Chronograph

The chronology settles this. The platform was built from 2016 as portfolio monitoring and data management, and the artificial intelligence layer arrived in 2024 as a synthesis and summarisation product over documents already held, extended in 2026 with semantic search.

Strip the models and a working system remains: data collection workflows, reconciliation, fund and company level analytics, valuation support, dashboards and investor reporting, which is what clients bought for eight years before any model was offered.

Extraction from unstructured documents is genuine model work and keeps this well clear of the centrality floor, but the company's own framing puts the reconciled dataset first and treats the models as what makes that dataset easier to interrogate.

The output is research and reporting consumed by investment and operations teams rather than a decision executed on a client's behalf, so the oversight burden is lighter than for vendors acting in markets, and the company describes a reconciled dataset as the foundation, which implies a checking step ahead of anything the models touch.

Semantic search is grounded in retrieval across the client's own documents, so answers point back to specific source material rather than being generated from a general corpus, which is a real safeguard against fabricated figures. What is not published is any account of who reviews an extracted value that fails reconciliation, or whether a synthesised answer carries a confidence signal.

Reconciliation and a trusted dataset are the published claims, and both describe an intention rather than a measurement. No extraction accuracy rate appears, no error rate by document type, no retrieval recall for the search product, no drift monitoring, no revalidation cadence and no model documentation of the kind an institutional risk function would request.

The company states its models were trained on a large universe of private markets data across years of research, which speaks to the training input rather than to production performance. This is the index norm rather than a failing unique to this vendor.

This is the strongest evidence surface located in this lane. A managing director at one of the largest listed alternative managers is quoted by name with a quantified result, describing more than seven million data points captured and structured from that firm's portfolio, and a managing partner at a European secondaries manager is quoted separately by name.

Platform scale is stated precisely rather than in round numbers: more than five point nine trillion dollars of invested client capital monitored, roughly fifteen thousand funds and two hundred and fifty eight thousand portfolio companies. More than two hundred limited partner and general partner clients attended the company's first user conference in 2026, which is a headcount that is hard to inflate.

The published architecture points outward rather than inward, which is the useful contrast in this lane. The artificial intelligence layer operates on the documents a given client already keeps on the platform, a data warehousing product delivers that client's data into its own cloud data warehouse where the client can build its own applications, and a model context protocol connector lets an institution reach its data from an assistant it controls.

Nothing in the published material advertises a benchmark or comparable dataset derived from pooled client portfolios, which two competitors in this lane do market. What is still missing is the explicit statement: no published boundary says whether documents from one client inform models serving another.

The material held is institutional, covering fund financials, partnership agreements, capital accounts and portfolio company performance, so consumer financial privacy law is largely out of frame and the grade reflects what is published rather than a penalty for the category.

Searched the platform and company material for a privacy statement addressing handling of client documents, retention, deletion on termination or the treatment of portfolio company information belonging to third parties who have no relationship with this vendor, and located none. Partnership agreements and side letters sitting on the platform carry explicit confidentiality obligations, and nothing published describes how those are honoured.

A third party comparison reports both a service organisation control type one and a type two report, running on a major public cloud platform. The type one report is the more interesting of the pair and is rare in this index, because it is the report written for the auditors of the customers rather than for the customers themselves, which is the right instrument when a vendor's output feeds financial statements.

The grade sits at B rather than higher because the evidence comes from a third party compilation rather than the company's own compliance material, no trust centre was located, and no penetration testing or incident response statement was found. Confirm directly on a later pass.

No licence is held or required and the company presents itself plainly as a technology provider to institutional investors, which the convention here does not penalise. What is absent is the regulatory chain its buyers stand in.

Pension plans, insurers and registered advisers running on this platform produce valuations, performance figures and investor reporting under supervisory and financial reporting obligations, and nothing published explains how an extracted or synthesised figure is evidenced to an auditor or examiner, or how the lineage from a source document to a reported number is reproduced on demand years later.

The lending fairness frame does not apply and the real exposure has two parts. Extraction accuracy tracks document format, and format tracks the size, age, jurisdiction and language of the manager producing it, so an allocator's data is likely to be weakest on its smallest and newest commitments.

The second part is specific to semantic search and sharper: a natural language query returns a confident synthesis built from whatever the retrieval step surfaced, and an omitted document produces an answer that looks complete and is not. No recall figure, no coverage statement and no indication of what the search did not see accompanies the output, so a user cannot distinguish an absence of evidence from an absence in retrieval.

Trust is the word the company and its quoted clients use most, and no published commitment stands behind it. Extracted values feed valuations, performance reporting and investor communications at institutions managing pension and sovereign capital, so a wrong number travels quickly into places that are expensive to correct.

Nothing published states a service level, a warranty, an obligation to correct or any remedy, and nothing states the opposite either, since no disclaimer was located placing responsibility for a reported figure back with the institution. Reconciliation lowers the odds of an error without settling who carries one.

One dependency is named openly and it is unusual to see it stated so plainly: the company publishes a connector to a specific external model provider through the model context protocol and describes the joint capability in its own announcements, so an institution can see exactly which outside party is reachable from its data and under whose account. The company also credits large language model technology generally for the capabilities it launched in 2024.

What is not disclosed is the stack behind its own extraction and summarisation, meaning which models perform the work when a client uses the platform directly rather than through the connector, and whether those run in house or are called out to a provider.

Integration runs in three directions and each is named. Inbound, the platform collects from managers and administrators and integrates with fund accounting and investor reporting systems. Outbound, a data warehousing product lands a client's monitoring data inside a major cloud data warehouse so the institution can build its own applications on it without exporting files.

Third, the company publishes a model context protocol connector to a major model provider, letting an institution reach its own portfolio data from the assistant it already uses. That third route is the machine addressable distribution pattern this index has tracked since its first builds, and it is the first appearance of it in the private markets lane.

Delivery is vendor hosted cloud software, reported by a third party comparison as running on a major public cloud platform, and the company does publish a route for a client to hold its own copy of the data in its own cloud warehouse. What is absent is everything the axis actually asks for: no hosting regions are stated, no residency commitment is offered and no single tenant or client hosted option for the platform itself is described. The client base includes European allocators and Gulf sovereign investors who commonly impose location terms by contract, so the omission is material rather than cosmetic.

No rates, tiers or unit of charge are published, and the route to a number is a conversation with the company. The gap is compounded by a product line that splits across at least four purchasable things, the limited partner platform, the general partner platform, the artificial intelligence layer and a data warehousing product, with nothing published on whether the models are included, bundled or charged separately. A buyer cannot tell whether adopting the newer capabilities changes the cost of what they already run.

The buyer list is enumerated rather than gestured at, covering funds of funds, pension plans, sovereign wealth funds, insurance companies, outsourced chief investment offices, foundations, endowments and family offices on the allocator side, and general partners on the manager side, with the company describing coverage across the assets under management spectrum, asset classes and geographies.

Both directions of the same relationship are served by separate products rather than one platform stretched to fit, and private credit is addressed specifically with compliance certificate extraction and add back calculations, which is a distinct workflow from private equity rather than a relabelled one.