StockUp Logo
Open StockUp Blog Risk Disclaimer Terms
About StockUp

Who publishes this research, and how

StockUp builds Quan, a finance-native AI reasoning system. This page explains who produces the analysis published on this site, how that analysis is generated and checked, which data sources sit underneath it, and — importantly — what the models cannot do.

Short version: research and educational content, produced by the StockUp engineering team using our own models, grounded in verified market data and primary filings where available. Not personalized financial advice.

Who writes this

Content on this site is published under the StockUp Engineering byline. It is written by the team that builds and operates the Quan models — not by an outside content agency, and not auto-published without review. We attribute to the organization rather than to individual analysts because the analysis is a product of the model plus the engineers who build, benchmark, and correct it.

We are a small engineering team, not a registered investment adviser, broker-dealer, or research house. Nobody on the team is a licensed financial adviser, and nothing published here is personalized advice. Where a page presents a rating, a target, or a forecast, it is model output being reported and explained — not a recommendation.

How Quan produces an answer

Quan is a finance-specific reasoning system rather than a general chatbot with a finance prompt. The behaviours that matter for accuracy:

Server-verified quote context

For supported requests, market context is verified server-side before the model reasons over it, rather than relying on the model to recall a price.

Explicit unavailable-data handling

When a required data point is not available, the intended behaviour is to say so rather than produce a plausible-looking number.

Deterministic calculation

Arithmetic is computed rather than generated, so a figure in an answer is not a token prediction of what the figure ought to look like.

Opt-in live grounding

Live web grounding is off by default (googleSearch: false) and billed per executed search when enabled. An answer is not silently searching the web unless asked to.

We publish our own adversarial benchmark results rather than only headline scores: see Why finance AI needs truth boundaries for the methodology behind Quan 3.4's reported 93/100, and Quan vs ChatGPT on a 10-K for a worked filing example with the prompt and recorded outputs. Both are our own internal benchmarks, not third-party audits — treat them as vendor-reported.

Data sources

Analysis on this site and through the API draws on:

  • Primary company filings — 10-K, 10-Q and 8-K documents from the SEC's EDGAR system. Filing-derived claims on our ticker pages link to the company's EDGAR filing index so you can check the source.
  • Market and quote data — supplied via Finnhub. In the consumer product you connect your own Finnhub key; API requests use server-side verified context.
  • Optional live web grounding — powered by Tavily, opt-in per request and billed per executed search.

We are not an exchange-licensed market data redistributor, and Quan is not a complete raw market-data feed. Where you need authoritative pricing or an exchange-licensed feed, use a market data vendor.

What these models cannot do

Stating limits plainly is part of the product:

  • They cannot predict prices. Targets and ratings are estimates conditioned on available information, not forecasts with a knowable error bar.
  • They can still be wrong. Truth boundaries and deterministic calculation reduce fabricated figures; they do not eliminate error, stale data, or misread context.
  • They are point-in-time. A published rating reflects the information available on its date and is not continuously revised.
  • They are not personalized. The models know nothing about your circumstances, tax position, time horizon, or risk tolerance.
  • They are not a trading venue and do not execute orders.
  • Retention: do not treat the public API as zero-retention. Enterprise data-handling behaviour is established through Enterprise configuration and contract review, not assumed.

Corrections

If you find an error in something we have published — a wrong figure, a misread filing, a stale rating — tell us and we will correct the page and note the correction. Reach us on X, LinkedIn, or Reddit.

Full legal terms: Risk Disclaimer · Terms of Service · Privacy Policy

StockUp Developer API Docs Blog Pricing