user confirms each position. The review surface is the artifact itself, since every proposed edit arrives as a tracked change in Word that a lawyer accepts or rejects, with internal comments held back from the counterparty, and the vendor describes the product as recommending focused changes via a checklist with a human in the loop while the user controls strategy.

The route back to human judgement is contractual as well as descriptive: Subscription Terms section 5.03(b) places evaluation of Output on the Client and names human review as the means. R37 rule 2 governs what holds it off A. The same estate markets an operations layer that runs itself, invoice verdicts computed rather than guessed, and agents that solve any workflow, and those claims cannot all be true alongside confirm-each-position.

The conflict is not itself a downgrade; what it reveals is that no threshold, confidence boundary or class of work is published at which the system proceeds without a reviewer, and that is the limb the B band names as commonly absent. The note carries both claims so a reader weighs them. Verified 12 September 2026.

Real deployment evidence with unusual substance, held at B on the limbs that cannot be established rather than on thinness. Eight customers are named with a figure attached to each: Flex at contract review falling from eight days to five minutes, described as 99 per cent less review time; Siemens at ten times more consistent review; GEODIS at 75 per cent faster approvals from remote teams; Ancestry at twice faster execution across procurement workflows; Purolator at over $300,000 saved and 50 per cent lower legal-technology spend; Careington at five times more consistent review against standard terms; CMA CGM America at 90 per cent faster turnaround; ECS Limited at twice the review speed including on locked and scanned files.

Three carry a named executive quoted on the record: the General Counsel of Siemens APAC, the Deputy General Counsel of CMA CGM America and the Head of Legal Operations at Purolator. That is fuller attribution than the B band's examples, which describe a named customer without figures or figures without the name. Two A limbs are not established: no date appears against any of the eight on the surfaces read, and no method is published for any figure, so a reader cannot assess how 99 per cent or ten times was computed.

The full case-study library is published and was not opened; under R25 it is not load-bearing for B, which stands on the strip itself, and it is the artifact that would move this row to A if the studies carry dates and a stated basis. Verified 12 September 2026.

Substantive published commitments on confidentiality, readable before signing, short of most of the rest of the A list. The real commitment is Subscription Terms section 4.01(e), which is more specific than this corpus usually sees: DocJuris will hold all Client documents and playbooks confidential using precautions as stringent as it applies to its own confidential information, will not disclose them in a way that permits identification of the Client for the term and a further five years, and where a government authority compels disclosure will notify the Client where permissible and cooperate to obtain a protective order.

Tenant-level separation, which is the segregation limb for an in-house buyer, is documented and architectural: an isolated single-tenant deployment on the customer's own subdomain with its own service stack, role-based permission groups scoped to individual repositories and applications, and audit logging of logins and permission changes. Three A limbs fail. Privilege and work product are addressed nowhere, and R33 makes that limb decisive on its own.

There is no contractual position on training: the marketing states the models are never trained on customer data while section 4.02(f) grants a perpetual, irrevocable, sub-licensable licence over Client Data and Output to improve or enhance the Services, and the agreement governs the reading. And no position of any kind is published on what a third-party model provider may retain, which matters more here than usual because no provider is named. Retention and deletion are absent from the agreement entirely. Verified 12 September 2026.

A real published position on advice versus tooling, stated more precisely than the ordinary disclaimer, short of the supervision and competence dimension. The same sentence appears in the footer of every page read: DocJuris is not a law firm or a substitute for an attorney or law firm, and cannot provide any kind of advice, explanation, opinion, or recommendation about possible legal rights, remedies, defenses, options, selection of forms or strategies.

That is specific about the acts disclaimed rather than a bare not-legal-advice line, it sits on the marketing surfaces rather than buried in the terms, and Subscription Terms section 5.03(b) supports it by directing human review of Output. C does not fire: the audience is named rather than ambiguous and the marketing does not describe the product in advice terms. What holds it off A is the dimension this product raises most sharply and does not answer.

The platform is sold deliberately to non-lawyers, with procurement and sales operations working in the same tenant and marketed as safe self-serve, generating redlines and negotiating positions on legal documents, and nothing published addresses how a supervising lawyer discharges competence and oversight duties over that work, who may use which application, or any jurisdictional limit. A separate Website Disclaimer page is published in the footer and was not opened; under R25 it corroborates the sentence already read in full rather than carrying the grade. Verified 12 September 2026.

No governance mechanism, testing regime or accountable owner is published, which lands on C rather than lower only because a position of some kind exists. What is located: a privacy-first posture asserted site-wide, a statement that customer data never trains AI models, SOC 2 Type II certification, and two contractual provisions that function as governance even though they are not labelled as such. Section 4.03(c) publishes a scope boundary, stating that the Services are not designed with security and access management for medical records, US Munitions List data, defense articles or ITAR-related data, and barring the Client from putting any of it through the product; a vendor naming work its system should not be used for is a real disclosure and is rarer in this corpus than a principles page.

Section 5.03(b) acknowledges probabilistic output on the record. Everything the A and B bands ask for is absent. There is no responsible AI or AI governance page, no framework and no certification against one such as ISO 42001, nobody inside the vendor is named as accountable for AI, nothing is published about what is evaluated before an application ships, and nothing addresses uneven output across contract types, counterparty language or drafting conventions, which is the bias question for a product that applies one company's playbook to everybody else's paper. The SOC 2 attestation is security governance and is not credited here. Verified 12 September 2026.

Access control is documented and real; what happens to documents and prompts after processing is not addressed anywhere, which is the C band. Published and specific on access: single-tenant isolation with the customer's own data boundary, single sign-on through the customer's identity provider over OpenID Connect with invite-only access when it is off, role-based permission groups scoped down to individual repositories and applications, and audit logging of logins and permission changes, alongside usage metering visible to customer administrators.

Against that, four of the five limbs are simply unanswered. The Subscription Terms were read in full and contain no retention period, no deletion or return of Client Data on termination, no incident notification commitment and no security programme running to the customer; section 4.03(a) and (b) allocate security responsibility to the Client for its own systems, credentials and uploads, and section 3.02(g) runs the only audit right in the agreement in the vendor's favour rather than the customer's. No subprocessor list and no data processing addendum exist: the footer is the complete legal inventory and carries neither, which is a page-inventory finding under R20 rather than a failed search.

One genuine reader limit is recorded and does not carry this grade: the privacy policy is a stub linking to an iubenda-hosted document that returned a robots refusal to this index's fetcher on 12 September 2026, published and not gated, and a website privacy policy is in any case not where customer-document retention would live. The grade rests on the agreement, which was read. Verified 12 September 2026.

Liability is addressed only through a limitation clause that disclaims the exposure this product creates, which is the C band, and the drafting is more specific than most records at this grade, so the grade turns on scope rather than on clarity. What is published and precise: section 5.01(b) caps each side's aggregate liability for direct damages at the greater of fees paid in the preceding year or USD 50,000, section 5.01(a) excludes consequential, incidental, indirect, special and punitive damages including loss of data and breach of data or system security, and section 5.01(c) shortens the limitation period to two years, with the cap disapplied only for fraud or wilful misconduct.

Section 5.02 is a genuine indemnity with named exclusions, but it runs to third-party intellectual property infringement in the IP or Services and nothing else, and its remedy ladder ends in a refund of the unearned portion of the fee. The exposure a buyer actually carries here is a bad clause accepted, a missed obligation, a redline that concedes a position, or an invented authority in a research brief, and the agreement allocates all of it to the Client: section 5.03(a) supplies the Services as is and as available and disclaims every warranty including accuracy and fitness for purpose, and section 5.03(b) makes evaluation of Output the Client's responsibility.

No indemnity, no insurance position, no service level and no warranty of the work product were located. Verified 12 September 2026.

Real integrations, named and documented, with genuine depth on the surface that matters most, short of the documentation an implementer could work from unaided. The integrations page publishes more than fifty connectors across CRM, ERP and storage categories and names the tools individually: Microsoft Word, SharePoint, Salesforce, Excel, Google Drive, OneDrive, Dropbox, Box, Asana, Trello, Jira, monday.com, ClickUp, Typeform, Microsoft Forms, Google Forms, plus Zapier and Power Automate as routes to several thousand more.

The depth is described rather than asserted on the document path, which is the one legal work actually lives on: contracts go in and come back as Word files with tracked changes, comments and formatting intact and no conversion step, and the vendor publishes a separate account of embedding intake through Microsoft Forms, redlining inside Word and version control through SharePoint. An API exists and its semantics are published at the event level, with triggers named for turn added, turn completed and internal comment so a customer's systems react when review work happens, and a documented round trip that pushes a PDF or Word file from a contract record into DocJuris and returns the redline to the originating system, with OAuth 2.0 authentication and agent access controls.

What holds it off A: the API documentation is not public and is obtained by emailing support, the connector inventory is presented as a sample rather than a register, and what syncs in which direction is described for the Word path but not for the CLM and ERP connections. A support portal at support.docjuris.com carries integration documentation and was read only in part through the search index. Verified 12 September 2026.

The tenancy model is stated clearly on its own with no residency detail published, which the B band now names expressly. Tenancy is the strongest thing on this record and it is stated as architecture rather than as policy: an isolated single-tenant deployment on the customer's own subdomain in the form company.docjuris.ai, with its own service stack, described in the vendor's words as never a shared multi-tenant black box and as isolated from other customers by architecture rather than by policy alone.

The deployment path is set out in three published steps covering the dedicated tenant, identity provider integration, and administration through audit logs and usage meters. Residency is absent altogether. No region is named anywhere on any surface read, no menu of regions is offered, no single-region or in-country option is described, no cloud provider is identified, and nothing distinguishes where processing happens from where data is stored, which matters here because the customer's documents are the input and because the model providers behind the applications are also unnamed.

The only geographic datum located anywhere is a place-of-processing line for website visitors in the search-indexed fragments of the hosted privacy policy, which concerns tracker data rather than customer contracts and is not credited. Verified 12 September 2026.

Certification is real and stated, and the evidence behind it is not reachable without a sales conversation, which is the B band. SOC 2 Type II is named on at least four surfaces including the dedicated Privacy and Security page and the site-wide banner, and it is stated plainly in answer to a direct question rather than displayed as an unexplained badge, so C does not fire. Around it sit administrative controls documented control by control: OpenID Connect single sign-on, role-based permission groups scoped to repositories and applications, audit logging of logins and permission changes, invite-only access where single sign-on is off, and usage metering visible to administrators.

R5 governs the access tier and it is the sales-gated one, earning no credit: the published route to the evidence is to book a demo and bring the security team, on which the vendor will walk the architecture and share the documentation the review needs. No request was submitted. What is missing for A is the whole evidentiary apparatus: no auditor is named, no report period or observation window is given, no certificate or report number is published, there is no trust portal of any kind, no second standard such as ISO 27001 is claimed, and no penetration testing, vulnerability remediation target or incident response function is described anywhere. Verified 12 September 2026.

The vendor refers to advanced models without identifying what sits underneath, which is the C band in its plainest form. The disclosure stops at a category and the category is a marketing term: the pricing FAQ states that the vendor runs the best frontier model per task, the catalogue describes agents as frontier models solving any workflow, and the platform is described as proprietary AI. No model is named, no provider is named, no architecture is described beyond the phrase, and no commitment to notify customers when any of it changes was located, so none of A's four limbs is reached and B is not either, because a provider is never named.

One adjacent fact is recorded rather than credited, because it answers a different question: the same FAQ offers to run inside the customer's own approved AI deployment, and names Claude and Copilot only as tools a customer may already own rather than as anything DocJuris uses, so it evidences deployment flexibility and not supply chain disclosure. There is no subprocessor list anywhere on the estate and no data processing addendum, so a customer cannot learn the providers by any published route, on request or otherwise.

That is the sharper half of this row: on most records at C the identity is obtainable by asking, and here nothing indicates it is obtainable at all. Verified 12 September 2026.

Pricing is gated behind a demo while the shape is published in detail, which is the C band and the Bloomberg Law precedent at R17. What is published is genuinely more than most gated pages carry: two named ways to buy, an annual platform subscription covering the full catalogue with applications, agents and enterprise integrations bundled, or a professional-services route of a one-time build fee followed by hosting and usage credits with change orders; the basis on which the subscription is sized, being contract volume and the applications enabled rather than headcount; an express statement that nothing is charged per seat, per document or per question; two comparison tables of feature splits; a stated reason for withholding the number; and a launch discount of 20 per cent on a first scope of work engaged within thirty days of a demo.

No figure of any kind is published for DocJuris. The dollar amounts that appear on the pricing page describe competitors rather than this vendor and are not the vendor's own pricing. What the record must carry, under R37 rule 1, is that the pricing page and the agreement disagree about the unit of charge, and the agreement governs: the page says unlimited users and no per-seat fees, while Subscription Terms section 3.02(a) sells a subscribed number of Authorized Users stated in the Order and charges any excess at the per-Authorized User rate.

Commercial terms otherwise sit in the agreement, including fees payable in advance and irrevocable and non-refundable, ninety days' notice to terminate for convenience, and amendment by public posting with silence for thirty days deemed acceptance. Verified 12 September 2026.

Segment and practice coverage is described with real substance and the boundaries are mostly left open, which is B. Who it is for is stated precisely and repeatedly: in-house legal departments first, with procurement and sales operations named as co-users inside the same tenant, addressed on dedicated pages of their own and on an all-teams page, and sized explicitly from a department of twenty to forty lawyers up to a global enterprise.

The exclusion is positional rather than stated as a limit but it is consistent, the vendor defining itself against law-firm economics and describing its design as built around in-house workflows rather than law-firm billing models. Practice coverage is enumerated rather than gestured at, across five named groups, contracts and commercial, legal operations, compliance and regulatory, real estate and facilities, and finance and treasury, with roughly two dozen applications named individually beneath them and industry pages for healthcare contracts, SaaS providers and vendor contracts.

One real limit is published, and it is contractual: section 4.03(c) states the Services are not designed for medical records, US Munitions List data, defense articles or ITAR-related data. What is left open holds it off A. The eight jurisdictions covered by the legal research application are counted and never named, no practice area is identified as unsupported, nothing states which applications a non-lawyer may operate, and the catalogue count itself is inconsistent across the site, given as more than twenty-five in the navigation and as twenty-two on two in-page links. Verified 12 September 2026.