Building a Reliable Source Register

Hello! Your repository now has a place for each company’s evidence. The next step is to make that evidence traceable. A link to an annual report is useful, but an interviewer should also be able to see which claim you drew from it, where you found that claim, and when you accessed the document.

In about 40 minutes, you will create a source register for your first company case. It will be small enough to maintain while you work toward a first resume-ready brief, but structured enough to support later charts, models, and transaction studies.

Treat a source as evidence for a specific claim

A source register is not a bookmarks list. Its basic unit is a claim–source record: one statement you might use in your analysis, paired with the document and precise location that support it. The same annual report may therefore appear in several rows if it supports several claims.

A literature matrix demonstrates the organizing principle. The video uses academic papers rather than company filings; watch for how its columns make information findable, not for columns to copy into a finance project.

For your portfolio, the useful question is more exact: “If I put this sentence in a report, what could someone open to check it?” A URL alone may lead to a 200-page PDF. A document title, page or section, reporting period, and narrowly worded claim make the check practical.

Build a register that preserves dates and context

Create companies/<company-slug>/sources/source_register.csv. You can edit the file in Excel and save it as CSV UTF-8 so it is also readable directly on GitHub. Start with this header row:

record_id,source_id,issuer,document_title,source_type,url,publication_date,access_date,reporting_period,locator,claim_supported,reliability_notes

Use one row per claim–source pairing. Give each row a unique record_id (R001, R002) and each distinct document a stable source_id (S001, S002). If two rows draw on the same annual report, repeat its document details and source_id; change the claim and locator as needed.

The fields serve different purposes:

• url, issuer, document_title, source_type identify what to open and who published it. Prefer a direct document or disclosure link over a search-results page.
• publication_date is when that version of the document was issued, if you can establish it. access_date is when you opened it. Enter dates as YYYY-MM-DD; write Not stated rather than inventing a publication date.
• reporting_period states what period a figure describes, such as the financial year named in a report. It is not the publication date.
• locator gives a printed page number, table name, or named section. If the PDF viewer’s page number differs from the number printed on the page, say which one you mean.
• claim_supported is a specific statement the cited passage actually supports—not your entire investment thesis.
• reliability_notes records a relevant strength, limitation, definition, or follow-up check. “Credible” by itself tells a reviewer little.

These distinctions become especially important when you reuse a number in a spreadsheet. The following short reading explains why a retrieval date cannot stand in for a reporting period, and why company, units, and accounting basis must remain attached to financial figures.

You do not need a separate column for every possible financial definition today. Put an important distinction in the locator or reliability note: for example, whether a figure is consolidated or standalone, what units the table uses, or whether a measure is reported or adjusted. When those distinctions become recurring model inputs, you can add dedicated columns without losing the original source records.

Make reliability notes useful, not ceremonial

The CRAAP framework gives you five prompts for a short, practical note: currency, relevance, authority, accuracy, and purpose. At this stage, use them to flag what you know and what needs checking. A later lesson will examine source credibility more systematically.

For a company-issued annual report, a useful note might be: “Issuer filing; consolidated figures; historical table includes a comparative year; check whether the later filing restates that year.” For an investor presentation, it might be: “Issuer’s explanation of its strategy; verify numerical claims against filings; forward-looking targets are not achieved results.” Neither note dismisses the source. Each identifies what kind of claim it can support.

Keep the wording of a claim within the evidence. If management says higher volumes drove sales, your register can say, “Management attributes sales growth partly to higher volumes.” It should not silently become “Higher volumes caused all sales growth.” Likewise, a table of two reported revenue figures can support those figures directly; a growth rate you calculate from them is your calculation, which must retain the input locations and method.

Populate the first company’s register

Use the publicly listed company you selected earlier, or the leading candidate if you have not finalized it. Aim for two or three real documents, not an exhaustive bibliography. An annual report is a sensible starting point; add another annual report or an investor presentation if it is available and relevant. Work with public links rather than uploading PDFs whose redistribution rights you have not checked.

1. Open and identify a document. Record its actual title, publisher, direct URL, document type, publication date if stated, and today’s access date. Do not substitute the financial year-end for an unknown publication date.
2. Choose one checkable claim. A figure from a named financial-statement table or a narrowly described business segment is easier to verify than “the company has strong growth prospects.” Record the financial period and page or section alongside it.
3. Write one reliability note. Include a strength or limitation that matters to this claim: issuer-reported versus third-party, consolidated versus standalone, historical versus forecast, or a possible revision to check.
4. Add another claim or document. If it comes from the same PDF, keep its source_id and give the new row a new record_id. If you find a revised version of the document, assign that version a new source_id and note the relationship.

For example, a good claim_supported entry would have the shape “Consolidated revenue for FY[year] is [figure and units]”, with a locator such as “Consolidated statement of profit and loss, printed p. [number].” Those brackets are placeholders, not evidence: replace them only after checking the actual document. “Revenue is growing rapidly” would be a poor entry unless you specify the periods, figures, calculation, and meaning of “rapidly.”

In Excel, select the filled range and enable filters, as shown in the video. Filtering by source_id should show every claim taken from one document; filtering by reporting_period should help reveal claims that might accidentally mix years. Save the CSV, reopen it, and check that long URLs and notes survived the export.

A common time-wasting mistake is to register dozens of links before deciding what they support. For this first release, a small register with precise locators is more valuable than a large list you cannot audit.

Publish and test the evidence trail

In sources/README.md, add a link to [Source register](source_register.csv) and one sentence explaining that each row connects a claim to a public document and locator. The company README already points readers to sources/README.md, so this completes the route from case overview to evidence. Commit and push both files.

Before stopping, open your register on GitHub and test one row as if you were an interviewer unfamiliar with the company. Can you open the URL, find the cited page or section, identify the correct period and units, and see that the wording of the claim does not exceed what the source says? If the link fails or the passage does not support the sentence, fix the row now. A registered source is traceable evidence only when the connection has been checked.

If time is tight, finish one thoroughly verified document today and add the next tomorrow. Leave unknown fields visibly unknown rather than filling gaps from memory.

Takeaway

Your source register connects a specific claim to a specific version and location of a public source, while keeping publication date, access date, reporting period, and reliability notes distinct. That is the foundation for source-linked charts and models—and for explaining your work under interview questioning.

Next, you will learn to navigate annual reports and other company materials efficiently, so you can find the disclosures your register needs without reading every document from beginning to end.