VerdictCapital

14 analysts.One verdict.

A long-term research committee that runs inside Claude Code. Thirteen AI analysts check a stock or ETF against one checklist, a fourteenth — a devil's advocate — attacks their work, and a lead AI agent writes one reasoned conclusion.

One-time purchase. Runs in your own Claude Code. Educational research, not financial advice.

Deliberation

Example: HALDEN, a fictional company

    Hearing the analysts…

    Criteria met, but not at this priceScore 10.5 of 13

    Illustrative only. The company, ticker and results are invented to show the format. Real runs research current public data when you use the project.

    Fourteen specialists, one long-term question

    Is this a business worth owning for five to ten years, at this price? Each agent answers one part of that question, with its own brief, its own thresholds and its own sources.

    Business quality

    Is this a good business, run by the right people?

    01Revenue model
    How the company makes money, how recurring it is, and whether it will keep making it.
    02Moat
    Network effects, switching costs, brands, cost advantages, and whether they show up in the numbers.
    03Industry
    Growth, structural trends, competitive dynamics and the company's position.
    04Management
    Track record, insider ownership, capital allocation and red flags.

    Financial health

    Do the numbers back up the story?

    05Growth (CAGR)
    Revenue, EPS and free cash flow growth over at least five years, and how consistent it was.
    06Profit margins
    Gross, operating, net and FCF margins, against direct competitors.
    07Return on equity
    ROE with a DuPont breakdown, so leverage cannot pass for quality.
    08Debt
    Debt-to-equity below 1, interest coverage and upcoming maturities.
    09Free cash flow
    Positive, growing, and backed by real earnings rather than accounting.

    Valuation

    Is the price reasonable for what you get?

    10Relative valuation
    P/E, P/S, EV/EBITDA and PEG against peers and the company's own history.
    11DCF valuation
    Intrinsic value in bear, base and bull scenarios, with the assumptions stated.

    Risk and discipline

    What could go wrong, and what would you do about it?

    12Major risks
    Business, financial, legal and geopolitical risks, rated by probability and impact.
    13Exit plan
    The price levels at which the criteria would be met, the signals that break the thesis, and what is only noise.
    14Devil's advocate
    Runs last. Reads the other thirteen reports, hunts for bias, names the analyses it dismantled, and writes the strongest case against.

    Then, the lead investor agent

    A lead agent with a disciplined, long-horizon brief reads all fourteen reports, answers the devil's advocate point by point, and writes the conclusion, together with the price levels that would meet the criteria and the signals that would break the thesis.

    A verdict, not a vote count

    A simple tally would treat a missed margin target like a dishonest management team. VerdictCapital weighs which checks failed and why.

    Half points, not a tally

    Each of the thirteen checks scores 1 for pass, 0.5 for partial, 0 for fail. A great business can fail a few and still clear the checklist; another can score 11 and not clear it. Which ones failed matters more than the total.

    Some failures are close to final

    An unsustainable revenue model, dishonest management, dangerous debt with negative cash flow, or a risk that could wipe out the investment.

    The devil's advocate is a gate

    It is not a fourteenth point. If it raises a high alarm, or takes apart two of the reports behind the conclusion, the conclusion moves one step down — unless the lead agent rebuts each one point by point, in writing, in the report.

    CRITERIA MET

    Both the business and the price clear the checklist.

    CRITERIA MET, BUT NOT AT THIS PRICE

    The business clears it, the price does not. The report gives the price at which it would.

    CRITERIA NOT MET

    Decisive checks failed. The report says which ones, and why they settle it.

    The report states what the checklist shows, never what you should do. These labels, and any price levels, exit levels or position-size figures in a report, are the output of an automated AI research process that knows nothing about your finances, goals or risk tolerance. They are not investment advice or personal recommendations, and they can be wrong. Treat them as a structured starting point for your own research.

    From download to first report in minutes

    1. Unzip and open

      Open the folder in Claude Code and put your email in .env, which the SEC requires for automated requests. No API keys, no servers, no accounts beyond your own Claude Code setup.

    2. Name a ticker

      Type /analyze MSFT, an ETF like /analyze VWCE, or even an ISIN. One dated price and one set of filed figures are fetched first, thirteen analysts then work from them in parallel, and the devil's advocate reviews the result.

    3. Read and verify

      The conclusion appears in the conversation and is saved as Markdown in analyses/, next to all fourteen analyst reports, so you can audit the reasoning behind it.

    $ cd verdictcapital
    $ claude
    > /analyze MSFT
      Price snapshot: 1 dated quote, labelled live / today's close / last close
      Fact sheet: as-filed SEC figures, six fiscal years + TTM
      13 analysts running in parallel on that same data…
      Devil's advocate reviewing the synthesis…
      Report saved to analyses/MSFT-<date>/report.md (+ 14 agent reports)
    /analyze TICKER

    The full run: fourteen agents, one conclusion, saved with every underlying report.

    /quick TICKER

    A five-analyst screen that answers one question — is a full run worth it? No conclusion on the instrument.

    /review TICKER

    Re-checks a saved thesis against today's price: intact, watch, or a trigger fired.

    A report you can check line by line

    Every analyst uses a fixed format with a verdict and a score out of ten, and the research analysts cite their sources, so disagreements are easy to spot. A checker runs over the finished report and flags a score that does not add up, a missing analyst or a missing disclaimer.

    When data is missing or sources conflict, the agents are instructed to say so rather than fill the gap.

    • The conclusion, the score out of 13 and the alarm level
    • One-line summary from every analyst, plus their full reports
    • The case for the business, or the reasons it fails
    • The devil's advocate's objections, and the answers
    • Corrections: factual errors found in the analyst reports
    • The monitoring plan and the signals that break the thesis
    Read the full example

    HALDEN

    Illustrative excerpt · fictional company · invented figures

    Criteria met, but not at this price

    Score 10.5/13 · devil's advocate alarm: medium. A strong business whose price already reflects most of its expected growth.

    Selected analysts

    AnalystVerdictSummary
    MoatPASSHigh switching costs for customers
    DebtPASSDebt-to-equity 0.3, coverage 18x
    Major risksPARTIAL31% of revenue from one sector
    DCF valuationFAILBase-case value below the price

    Devil's advocate

    “Five-year growth is flattered by two unusually strong years.” Answer: the base case already assumes slower growth than that.

    Thesis-break signals

    • Recurring revenue below 70% for two quarters
    • Debt-to-equity above 1 to fund an acquisition

    Invented example to show the format. Not research on a real security. Not financial advice.

    Is it for you?

    A good fit if you

    • Hold stocks or ETFs for years, not days
    • Want a structured checklist before you commit money
    • Like to see the argument against your idea
    • Already use Claude Code, or want a practical reason to start
    • Want to learn how multi-agent Claude Code projects are built
    • Are comfortable running one command in a terminal

    Not a fit if you want

    • Day-trading or short-term signals
    • Technical analysis, charts or options data
    • Personal advice based on your own finances
    • Guaranteed returns or predictions (nobody can offer these honestly)
    • A web app. This runs in Claude Code on your computer.

    Why it is built this way

    Not a black-box signal service. A transparent research process you own and can change.

    You hear the case you would skip

    Confirmation bias is expensive. One agent exists only to argue against the idea you already like, and the verdict has to answer it.

    One price, one set of figures

    Before any analyst starts, the project pulls a single dated quote and a fact sheet of as-filed SEC figures, and every agent works from those. No analyst quietly uses a different number.

    A plan made while you are calm

    Every report defines the signals that would break the thesis and what is only noise, before a market panic or a rally tests your nerves.

    You own every prompt

    All fourteen agents and the lead investor are plain text files. Change a threshold, rewrite a brief, or add your own analyst.

    No subscription

    Pay $29 once. It runs on your own Claude Code access, as often as your Claude usage allows.

    The full committee

    $29

    One-time payment. VAT included where it applies, so $29 is what you pay.

    Buy VerdictCapital for $29

    Requires Claude Code with an active Claude plan or API access, which is not included, plus Python 3.9 or newer for the data tools. On our own test runs a full analysis took roughly 400,000 to 600,000 tokens and 5 to 15 minutes; your usage will vary.

    What you get

    • 14 specialist agents, including the devil's advocate
    • Three commands: /analyze, /quick, /review
    • Price snapshot and SEC fact-sheet tools, no API keys
    • A report checker and 33 offline tests
    • Works for stocks and ETFs, with adaptations for banks, insurers and REITs
    • Every analyst report saved, so the reasoning can be audited
    • Setup guide and an illustrative example report
    • Free updates for version 1.x
    • Personal license with unlimited runs (within your Claude usage limits)

    Questions

    Is this financial advice?

    No. VerdictCapital is an educational research tool. It organizes public information from fourteen angles, but it knows nothing about your finances, goals or risk tolerance. We are not authorised by any financial supervisory authority to provide investment advice. Verify important figures yourself, and consider a licensed advisor for personal decisions.

    Does it tell me to buy or sell?

    No. A report says whether the checklist was met, not what you should do: "criteria met", "criteria met, but not at this price", or "criteria not met". That conclusion, and any price level in it, is the output of an AI process with a fixed checklist, not a recommendation to you. It can be wrong, and the decision is always yours. If you share a report publicly, keep its disclaimer and say it was AI-generated.

    What do I need to run it?

    Claude Code installed on your computer, with an active Claude plan or API access and web search available, plus Python 3.9 or newer, which comes preinstalled on macOS and most Linux systems and can be installed on Windows from python.org. Python runs the two data tools and the report checker; without it the analysis still runs, with the analysts researching the figures on the web instead. Setup is one step: put your email address in a .env file, because the SEC requires automated requests to identify who is making them. You don't need to write code; you type /analyze AAPL.

    Does it work for ETFs?

    Yes. For ETFs, company-level checks are adapted (for example, the fund issuer and costs replace management) or marked not applicable, and the score counts only the checks that apply.

    Where does the data come from?

    Three places. The price comes from one dated snapshot taken when you start the run, from Yahoo Finance's public chart data, and it is labelled live, today's close or last close. The financials come from what companies actually filed with the SEC, read from EDGAR. Everything else — segments, guidance, peers, news — each analyst researches on the web and cites. Nothing is fetched, stored or redistributed by us: the tools run on your machine, under your own responsibility, for your own personal use, and you are responsible for the terms of any third-party service you access through them. The Yahoo Finance endpoint is a public but unofficial one and can change or stop working without notice; when it does, the run falls back to a dated web quote and labels it as such. We are not affiliated with, endorsed by or licensed by Yahoo or the SEC. AI can still misread figures, so check anything important against the original filings.

    Which markets does it cover?

    The price tool handles US tickers, share classes, ISINs and the main non-US exchanges. The SEC fact sheet covers companies that file under US-GAAP, which includes foreign filers such as ASML; for IFRS filers, ETFs and non-US listings it says so plainly and the analysts research the financials on the web instead. Banks, insurers and REITs are flagged, and the checklist adapts to them. Smaller or thinly covered companies produce more "data not found" notes.

    How long does a run take, and what does it use?

    On our own test runs, with every analyst on the main model, a full analysis took roughly 400,000 to 600,000 tokens and 5 to 15 minutes, depending on the company. That is far more than a single chat. Your figures will differ with the company, your model and your Claude Code setup, so treat this as a rough guide rather than a promise. To keep usage down, nine of the analysts ship set to a lighter model, and /quick screens a company with five analysts for about a third of the cost before you commit to a full run. Your own plan decides what that usage costs you.

    Can I change the agents?

    Yes. Every agent is a plain Markdown file you can edit for your own use. Redistributing or reselling the project is not allowed under the license.

    Can I get a refund?

    Yes. Within 14 days of purchase you can ask for a refund for any reason, and no reason is needed. Email hello@verdictcapital.ai from the address you bought with, or reply to your receipt; the refund is issued by Lemon Squeezy, our merchant of record, to the payment method you used, and your licence ends when it is issued. This matches the EU 14-day right of withdrawal, which we honour in full even though the download starts immediately — so if the checkout asks you to confirm immediate delivery, that confirmation does not take this refund away. Details in the refund policy.

    Hear all fourteen before you decide.

    Get VerdictCapital for $29