Carta

The established platform shape again, graded the same way as the data platform built alongside it this session, which is a useful consistency check on both. The company ran for over a decade as a capitalisation table and fund administration system before any of this existed, and strip the agents out today and the ledger, the cap table, the tax product, the compliance tools and the investor portal all continue to work.

Worth flagging for a future re grade though: unlike a chat layer bolted on top, the agentic ingestion here feeds the general ledger directly and the monitoring agents sit in the transaction path, so the line is moving toward the core rather than staying beside it. If the accounting close genuinely stops working without the agents, this becomes a B.

Two genuine signals. Agents process each detected signal in the context of the governing partnership agreement, so behaviour is grounded in the document that actually controls the fund rather than in a generic rule set, which is the strongest form of context binding this index recognises. And the company states plainly that agents run continuous background health checks while the last mile belongs to experienced human experts.

Held off A because the last mile statement is positioning rather than a described mechanism: no threshold, no review rate, no escalation criterion, and nothing that says which agent outputs reach the ledger without a person seeing them. Set against a stated goal of removing the need to log events manually, that is the gap that matters.

The auditability claim is real but it is about the substrate rather than the models: a deep layer of auditable data running from a single company's capitalisation table to a fund's general ledger. Nothing describes how the models themselves are evaluated, no accuracy or extraction error figures for the ingestion agents that write to the ledger, no documented failure modes, no drift policy and no model documentation.

The continuous health checks are agents doing monitoring rather than monitoring being done of the agents. For a product whose output lands in fund accounting, extraction accuracy is the standing question and it is unanswered.

Adoption evidence for the generative products specifically, which is rarer than it sounds: more than 1,500 companies and firms using the assistant plugins within months of launch. Platform scale is documented in volume terms that are checkable in kind, with monthly document processing in the tens of thousands and over a million journal entries a year. Customer quotes appear on the platform pages. Held at B because no named institution is attached to a measured outcome from the agentic products, so adoption is evidenced and performance is not.

The pooled corpus question is answered in the negative and in plain words, that customer data is never used to train models and stays private and internal, with contractual protections named with the model providers. Held off A, and the contrast with the data platform built in this same session is the instructive part: identical commitment, delivered as one marketing sentence rather than as a governance document.

The A grade there rests on enumerated data sources, a stated isolation guarantee between firms, named retention and purge periods, model hosting policy and supplier posture. Here there is the promise and none of the surrounding detail that would let a buyer verify or enforce it.

The central commitment is made plainly and without hedging: customer data stays private and internal and is never used to train models, backed by stated contractual protections with the providers involved. That is the sentence most vendors in this index will not write. Held off A because it stands alone, with no data processing agreement, subprocessor list, retention or purge terms, residency commitment or named privacy regime accompanying it.

A named attestation with imprecise qualifiers, which is the familiar middle of this axis. The generative material states that the intelligence runs in a closed service organisation control environment with strict contractual protections. Catalogue the phrasing as a new imprecision variant worth watching: describing an ENVIRONMENT as attested rather than stating that the company holds a report of a given type, scope and period.

Held off A on exactly that, plus no located trust portal, no report availability statement and no penetration testing disclosure. Banked check: a fund administrator of this size holding client books and records will hold both a service organisation control report over financial reporting controls and one over trust criteria, so look for a dedicated trust page.

No supervised status credited, because none was verified in this pass. Banked check, and it is a cheap one that could move this grade materially: this company has historically operated regulated entities in the United States, plausibly including securities transfer agent registration and a broker dealer subsidiary for its private market trading venue.

Transfer agent registration would matter here in a way that most group licences do not, because it attaches to securities recordkeeping, which is the graded product itself rather than an adjacent business. Verify before crediting.

No fairness testing or governance disclosure. Exposure is lower than in consumer facing lending or pricing, since the subjects are funds and transactions rather than individuals, but it is not absent: the relationship intelligence product ranks leads and evaluates targets from years of historical data, and ranking which opportunities and which people surface first is a consequential ordering even when no individual is being scored for credit.

Stronger than silence and worse for a buyer: the published material carries an express disclaimer that the company does not assume any liability for reliance on the information provided. Nothing allocates responsibility for an agent that misreads a capital account statement and writes a wrong figure into a general ledger, and nothing describes detection or correction when that happens. Worth recording as a distinct shape on this axis, since most entries here are absence of a stated position rather than an affirmative disclaimer of one.

The assistant platforms the product interoperates with are named openly, which is unusual candour, and the open protocol server architecture is named as the mechanism. Contractual protections with providers are asserted. Held at B because naming the assistants a customer can drive the platform from is not the same disclosure as naming the model behind the proprietary ingestion agents that write to the ledger, and that second question is unanswered. Same inversion as the data platform built this session: the how is described and the whose is not.

The strongest integration position found in this session, and it is structural rather than claimed. The platform is itself the system of record, holding the general ledger and the capitalisation table, and it connects outward to bank feeds, portfolio company data and investor records, with payment authorisation running through named financial institution partners at around three billion dollars of volume.

On top of that it exposes an open protocol server architecture and ships plugins that let a firm run fund administration, cap table and dealflow workflows from inside general purpose assistants, with adoption in the low thousands of firms. A data warehouse serves the same data to natural language queries. Owning the record and reaching outward from it is what an A on this axis is supposed to describe.

Software as a service with nothing published on region, residency, tenancy or hosting arrangements. For a platform holding the books and records of funds across multiple jurisdictions, and now routing that data through general purpose assistant platforms via plugins, residency is a question a buyer will ask and the material does not answer it.

No published price located for the fund administration, loan operations or analytics lines. Banked check that could move this: the company has historically published tiered pricing for its capitalisation table product aimed at early stage companies, and if those tiers are still public then at least one product line is priced openly, which is more than the rest of this pocket manages.

Venture and private equity funds, private credit funds, funds of funds, institutional limited partners and allocators, general partners, portfolio companies and corporates, across fund administration, fund accounting, tax, compliance, capitalisation table management, loan administration, investor reporting and relationship management. Roughly 9,000 funds and more than 203 billion dollars of assets, about 125,000 allocators, more than 50,000 capitalisation tables. Both sides of the private markets relationship are served on one platform, which is unusual.