The academic citation manager exists to solve a specific problem: formatting bibliography entries and deduplicating sources from known databases. Zotero, Mendeley, and their peers do this reliably. Feed them a DOI or a PubMed ID and they return clean metadata, populate author fields, format entries across citation styles. For a researcher working almost entirely within academic literature, these tools are genuinely useful.
The problem is that most serious research does not look like a PhD dissertation. A policy analyst working on a six-month infrastructure report is pulling from peer-reviewed papers, yes, but also from city council meeting transcripts, regulatory filings, a spreadsheet of notes from phone calls with officials, archived news coverage from the 2000s, and pricing pages from three competing vendors. A journalist investigating a supply chain story works from corporate filings, exported emails, source interview notes, court records, and documents that have no DOI and no stable identifier. A consultant synthesizing a market landscape has analyst reports, internal decks, and a rotating collection of tabs in varying formats.
Citation managers handle maybe a third of that. The rest ends up distributed across whatever capture tools were convenient at the moment: bookmarks, a Notion page that was supposed to be temporary, a Downloads folder, a Google Doc named "misc research," and a desktop folder with 47 files that has not been opened in three weeks.
What Citation Managers Actually Solve
It helps to be precise about what academic citation tools are built for, because conflating them with general research tracking leads to bad tool choices.
They solve: bibliographic metadata capture for sources with stable identifiers, deduplication across known databases, citation style formatting, and PDF attachment with basic annotation. They assume you are working within a corpus of sources that have known provenance and machine-readable IDs. They treat sources as independent entries in a flat list. The relationship between sources, whether they support the same claim, contradict each other, or address different parts of your argument, is not tracked. The context of why you saved a source is typically stored in a notes field that is separate from both the source and your draft.
For research outside the academic corpus model, they are essentially expensive bookmarks with better metadata for the sources they happen to recognize and incomplete handling for everything else.
Where the Failure Happens
The failure does not occur at the moment of saving. It happens three weeks later, when you need to find something specific. You read something about a regulatory proceeding, you are reasonably certain it was a PDF, you have a vague memory of where you found it. You spend 40 minutes searching. You either find it after a significant time cost, or you do not find it and either drop the reference or re-research the point from scratch.
This is not a memory failure. Your memory of reading the source is intact. What failed is structural: the system you used to save the source did not capture the context that would make it retrievable later. The context that goes missing is threefold. First, where in your argument you thought this source would be useful. Second, what specific claim or data point prompted you to save it. Third, how it related to other sources you were already tracking.
The gap between saving and contextualizing is where the problem lives. You bookmark something in a moment of forward momentum, intending to annotate it properly later. Later is precisely when context has already degraded. The annotation you planned to write during a dedicated pass does not get written, and the intent that seemed obvious in the moment of capture becomes invisible two months into a project.
What Long-Form Source Tracking Actually Requires
After building Pilgrim and working closely with analysts, journalists, and policy researchers, we have found three structural requirements that distinguish effective source tracking from citation management.
The first is argument-linked capture. When you save a source, you should be anchoring it to a question in your developing argument, not dropping it into a flat list. Sources stored as independent objects require you to reconstruct their relevance every time you return to them. Sources tied to specific questions can be retrieved by those questions, and you can see at any point which parts of your argument are well-sourced and which are thin.
The second is relational visibility. When Source A makes a claim and Source B contradicts it, you need to see that tension while you are still building your source base. Most tools store sources independently: they give you no mechanism to surface contradictions except through manual comparison, which happens inconsistently under deadline pressure and often not at all.
The third is durable context. The note you write at the moment you encounter a source carries information the source itself does not: why you read it, what claim it was going to support, what surprised you, what it changed in your thinking. That note is worth more than the citation when you are drafting two months later. Most citation tools treat it as a secondary field that does not travel with the source when you move into writing.
The Scale Problem
At ten sources, none of this matters much. You can hold the relationships in your head. At thirty, the relational structure starts to slip. At sixty, you are working from a system that was not designed for sixty sources, and you feel the cost as a constant low-grade tax: time spent re-locating things you already read, re-reading passages to remember why they mattered, reconstructing comparisons you made informally in your head weeks ago.
Researchers who spend time on long-form projects often describe a specific feeling: losing access to their own thinking as a project progresses. Notes become cryptic. Sources that seemed obviously relevant when saved look unclear in retrospect. The early insight that shaped the project direction is buried somewhere that cannot be located. This is not a discipline problem. It is what happens when project scale exceeds the design assumptions of the tools.
The Case for Light Capture
There is a real argument that capturing too much context during research creates its own overhead. Some researchers prefer the lightweight inbox model: save first, annotate later, let the real synthesis happen at the writing stage rather than during reading. This has genuine advantages. When you are in a reading flow, stopping to write two sentences about a source's relevance to your argument can break the thread of attention. Light capture respects the momentum of active reading.
We are not arguing this approach is wrong. It works well for shorter projects where the volume of sources stays low enough to manage in memory, or for researchers disciplined enough to run a dedicated annotation pass before drafting. The problem is that most long-form research does not have a clean annotation pass built into its schedule. The pass you intended to run gets skipped. Two months later, you have a citation and no memory of why you added it. Light capture assumes a follow-up step that frequently does not happen.
What We Built Pilgrim For
We built Pilgrim with the source-to-argument chain as the core data structure, not as an optional layer on top of a flat list. When you add a source to a Pilgrim project, you anchor it to the questions your project is investigating. The system tracks which sources address which parts of your developing argument. You can see which positions are well-supported and which need more sourcing, and you can retrieve sources by the question they address rather than by metadata that was incomplete when you saved them.
The context note lives alongside the source, not in a separate field. When you pull a source into a draft section, the note travels with it. The intent you captured at the moment of reading remains attached to the object through the drafting stage.
If you are writing a dissertation and working almost entirely in peer-reviewed literature, a dedicated citation manager may serve your bibliography formatting needs better. What Pilgrim was built for is the analyst, journalist, or policy researcher working across heterogeneous source types over months, where the failure mode is not citation formatting but context loss. That is a different problem, and it requires a tool designed with that failure mode as the central concern.