Maik's Digital Garden https://profmanagement.github.io/digital-garden/en/index.html Notes on knowledge management, learning, and digital gardening en Beyond the Filing Cabinet: Do Organisation Methods Actually Solve the Retrieval Paradox? https://profmanagement.github.io/digital-garden/en/20260614_beyond-filing-cabinet.html Sun, 14 Jun 2026 00:00:00 +0000 Maik@example.com (Maik) PKM Beyond the Filing Cabinet: Do Organisation Methods Actually Solve the Retrieval Paradox?

Maik · 2026-06-14 · 60% human

Where the last post left off

In my last note about the The Retrieval Paradox, I argued that the uncertainty in my PKM and second brain does not lie in storage, meaning that the notes are still available where they should be, but in the gap between the way I encoded a note in the past and the way I try to retrieve it in the present. Tulving and Thompson’s (1973) encoding specificity principle says retrieval succeeds when present cues simply resemble the cues that were present at encoding. It often happens that, months later, my cues have changed, so the note stays available but not accessible. And I made one further claim that matters here: links solve navigation (getting from a known note to another known note), but the paradox lives one step earlier, in recognition — realising that a relevant note exists at all.

After discussion, a dear colleague suggested I have a look at the LATCH method as a possible future pathway to solve this issue. That intrigued me to do a little research on PKM organisation schemes. This post is the result, but not in the sense of “here are five ways to better organise your life,” but as a test. An organisation method only addresses my retrieval paradox if it does at least one of two things — bridge the encoding–retrieval gap, or support recognition rather than mere navigation.

  1. Encoding–retrieval match — Does the method help when my present search cue differs from the past term, tag, or context I filed under? Or does it assume I already remember the dimension I stored it on?
  2. Recognition vs. navigation — Does the method help me realise a forgotten note exists (recognition / recall)? Or does it only help me travel efficiently once I already know what I’m looking for (navigation)?

It turns out: Most of them, it turns out, do neither. Most organisation approaches are build for something different. Some are storage architectures wearing only the costume of a retrieval solution.

The comparison

Table 1

Assessing PKM Organization Methods in Light of the Retrieval Paradox

Approach Organises by The Retrieval Paradox: Why My Own Tools for Thought Increasingly Frustrate Me

Maik · 2026-06-07 · 90% human

Over the past few weeks, I’ve repeatedly grappled with a problem that increasingly angers me. It concerns precisely those tools for thought that are supposed to help me organize knowledge, make connections visible, and keep information available in the long term.

For years, I have been collecting notes, literature references, quotes, ideas, and observations in digital knowledge systems. Like many other users of Tana, Roam Research, Obsidian, or similar tools for thought, I conscientiously follow the conviction that well-networked knowledge is easier to retrieve later. So I’ve linked notes, assigned tags, created references, and tried to establish connections between atomic notes in the form of relationships.

Yet I increasingly catch myself having this frustrating thought:

I know that I noted something important about this at some point. But I have no idea anymore how I thought about it back then, under what term, with which tags, and especially in which vault I stored it. Where should I even start to search?

The problem is not that the information or idea has been lost. It’s already located somewhere in my tool for thought. The problem is rather that I can’t search precisely because I can no longer remember the note—because it’s somehow slumbering in the sea of my second brains.

The larger my knowledge archive grows, the more frequently this uncomfortable feeling arises. And therein lies the paradox: While the volume of my stored knowledge grows, my confidence in accessing it at the right moment decreases. Often, for quickly retrievable knowledge, I unconsciously resort to other storage methods, such as browser bookmarks.

Storage Is Not Retrieval

Many personal knowledge management systems and tools for thought are based on the assumption that good links and metadata should make later retrieval easier. The underlying assumptions initially seem plausible:

  • Information is stored.
  • Information is connected with other information.
  • Information can later be retrieved through these connections.

In practice, however, this only works under one important condition: I must know what I’m searching for.

This is exactly where the problem begins.

Most notes are created in a specific situational context. I read an article, listen to a podcast, work on a publication, or engage with a concrete question. The terms I use to tag the note reflect this context.

Months later, I encounter a similar problem, but from a completely different perspective. The search terms that come to mind now are different from those I u]]> https://profmanagement.github.io/digital-garden/en/20260607_the-retrieval-paradox.html Knowledge Synthesis from PDFs: Discourse Graph & Structured Parsing with Claude Code https://profmanagement.github.io/digital-garden/en/20260517_discourse-graph-pdf-synthesis.html Sun, 17 May 2026 00:00:00 +0000 Maik@example.com (Maik) Research Knowledge Synthesis from PDFs in the LLM Wiki: Discourse Graph & Structured Parsing with Claude Code

Maik · 2026-05-17 · 95% human

The Problem: Information Overload in Scientific Papers

When building an “ingest workflow” for scientific papers (PDFs) into my LLM wiki, I realized summaries that contain everything are counterproductive. Instead, I needed to extract only genuinely relevant information. The context “clogs” processing, creating enormous noise. Many details in a paper are simply irrelevant to specific research questions, but the model doesn’t know that on its own.

In the analogue world, the solution is straightforward: linear reading, highlighting, extracting. Time-consuming. Error-prone. Not scalable.

But there is an approach: extracting information through structured parsing using the Discourse Graph methodology.

The Discourse Graph: A Method for Structured Knowledge Synthesis

The Discourse Graph is not merely an ontological knowledge concept—it’s a practical method for systematically extracting and linking knowledge from unstructured sources. The idea: decompose every text into granular, interconnected units rather than treating it as a monolithic block.

I applied the five core components of the Discourse Graph to my wiki as follows:

1. Sources

The primary document itself. Who wrote it? When? In what context? This anchors all downstream information.

2. Evidence

All usable information from the source:

  • Methodological details (research design, sample size, instruments)

  • Theoretical foundations (concepts, frameworks)

  • Empirical findings (results, metrics, observations)

Important: Evidence is facts or observations, not yet interpretations.

3. Claims (Assertions/Findings)

The conclusions authors draw from evidence: “If evidence X shows, then statement Y follows.” Claims are the scientific findings a paper propagates.

4. Questions

The scientific questions the text answers or raises. These emerge inductively from evidence and claims. They map the research space.

5. Concepts

Overarching ideas, theories, constructs linking multiple claims and evidence. They are the “building blocks” of my knowledge graph.

This structure is also called atomization https://profmanagement.github.io/digital-garden/en/20260517_discourse-graph-pdf-synthesis.html On the Way to an Agentic OS — First Steps with the Everything Claude Code Repo https://profmanagement.github.io/digital-garden/en/20260510_everything-claude-code.html Sun, 10 May 2026 00:00:00 +0000 Maik@example.com (Maik) AI One Step Closer to a Personal Agentic OS for Research and Teaching

Maik · 2026-05-10 · 80% human

For some time now I’ve been thinking about how to deploy an agentic OS — or rather an OS kernel for research — to handle parts of routine tasks and support creativity in research and teaching. It should be capable of executing valid workflows with defined skills, and have access to learnable libraries of skills, plugins, routines, and so on (so-called research stacks). This has some similarities with the Architecture of AIOS, an “AI Agent Operating System, which embeds large language model (LLM) into the operating system and facilitates the development and deployment of LLM-based AI Agents” (Agiresearch/Aios on GitHub).

Fig. 1. Architecture of AIOS

In the course of my research I came across Everything Claude Code (ECC), which brings a collection of 75 skills, 71 agents, 33 hooks, and countless commands to Claude Code. The repo is capable of executing professional research tasks semi-autonomously.

Fig. 2. Everything Claude Code

Plan first, code later — as a plugin

Central to the core workflow is the /everything-claude-code:plan command, which launches a specialized skill that asks clarifying questions, weighs options, and waits for explicit confirmation before a single line of code is written. This can be used both for planning a new feature in the current project and for designing a new system architecture.

The same applies to test-driven development: the /tdd workflow automatically sets up a test framework, writes tests before the implementation, and targets 80% coverage. It repeats the procedure until the desired target is reached, refining at each step.

Externalizing judgment

ECC raises an interesting question: if a tool like this makes it possible to externalize good practices for working routines, does it not replace my judgment? What is the actual added value beyond saving time?

I lean toward a nuanced answer. Tools like ECC significantly lower the barrier to structured work. Beginners get a professional scaffold without years of learning experience. Those who have been at it for a long time have to exercise less discipline to maintain familiar patterns.

What no tool can replace: knowing when to deviate from the]]> https://profmanagement.github.io/digital-garden/en/20260510_everything-claude-code.html New WikiProject for Personal Knowledge Management on Wikipedia https://profmanagement.github.io/digital-garden/en/20260502_wikipedia-project-pkm.html Sat, 02 May 2026 00:00:00 +0000 Maik@example.com (Maik) PKM New WikiProject for Personal Knowledge Management on Wikipedia

Maik · 2026-05-02 · 100% human

Wikipedia already has many articles that can be assigned to the topic of Personal Knowledge Management (PKM). Some key articles already exist — such as Personal knowledge management, Zettelkasten, Information literacy — but are often incomplete, methodologically outdated, or insufficiently sourced. By contrast, other relevant concepts, approaches, and related topics are missing entirely.

Today I created the WikiProject Personal Knowledge Management to systematically close these gaps, together with other interested contributors.

Figure 1: Screenshot New WikiProject PKM

The project coordinates work on existing and new articles in the following areas:

  • Methods and approaches — Zettelkasten, excerpting, mind mapping, Cornell method
  • Tools and software — Obsidian, Logseq, Notion, Roam Research
  • Foundational concepts — Knowledge management, information literacy, metacognition
  • Historical precursors — Commonplace book, Memex, card index

The project inventory includes core articles and so-called stubs — along with working notes on relevance and specific areas for improvement. The goal is to revise existing articles and those yet to be identified, rewrite others where needed, and translate articles from other language versions where appropriate.

The project is currently in its early stages. Anyone who researches, teaches, or is simply interested in the field and has some experience editing Wikipedia is warmly invited: contributions of any kind are welcome — a citation, a paragraph, a new article. If you are interested, feel free to reach out by email.

“The Whole Wide World” by Wreckless Eric, 1977. A song about searching — which feels right for anyone building a knowledge system. Listen on: YouTube | Spotify

Written from my desk, May 2026.