Reporting is mostly a memory problem. You talk to twenty people, read forty documents, attend a hearing, sit through a press briefing, get a tip on background, find a ten-year-old filing on a court docket, and then write 1,400 words by Friday that has to be accurate against all of it. The actual writing is the smallest part. The bulk of the week is gathering, cross-referencing, and remembering — under deadline.

This guide is about putting the parts of a reporter's notes that benefit from being together — sources, interviews, documents, the running file on the story — into one vault where an agent can read across the record.

## What a reporter's notes have to hold

Reporting work fails when the right detail can't be found at the moment of writing. The quote you remember but can't locate. The document you read three weeks ago that has the answer to the question you're now asking. The contact who said something useful on background that you forgot to write down.

Four loads a reporter's vault has to carry:

- **A source file** — contacts, what they cover, history with you, on-record versus off-record context, last contact.
- **Interview transcripts** — every recorded conversation, with speaker labels, searchable.
- **Document research** — court filings, FOIA returns, leaked memos, public reports, regulatory filings, dropped in and converted to searchable text.
- **Story files** — one per active story, holding the working theory, the reporting log, the draft, and the running list of what's still uncertain.

A vault holds all four together. Plain markdown so the agent is fast. PDFs auto-converted to text so the docket sheet you grabbed yesterday becomes searchable today. Audio transcribed with speaker labels. Inline databases for the trackers that benefit from being tabular.

## The source file — built quietly, used constantly

The source file is the long-running asset. A page per source. Built up over years.

A typical source page holds: name, role, organization, contact details, the topics they're useful on, your interaction history (one entry per conversation with date and one-line summary), what they've said on the record versus on background, and a private notes section for your own observations — what they care about, what they're touchy about, what their reliability looks like over time.

For a recurring source you've talked to twenty times, the page becomes a real working document. Before a new call, ask the agent: "Pull the last three conversations I had with this source. Summarize what they've said about the regulatory beat across those calls, and flag any contradictions." You walk into the call having actually re-read the relevant context, not having reconstructed it from memory.

For sources you talk to once a year on a specific topic, the page is a way to not start from scratch. The bio you wrote the first time you called them, the topic you talked about, the way they wanted to be quoted — all there. The next call doesn't begin with you fumbling for context.

The on-record / on-background / off-record distinctions matter and need to live in the source page itself, not in your head. Tag each interaction with what was agreed. The agent respects the tags when it pulls quotes for a draft. If a source said something on background, the draft shouldn't quote it on the record — and the tagging in the source page is what enforces that.

## Interview transcripts that survive the deadline

Recording interviews is standard practice when sources agree. The transcription pipeline is what makes the recording useful three weeks later instead of a folder of audio files you'll never relisten to.

Drop the recording onto the relevant story page or source page. It transcribes with speaker labels — so you know who said what across a four-person background meeting. The transcript lives in the page, searchable, alongside your own notes from the conversation.

Now under deadline, when you remember someone said something useful but you can't find which interview, you ask the agent: "Find every time anyone in my reporting on the FERC story said anything about the timeline for the rule's implementation. Pull the relevant exchanges with who said what." You get the moments back, with attribution, in seconds. The hours of relistening compress.

For long interviews where the news comes in the middle, the transcript is the only practical way to find the news again. Forty-five minutes of conversation, three minutes of which contained the lede — the transcript lets you find the three minutes without scrubbing through the forty-five.

A discipline that pays off: after each interview, type three lines about what surprised you, what felt off, and what you should follow up on. Those three lines are the most useful annotation on the transcript when you come back to it under deadline. The same capture-after-meeting habit shows up in [how to capture action items so they actually get done](/guides/meetings-people/action-items-actually-get-done/) — different output, same shape.

## Document research and the FOIA pile

A serious story usually involves documents. Court filings, agency records, regulatory submissions, leaked internal memos, the corporate filings you grabbed from EDGAR last week. The pile grows fast and stops being usable around document twenty unless you have a system.

A "Documents" sub-vault per story, with one page per document. The PDF goes onto the page; it auto-converts to markdown via docstrange, so the document becomes searchable text. Add a one-paragraph summary at the top — what the document is, where you got it, what's significant in it.

When you're writing the draft and need to cite the third footnote of the agency's response, you don't dig through forty PDFs. You ask the agent: "In the agency response document, find the section that addresses cost projections. Pull the relevant paragraphs and the page number." You get the citation back, in plain text, ready to verify against the original.

For document-heavy investigations, a "Documents" inline database via the `:::database:::` directive is useful for the working catalog. Columns for source, date received, document type, status (read / partially read / unread), and significance. The whole document trail becomes a single searchable list.

The PDF→markdown conversion matters most when documents are scanned or formatted weirdly. Court filings with redactions. Agency reports with two-column layout. Regulatory filings with embedded tables. The conversion makes them searchable — you stop relying on Cmd+F on a PDF that doesn't have a real text layer.

## Story files — the working surface

One page per story. The page holds the working theory of the story (what you think it's about, which can shift as reporting develops), the reporting log (a dated list of who you've talked to, what documents you've reviewed, what's still outstanding), the draft, and a "what's still uncertain" list at the bottom.

The reporting log is the part that prevents the most regret. Three weeks into a story, the log tells you who you've already called, what they said, and which leads you said you'd follow up on but haven't. Without it, you re-call sources you already talked to and forget the lead the press secretary dropped on background two weeks ago.

The draft section is where the writing happens. The agent can help with the structural work that's hard to do staring at the page: "Read the reporting log and the relevant interview transcripts. Draft a 600-word version of the story focused on the regulatory implications angle. Quote the FERC source on background only as an attributed paraphrase, never on the record." You get a draft to react to. You rewrite most of it, but the structural assembly took seconds instead of an hour.

The "what's still uncertain" list is the discipline that keeps the story honest. Every claim in the draft that isn't fully sourced gets a line in the list. Before filing, the list goes to zero — either by sourcing the claim or by removing it. The agent can help: "Read the draft and the reporting file. List every claim in the draft that isn't supported by an interview or a document on file." That's a real check before filing.

## The running beat and the deadline draft

For reporters covering a beat over years, the vault is more than the sum of stories. It's the running file on the institution. Every committee hearing, every annual report, every leadership change — captured incrementally, building up the contextual background that makes the next story faster. A beat sub-vault holds pages for each major institution, the recurring topics, the legislative tracker, the regulatory calendar. When a new story lands, you start from a vault that already knows the players.

The hardest writing is the deadline draft. The agent compresses the assembly part of the work, not the judgment. "From the reporting file, pull the three quotes that most directly contradict the agency's official position. Pull the document section that supports each. Suggest a structural outline for a 900-word piece that leads with the contradiction." You get something to work against. The blank-page problem is gone. For drafting workflows broadly, see [drafting emails, proposals, and newsletters inside your notes app](/guides/creatives-content/draft-emails-proposals-in-notes/) — the deadline-news-draft case is a faster, higher-stakes variant.

A practical note: nothing the agent says about a document or a quote is automatically true. The agent finds the quote, summarizes the section, surfaces the contradiction — but the verification, the actual reading of source material in context, is still the reporter's job. Every claim in a draft needs a citation in the reporting file that an editor could check. The agent helps surface the citation; the reporter confirms it's real. The capture-after-meeting habit is the same shape as in [how to capture action items so they actually get done](/guides/meetings-people/action-items-actually-get-done/) — different output, same discipline.

## A calmer way to work the beat

Beat reporting is going to be intense whether your notes are good or not. But the cognitive load between stories — remembering who said what to which colleague three weeks ago, finding the document that has the answer to today's question, keeping the source file current — that part is fixable. The vault holds the running record. The agent does the searches. The reporter does the reporting.

Try Docapybara free — [sign up](/accounts/signup/), drop in the transcripts of your last two interviews on an active story, the documents you've gathered, and the working draft, then ask the agent for a list of every claim in the draft that isn't yet sourced.