P

Precedent

The models are the engine of a core capability, layered on a service platform that would still function without them. What would survive their removal is substantial and separately saleable: Claim Setup submits claims to carriers and returns verified policy limits, an automatically generated letter of representation notifies the insurer, Record Retrieval requests and tracks medical records, and Policy Verify establishes coverage.

Those are operational services with a human and workflow core, and the vendor presents them as the first three of four steps before AI enters. What the models power is the fourth step and the flagship: Demand Composer drafts the demand package from the case documents, MedChron builds medical chronologies, Global Demand generates multi-plaintiff packages, Exhibit Manager categorises, describes and renames incoming files automatically, Case Intelligence analyses and benchmarks demands, Adjuster Intelligence predicts adjuster tactics, and Ask Casey is a voice assistant.

Medical billing is broken to CPT-code level. A is unavailable because a buyer could purchase the claim setup, policy verification and records side and receive a working service, and because the vendor itself describes trained expert reviewers as part of the production path rather than the models alone. Checked 8 September 2026.

Grounding is real, documented and verifiable by the reader, short of A on the limbs asking for measurement and named failure modes. The architecture is described rather than asserted: what the vendor calls a Closed Loop means every medical fact, diagnosis and billing code in a demand letter is hyperlinked directly to the source page in the raw medical record, and the vendor puts the point plainly, telling a buyer they do not have to trust the AI because they can click the citation and see the source document.

That is linked primary sources in the idiom that matters here, the primary source being the medical record the demand rests on. A second control is published: trained expert reviewers check every draft for narrative logic and factual integrity before the firm receives it. A tension on the same page is recorded rather than resolved in the vendor's favour, because it is the most informative thing on this axis. The section is headed Zero Hallucinations, and a system with zero hallucinations would not need a review team checking factual integrity before delivery.

Under R37 rule 2 the conflict is not itself the grade; what it shows is that the accuracy assurance rests on source-linking and human review rather than on model behaviour, which is a sound architecture and is not what the headline claims. R40 governs the floor: the bare-claim limb does not fire where real architectural controls stand behind the claim. A is unavailable because no measured accuracy figure, test set or method is published anywhere and no failure mode is named. Checked 8 September 2026.

A written commitment that the models work alongside human review, with real review surfaces, short of the full control structure. Two review stages are published and both are specific. The vendor's own trained expert reviewers check the output for narrative logic and factual integrity before the firm ever sees the draft, which is an oversight step inside the vendor rather than an obligation pushed onto the buyer, and it is unusual in this corpus.

The attorney then receives a draft the vendor describes as ready for review, and the verification surface is concrete: any medical fact can be clicked through to its source page in the record, so checking the output is a designed workflow rather than an instruction. Role-based access adds a further control over who inside the firm touches what. What is missing is the rest of the structure. No mode is described, no threshold is stated at which the system stops or escalates, and nothing published says what happens when the output is wrong or what recourse follows.

The Closed Loop is named as an architecture rather than described as a control with limits. A requires modes, thresholds, review surfaces and the route back to human judgement all published; two of the four are here. Checked 8 September 2026.

Real deployment evidence with substance, held off A by what could not be verified rather than by what is missing. Published: a named case study with Hines Law Firm reporting settlements up 16 per cent and the likelihood of tenders up 71 per cent, stated to be measured across hundreds of personal injury cases; named customer logos including Catania, Rodden, Lerner and Rowe, Dubin Law, Ramos Law and Sand Law; and attributed testimonials carrying both a person and a role, among them a Personal Injury Director at Hines Law describing drafting thirteen demands in a day, a firm COO, and a partner.

Aggregate figures are published alongside: more than 100 law firms, more than 50,000 demands sent, a 95 per cent retention rate and a 4.7 star rating. A named customer with figures attached would ordinarily reach for A. Two things hold it at B. The case study page was not opened, so the method behind the 16 and 71 per cent figures, the comparison basis and the period are unestablished, and A requires a method a reader can assess.

And the vendor's own surfaces disagree on the headline number: the home page banner claims 25 per cent higher settlements while the case study it links to reports 16 per cent, with nothing reconciling them. That discrepancy is named here rather than left for a reader to find, and the aggregate claims carry no stated basis. Checked 8 September 2026.

Substantive published commitments across most of the ground, held at B by the limb R33 makes mandatory. Published and readable before signing: each firm's data is logically isolated with case files described as held in their own private vault; role-based access assigns permissions to partners, paralegals and case managers so staff reach only relevant files; encryption is AES-256 at rest and TLS 1.2 or better in transit with perfect forward secrecy; all processing and storage occur within the United States; the firm owns its data, controls retention through provided tooling, and can have raw data permanently purged within thirty days of a request; and client case files are stated not to be used to train models or shared across clients.

The Master Service Agreement puts Protected Health Information inside the definition of Confidential Information, binds both parties to strict confidence surviving termination, requires subcontractors to be bound, and commits to HIPAA-compliant handling with a business associate agreement available. For a product ingesting a claimant's complete medical file that combination is materially stronger than most of this lane.

A is unavailable on two limbs. Privilege and work product are addressed nowhere, expressly or by implication, which under R33 forecloses A on its own. And no position is stated on what an underlying model provider may retain, because no model provider is named anywhere. Checked 8 September 2026.

This record sits in a documented gap between the C and D bands, and the grade is the band nearest the buyer's position with the gap stated rather than hidden. Neither C limb is literally true. There is no boilerplate advice disclaimer in the terms, the Master Service Agreement containing no statement that the service is not legal advice; the marketing does not describe the product in advice terms, presenting it as a drafting and workflow service; and the intended audience is not ambiguous, the product being sold exclusively to plaintiff personal injury law firms.

D's words are literally true, since nothing published states the advice line. But D describes a vendor that has said nothing about professional responsibility, and that is not this record. The workflow is published as routing every draft through the vendor's expert reviewers and then to the attorney as a draft ready for review, so review by the lawyer is designed in and stated. And the vendor publishes a substantive article on the professional conduct dimensions of using its own services, engaging ABA Model Rules 1.5 and 1.8, mapping requirements across six named states, and telling firms their technology use must not compromise duties of loyalty and communication.

What is absent is the specific statement this axis asks for: what the product is and is not, who may rely on it, and how it supports competence and supervision. Jurisdiction limits are not named either. A proposed band amendment is logged rather than escalated. Checked 8 September 2026.

No governance position is published for a system that drafts the document a claimant's recovery is negotiated from. Nothing identifies who inside the vendor is accountable for the models, no pre-release testing or evaluation regime is described, no results are disclosed, no policy or principles statement exists, no certification such as ISO 42001 is claimed, and nothing addresses uneven output across injury types, treatment patterns, jurisdictions or claimant demographics, which is where bias would bite hardest on a product that values a claim.

The security page does carry a section headed AI Safety, and it is real content rather than a badge, which is why the C band was tested carefully. C describes responsible AI principles published without a mechanism; this vendor publishes the inverse, a mechanism without principles, so C's words are false of it. The mechanism itself, the Closed Loop architecture with hyperlinked citations and expert reviewers, is graded on Citation Accuracy and on Autonomy where it answers those bands directly, and under the no-double-spend rule it cannot also stand as a governance position.

Product quality control and AI governance are different questions: one is about whether this output is right, the other about who is answerable for how the system is built, tested and corrected. Only the first is published. Checked 8 September 2026.

Substantive published policy covering most of the ground, short of the full set in the two ways the B band names. Published and specific: retention is customer-controlled rather than a fixed vendor window, with tooling provided so a firm can apply its own retention policy and permanent deletion options available; offboarding is stated concretely, with all generated work product downloadable and raw data permanently purged within thirty days of a request; access control is described at the level a firm operates, with logical tenant isolation and role-based permissions for partners, paralegals and case managers; encryption is AES-256 at rest and TLS 1.2 or better in transit with perfect forward secrecy; all processing and storage are within the United States; and the training position is stated plainly, with case files used only to produce that case's demand package and never shared across clients or used for model improvement.

HIPAA-compliant architecture and an available business associate agreement matter here because the product ingests complete medical files. Two elements are absent and they are exactly the B band's named shortfalls. No subprocessor is named anywhere, so a firm cannot enumerate who touches a claimant's records. And no incident or breach notification practice is published on any surface, which is a conspicuous gap for a HIPAA-facing platform whose agreement is otherwise detailed. Checked 8 September 2026.

A real published position on liability, short of the full picture. The Master Service Agreement sets out mutual indemnification, each party defending the other against third-party claims arising from insufficient consents, introduction of malicious code, negligence or misconduct, material breach, intellectual property usage or legal violations, with prompt notice and cooperation required. Aggregate liability is capped at amounts paid in the preceding twelve months, with infringement claims excepted, and neither party bears indirect, consequential or punitive damages.

Precedent warrants its authority to perform and non-infringement of registered US copyrights, patents and trademarks. That is scope and caps both published, which is more than most of this lane. What is not available is anything a buyer can invoke when the output itself is wrong. No warranty addresses the accuracy or completeness of a demand package, clause 6.3 disclaiming all warranties beyond the express limited one; no insurance is referenced; and a clause recovered from the agreement provides that Precedent shall have no liability for failure to identify or notify the client of any demand, which matters for a service that delivers demand letters to carriers on the firm's behalf and where a missed communication could carry consequences the firm rather than the vendor bears.

One limit is recorded and did not drive the grade: the agreement as published is an abridged summary rather than operative text, and fuller clause language exists in the search index than the page returns, so the full limitation wording was not read. Checked 8 September 2026.

Real integrations exist, into precisely the systems this buyer already runs, and are documented short of depth. Three are named and each has its own dedicated page: SmartAdvocate, where the vendor states Demand Composer runs directly inside the platform; Clio, where demand creation is described as streamlined within Clio Manage; and Litify, where demands are generated and managed in the firm's matter system. For plaintiff personal injury those three are close to the whole market of case management systems, so this is coverage of the systems legal work actually lives in rather than a list of adjacent tools.

Third-party recognition supports it: the vendor was a finalist for Best New App in the 2025 Clio Integration Awards, which is an assessment of the integration specifically rather than of the product generally. A dedicated integrations index page exists. A is unavailable on the depth limb and the reason is stated rather than inferred: the three integration pages were not opened, so what actually syncs, in which direction, and what a firm must configure are unestablished, and the navigation blurbs describe the function without describing the data flow.

No API or developer documentation was located, and no document management or e-signature connection is named. Checked 8 September 2026.

The deployment model is stated clearly with partial residency detail, which is the B band. Two limbs are answered better than most records in this lane manage. Processing is addressed distinctly from storage rather than collapsed into a single hosting statement: the vendor states that all data processing and storage occur exclusively within the United States and that client data never leaves US soil, framing it as a data sovereignty commitment.

And tenancy is stated rather than left to inference, each firm's data described as logically isolated with case files held in their own private vault under strict access controls, supported by role-based permissions. Where it stops short is granularity and choice. Residency is published at country level only: no cloud provider is named, no region or data centre is identified, and no failover or secondary location is described, so a buyer knows the country and nothing below it.

No deployment options exist to compare, there being a single cloud service with no private, dedicated or on-premise variant described, so the tier limb does not bite and is named rather than counted either way under R15. And residency is disclosed as the vendor's architecture rather than offered as a customer election, with no mechanism for a firm to require or verify placement. Checked 8 September 2026.

Certification is real and stated, short of accessible evidence, which is the B band. The vendor states it is SOC 2 Type II certified and describes the basis with more specificity than a badge: rigorous annual audits by independent third parties verifying internal controls over security, availability and confidentiality. It separately claims HIPAA-compliant architecture and, unlike a bare compliance claim, backs it with a checkable commitment, offering to sign a business associate agreement for firms handling Protected Health Information.

The security page itself is substantive and self-published rather than a hosted portal, covering encryption, tenant isolation, role-based access, retention control, US residency and the training position. What is missing is everything that would let a buyer verify the certification. No auditor is named, no audit period or report date is given, and the scope of the assessment is not described. Most notably, no route to the report is published at all: unlike records where access is sales-gated and therefore earns no credit under R5, here there is neither a download nor a stated request path nor a mention of a security review process, so the report's existence is asserted and its availability is not addressed.

No other attestation is claimed and no penetration testing or vulnerability programme is described. A requires reports, dates and covered standards reachable without a sales call, and none of those is present. Checked 8 September 2026.

The vendor refers to the models generically and identifies nothing underneath them, which is the C band. The one place the supply chain surfaces is the training answer on the security page, which states that the vendor does not train its public foundational models on private client data. That phrasing concedes that foundational models are in the architecture, and it is the only acknowledgement located, but it names no model, no provider and no version, and nothing states whether the generative layer is built in-house or supplied by a third party.

The Closed Loop is described as an architecture without identifying what generates inside it. The Master Service Agreement is silent on the point across its length, containing no AI, model or machine learning provision of any kind. No hosting provider is named either, so even the infrastructure layer is undisclosed, which is unusual: most records in this corpus at least name a cloud host, and naming one would not have reached this axis in any event, since under R29 test 1 a host says where a model runs rather than whose it is.

D was tested and does not fit, because the reference to public foundational models does disclose something about the supply chain a customer inherits. Change notification is not reached, nothing being named that a change could be notified about. Checked 8 September 2026.

Real pricing is published for part of the range, which is the B band's first limb. The flagship carries a full published rate: Demand Composer is a flat 275 US dollars per demand with unlimited pages and unlimited revisions, presented on the home page as a maximum price rather than a starting point. Surrounding terms are published too: a thirty-day risk-free trial, fixed pricing with no fees or add-ons, payment only for what is used with no prepaid tokens, and an express statement that firms of all sizes are served.

For a buyer wanting one demand drafted, the cost is knowable without a sales conversation, which is more than most of this lane offers. What holds it off A is coverage and one contradiction. Six other named products carry no published rate at all, including Claim Setup, Policy Verify, Record Retrieval, MedChron, Case Intelligence and Adjuster Intelligence, so a firm buying the platform rather than a single demand cannot price it.

And the marketing and the agreement diverge on the commercial commitment: the home page states there are no contracts and the security page that there are no long-term contracts, while clause 9.2 of the Master Service Agreement provides for automatic renewal for a further twelve months with a 7 per cent fee increase unless cancelled in writing thirty days before the service end date. Under R37 rule 1 the agreement governs. Both are stated so a reader can weigh them. Checked 8 September 2026.

Segment and practice coverage is described with substance, with the boundaries left open. The segment is stated without hedging and the product is built around it: plaintiff personal injury law firms, positioned explicitly against the carrier side. Firm size is addressed directly rather than assumed, the vendor stating it serves all firm sizes and handles small and large cases alike, which matters in a lane where products often silently target either high-volume intake shops or catastrophic-injury boutiques.

The workflow coverage is enumerated across the pre-litigation arc rather than claimed generally: claim setup, letter of representation, policy limit verification, medical record retrieval, exhibit organisation, medical chronology, demand drafting, multi-plaintiff demand packages, demand benchmarking and adjuster preparation. Multi-plaintiff matters are named as a distinct capability. Named customers span firm types from regional practices to high-volume advertisers.

Depth within personal injury is real; breadth is narrow by design and that is a product decision rather than a gap. A is unavailable because the limits are not stated: nothing says which case types, injury categories or jurisdictions fall outside coverage, no statement addresses whether the product suits litigation-stage rather than pre-litigation work, and while a testimonial references jurisdiction-specific requirements no jurisdictional scope is published. Checked 8 September 2026.