Reflexivity

The removal test leaves four licensed data feeds and nothing to do with them. Agents interpret a financial question, write and execute code against the underlying data and return an analysis, a knowledge graph infers relationships between companies, themes, geographies and counterparties that no feed states directly, and language models read filings, transcripts and presentations at volume.

Screening on qualitative signals and sentiment, scenario simulation against hypothetical shocks and hypothesis testing against history are all model work. The founders built it from the problem of managing a global macro book, and what they built is the analysis layer rather than the data underneath it.

Autonomy exposure is inherently limited because the product analyses rather than acts, with no execution, order routing or portfolio authority, and the company describes its purpose as empowering human decision makers. What lifts this above the ordinary is a mechanism most vendors in this index lack entirely: the system is engineered to acknowledge its own limitations and to state explicitly when data is unavailable or uncertain, so the analyst is told where the analysis is thin rather than receiving uniform confidence throughout. That is calibrated uncertainty exposed to the user. What is absent is any described review of agent produced research before it circulates inside a firm.

The fifth A on this axis and the most complete set of mechanisms recorded so far. Four things stack. The agent writes and executes code to reach an answer, so the analysis is reproducible and inspectable rather than asserted, which is a materially stronger guarantee than a citation. Every insight is traceable to its source data and carries a data quality rating, so the user sees both where a number came from and how reliable that source is.

The system is built to state explicitly when data is unavailable or uncertain rather than answering anyway, which is the failure mode that makes generative analysis dangerous in this setting. And a full audit trail records provenance, methodology and rationale. Inputs are restricted to licensed institutional data rather than open web retrieval. One claim should not pass unchallenged: zero hallucination is marketed as an outcome, and no generative system can guarantee it, so treat it as a design aspiration rather than a property.

The investor register is the strongest signal and it is unusual, because several backers are the customer type. Around 40 million dollars has been raised across two rounds, with a 30 million dollar Series B led by a venture firm and a major electronic broker, joined by two of the best known macro hedge fund managers, a large Japanese bank's venture arm and a multistrategy hedge fund.

Institutional money that also trades for a living choosing to fund a research tool is a different kind of endorsement from generalist venture capital. The broker relationship has produced shipped product, with a jointly launched investment themes tool and a stated intention to integrate the analysis into the broker's trading platform. Against that, no institutional client is named, no user count is published, and no accuracy or research quality outcome is measured.

The stated absence of persistent client data storage answers the question this axis exists to ask, and answers it structurally rather than by promise. The concern for a shared research platform is the Rogo problem, that competing funds use the same system and one firm's research direction could inform what another sees, and queries that are never retained cannot train a model or surface anywhere else.

Held below the top grade because nothing states whether ephemeral processing still passes through external model providers, what those providers may retain, or whether aggregate usage patterns are analysed even when individual queries are not.

One architectural commitment does the work and it is the right one for this product. The company states end to end encryption with no persistent storage of client data, which matters because the sensitive material here is not personal information but intent: the questions a fund asks, the securities it screens and the scenarios it stress tests reveal its positioning and its thinking, and that is among the most closely guarded information an investment firm holds.

A platform that does not retain those queries cannot leak, subpoena or repurpose them. Held at B because the claim is a single unqualified line with no supporting detail, no subprocessor list and no data processing terms published.

Two specific practices are published rather than asserted. Annual third party audits are stated as a standing commitment, which is an independent assessment recurring on a cycle rather than a one time exercise, and controls are described as meeting institutional due diligence questionnaire requirements, which for this buyer is the operative test since a hedge fund or asset manager will not onboard a vendor that cannot answer one. End to end encryption is stated alongside. Held at B because no individual framework or certification is named, so a reader cannot tell what the audits assess or against which standard.

No supervisor, statute or instrument is named. A full audit trail is offered for compliance requirements without identifying which requirements, and the gap has widened with the broker channel. Analysis distributed to institutions sits near research rules on substantiation and conflicts, and the same analysis surfaced inside a retail trading platform sits inside communications standards governing what may be shown to ordinary investors, including fair and balanced presentation and the treatment of anything resembling a recommendation. None of that is addressed in published material.

No consumer decision applies, so the axis adapts, and the governance question here is systemic rather than individual. It is also the question the company is named after. If thousands of investors, institutional and now retail through a broker integration, query similar models trained on the same licensed data and act on similar outputs, positioning becomes correlated, feedback loops shorten and crowding increases, so a tool that makes each individual better informed can make the market as a whole more fragile.

Independent commentary raised exactly this on the funding announcement. Nothing published addresses whether outputs are differentiated between users, whether crowding is monitored, or what happens when a large share of a broker's client base receives the same signal simultaneously. A second and narrower question also sits unanswered: which 40,000 instruments are covered, since the licensed providers' universes skew toward developed markets and larger issuers.

No guarantee or indemnity was located, and error detection is built into the output rather than promised beside it. Because every insight traces to its source with a quality rating attached, and because the agent's code can be inspected, a user who doubts a conclusion can follow it back to the data and to the method that produced it. The explicit acknowledgment of uncertainty means the system flags its own weak ground rather than leaving the user to discover it.

That is the Daloopa property with an additional layer. What is missing is anything after the fact: no correction, restatement or notification process is described for research later found to rest on a faulty step.

The data half of the chain is disclosed unusually well, with four premium providers named individually and the commercial arrangement stated, so a buyer knows exactly whose ratings, time series and exchange data sit behind an answer and that the licences are held by the vendor rather than required from them. That is more than almost any peer publishes. The model half is silent.

No provider is named for the language models reading filings or for the agent framework writing code, no hosting arrangement is described, and no subprocessor list appears, which matters because the code execution environment is where client queries actually run.

Both ends of the chain are named and both are substantial. Upstream, premium market data is licensed from four named providers covering ratings and fundamentals, historical time series and two major exchanges, and it is bundled rather than left to the customer to contract separately, which removes the usual procurement obstacle for a research tool.

Downstream, the analysis is being integrated into a major electronic broker's own trading platform, which is distribution through rails the customer already uses rather than another application to log into, the same structural position that lifts Signzy. That combination, institutional data included and delivery inside an existing execution venue, is what this grade is for.

No hosting provider, region selection, residency commitment or private deployment option was located. The stated absence of persistent client data storage reduces what is at stake, since there is less resting anywhere to place, but it does not answer where processing occurs, and a European or Asian institution subject to its own location requirements will ask. Nothing published addresses it.

No rate is published and a substantive commercial fact is. The platform states that premium market data from named providers is included with no separate data contracts required, which for this buyer is the largest variable in any research tooling decision, since licensing those feeds independently is a major cost line and a lengthy negotiation in its own right.

Knowing what is bundled tells a firm what it can stop paying for elsewhere, which is the shape of the deal even without the number. Nothing indicates whether charge falls per seat, per query or as a platform fee.

The buyer set spans institutional investors, asset managers, hedge funds, capital markets teams at investment banks, wealth management firms and exchanges, and asset coverage runs past 40,000 instruments across global equities and digital assets.

A newer channel reaches a different population entirely, since integration into a major electronic broker's platform puts the same analysis in front of independent traders and smaller advisory firms that never had access to institutional research tooling. What holds this at B is function rather than reach: this is research and analysis, and it does not extend into portfolio management, execution, risk systems or operations.