Ledger

Ledger — AI Disclosure

DRAFT — NOT LEGAL ADVICE. This document was drafted by the Ledger engineering team, not by a qualified lawyer. It must be reviewed by a UK-qualified solicitor before Ledger is offered to the public. Neither the author nor COTR Global Group Ltd guarantee its legal sufficiency in its current form.

Effective date: to be finalised at launch Last updated: 2026-07-28

1. What Ledger is — and isn't

Ledger is a research aid that uses AI to survey the public web for evidence about a market question you supply, and produces a structured narrative report. It is not:

  • Investment advice
  • Legal advice
  • Business advice
  • Tax advice
  • Medical advice
  • Professional advice of any kind
  • A substitute for talking to a human expert

If you are making a decision that involves money, contracts, health, or another consequential outcome, you should treat a Ledger report as one input among several — not as the answer.

2. How reports are generated

A Ledger report is produced by a pipeline that combines:

  • Large language models (LLMs) provided by Anthropic PBC (specifically models in the Claude family). LLMs are used to interpret your question, generate the set of hypotheses to investigate, reformulate search queries, classify collected evidence, and compose the final narrative.
  • Web search provided by Tavily. Tavily returns pages from the public web that match search queries derived from your question and its hypotheses. Ledger does not itself crawl the web.

Every hypothesis and every piece of evidence used in the report is recorded in the report itself, with the source URL from which it was retrieved. You can follow every citation back to its source.

3. What can go wrong

You should understand the following limitations before relying on anything a Ledger report says:

  • LLMs can hallucinate. They can generate text that sounds plausible but has no basis in the sources they were shown. Ledger's design reduces this risk by grounding every claim to a specific piece of cited evidence — but a mis-summary of what a source actually said is still possible.
  • Web search is noisy. Tavily returns what it finds. It cannot distinguish authoritative sources from unreliable ones, and it does not screen for factual accuracy. Sources cited in a Ledger report may themselves be wrong, biased, out of date, or promotional.
  • Evidence classification is judgemental. Ledger categorises each piece of collected evidence (for example: verified paying behaviour, claimed paying behaviour, expressed frustration, market existence, secondary opinion, or unclassified). This categorisation is done by an LLM and can be wrong.
  • Verdicts are deterministic — but only over what was collected. Ledger assigns each hypothesis one of three verdicts (strong signal, mixed, insufficient evidence) using a fixed rule over the evidence that was collected and classified. The rule is transparent and reproducible. But the verdict is only as good as the evidence set — if search missed something important, the verdict reflects that gap.
  • Facts stated in sources are recorded verbatim; missing facts stay missing. Ledger will not "fill in" a market number a source didn't actually give. If a report says a figure is absent, it means no source in the collected evidence stated it — not that the figure is zero, small, or unimportant.
  • Reports are point-in-time. The public web changes constantly. A report reflects what Ledger found when your run was generated; facts and citations can go stale or link-rot within days.

4. What we do to reduce risk

  • No silent dropping. Every hypothesis Ledger considers appears in the final report — including weak ones. If a hypothesis has insufficient evidence, it is shown as such, not quietly omitted.
  • Every claim is cited. Every piece of evidence in the report links back to the source URL from which it was retrieved. You can verify.
  • No fabricated numbers. We do not substitute plausible-looking values for missing data. If a source didn't state a figure, that figure is marked absent.
  • Deterministic verdicts. The verdict for each hypothesis is a deterministic function of the classified evidence, not another LLM judgement. The rule is documented and consistent across runs.

None of the above eliminates the risks in §3. They reduce them.

5. What you should NOT use a Ledger report for

You should not use a Ledger report as the sole basis for:

  • Investment decisions — buying or selling securities, allocating capital, valuing a company or asset. Ledger is not a financial adviser and reports are not a substitute for regulated financial advice (Financial Services and Markets Act 2000).
  • Legal decisions — evaluating contracts, regulatory positions, or litigation risk. Consult a solicitor.
  • Business decisions with material commitments — hiring, firing, major spend, entering or exiting a market, launching a product. Use the report to shape your thinking, then validate the specific underlying facts.
  • Any decision affecting someone else's rights or wellbeing — including medical, safety, or employment decisions.

If you use a report for any of these purposes anyway, that is your choice and your responsibility. See the Terms of Service §11 for the limits of our liability.

6. Sources cited in a report

A Ledger report cites external web pages via URL. Those pages:

  • Are not authored, endorsed, or verified by us.
  • May contain errors, opinions, advertisements, or paywalled content.
  • May become unavailable (404), be modified after the report was generated, or move to a new URL.
  • Are subject to their own terms of use, which you must comply with if you visit them.

The Report captures the URL and a snippet of the page as they existed when the run was generated. Following the link later may show different content.

7. Your responsibility to verify

Before acting on anything material in a Ledger report:

  1. Read the underlying source. If a source says the market is growing 30% per year, click the link and confirm it actually says that, in context.
  2. Check the source's credibility. Is it a trade publication, a promotional blog, a Reddit comment, a market-research vendor with an incentive to inflate numbers? Ledger cannot tell you this.
  3. Cross-reference. Does another independent source say the same thing?
  4. Consider what's missing. If a report has insufficient evidence on a hypothesis you thought was central, that's a signal — either for more research or for humility about how much you know.
  5. Consult a professional. For any decision in the categories in §5.

8. If you disagree with a report's contents

Because every claim is cited, disagreements are usually resolvable by following the citations. If you believe a report is materially wrong in a way that isn't just "I read the sources differently," email us at legal@goldrush.capital with your Run ID and what you think is wrong. We will investigate and, if appropriate, offer a re-run or refund per the Terms of Service §7.

9. Automated decision-making (UK GDPR Article 22)

We do not use your personal data to make automated decisions with legal or similarly significant effects on you. The report-generation process uses AI to produce a research document, but the document itself has no automatic legal effect — every consequential decision remains yours to make.

10. Contact

Questions about this AI Disclosure: legal@goldrush.capital

REVIEW POINT (launch-blocker): the legal@goldrush.capital alias must be configured and receiving mail before this Disclosure is published.