Part of Deep Processing
Deep Processing for Research
Research quality lives in the mental schema that can explain why each source matters, where it fits, what it fails to see, and what new knowledge could be created from the gap. Everything downstream — writing, methodology, argument — is an artifact of that internal organization. The field must be made intelligible inside your own head before any output becomes useful.
The analogy that captures it: each paper is a jigsaw piece. The useful move is to place it — where it belongs in the wider picture, what adjacent pieces it touches, what it leaves empty, and whether it distorts the section around it. Collecting pieces without building the picture is not research — it is filing.
Research Raises the Thinking Floor
Ordinary academic learning survives at the understanding and recall level. Research does not allow that floor. Each source has to be processed against the whole:
- Understand what the source claims.
- Locate it in the wider field — what it supports, challenges, or ignores.
- Compare it against other views and methods.
- Detect what it fails to see: its assumptions, measurement limits, framing choices.
- Evaluate its strength, generalizability, and relevance.
- Create something useful from what the field collectively reveals.
The temptation is to stay at the understand-and-summarize level because it feels like progress. In practice, more reading without active placement produces more confusion, not more clarity. Coverage without structure just adds to the pile.
Overwhelm Is the Raw Material
The feeling of being buried under sources often marks the first serious contact with a field — before expertise has organized it. This is not a signal of failure. It is the signal that the real work has begun.
The confusion has a characteristic shape:
- Many papers → many perspectives → contradictory claims → unclear gaps → cognitive overload → temptation to summarize and file.
The repair is not more reading. It is active organization:
- Stay with the overload instead of escaping it through passive re-reading.
- Build provisional groups — even rough, imperfect clusters.
- Compare claims directly: which authors agree, which disagree, and on what specifically.
- Name the disagreements rather than filing them as “different perspectives.”
- Identify what is missing — what the existing conversation has not thought to address.
- Explain the field simply — to a colleague, in eight minutes, without notes.
Expertise appears when the same mass of information starts feeling navigable. Author names, dates, and findings become easier to hold because they have places to sit in the structure you have built.
What Output Reveals
Weak writing almost always reflects weak internal organization rather than a writing problem. If the literature review has no narrative, if the discussion is hard to write, if the presentation cannot be compressed — the problem is upstream. Diagnose by type:
- Ideas are disordered → the higher-order schema is weak; re-organize the map before writing more.
- Examples are missing or vague → the lower-order detail is thin; go back to sources for concrete instances.
- Expression is clumsy despite clear thinking → the procedural writing skill needs practice, not more reading.
- Methodology keeps shifting → the research question may have been narrowed before the field was understood.
- Every new paper creates confusion → the current map cannot absorb new evidence; rebuild the structure before adding sources.
The repair often comes from reorganizing what has already been read, not from adding more to the pile.
Learn the Field Before Narrowing
Early narrowing feels efficient but creates drag. A specific question before a broad schema forces every source through a small frame — the learner then fights the field instead of understanding it. A stronger sequence:
- Broad orientation first. Systematic reviews, major competing perspectives, prominent authors.
- Contrary views. Which minority positions have not been absorbed into consensus, and why?
- Recurring methods and gaps. What does the field keep measuring? What does it keep not measuring?
- Provisional map. A rough structure you can draw, explain, and revise.
- Focused question. Narrow only after the field feels navigable.
- Study design or argument. Built on the map, not before it.
Once the field is intelligible, a narrow question becomes productive. Before it is intelligible, a narrow question just reduces the surface area of your confusion.
Evidence Still Has Limits
The hierarchy of evidence is useful for orientation — it is not a substitute for thinking. Systematic reviews and meta-analyses aggregate a field’s findings, but they also inherit its blind spots: the reviewers’ framing, the available measures, the populations studied, and what prior researchers thought to study.
A practical stance for evaluating any source:
- What was not measured? The absence of evidence is not evidence of absence.
- What assumptions made the research easier to run but less ecologically valid?
- Where do population differences, time costs, or transfer constraints limit the findings?
- Which authors are shaping the framing of the entire conversation? Following their citations often matters more than following citation counts.
The move from “what does the literature say” to “what can this literature see and not see” is when research thinking begins.
AI at the Boundary
AI is most useful in research at the edge of your existing model — after you have built the structure yourself. It can widen the searchlight. It should not carry the map.
Better uses:
- “Here is my current map. What perspectives might be missing?” — Gap-checking after your own model exists.
- “What keywords should I search to investigate this gap?” — Orientation for the next search pass.
- “What contrary positions exist around this claim?” — Stress-testing a provisional conclusion.
- “Find possible weaknesses in this explanation.” — Adversarial pressure on your own reasoning.
- “What prominent authors or concepts should I look for?” — Accelerating source discovery.
Poor uses:
- “Summarize this field for me.” — Outsources the jigsaw-placing work entirely.
- “Write my literature review.” — Produces the artifact without the schema that makes it usable.
- “Tell me what the consensus is.” — AI over-indexes on high-citation, mainstream material and cannot surface niche, recent, or low-citation work that may be most relevant.
- “Explain the best theory.” — Structures your understanding before you have done the structural work yourself.
For source discovery specifically, domain-specific tools (such as Research Rabbit) outperform AI because they map your existing papers and surface adjacent work algorithmically. AI hallucinates citations on niche queries and cannot replace specialized tools for this task.
Storing References While Encoding
Non-linear encoding creates a challenge for reference tracking: the structure being built does not preserve source attribution in the way linear notes do. Three strategies address this without disrupting the encoding process.
Synthesise as you go. After every two or three sources, write a short linear synthesis with citations — essentially a draft literature review growing alongside the non-linear map. This produces two parallel outputs: a non-linear structure for understanding and evaluation, and a cited linear document that accumulates as a reference bank. Best suited when a single piece of writing is the target, since the linear document is being built toward one output.
Second brain. Use a reference application (Obsidian, Roam, Zotero with notes) to create a page per source with a brief summary and carefully chosen tags. Tag networks and graph views then surface relationships between sources that would otherwise require memory. Higher setup cost, but better suited for long-term, multi-project reference use where the same sources will be needed repeatedly.
Reference chunking. Categorise each source by its importance and role rather than by topic or date. Useful categories include core foundational works, high-leverage applied studies, contextual but peripheral works, counterpoint or minority-view references, and methods or measurement references. Store these in citation software (Zotero, EndNote) with tags and brief notes explaining why each source belongs in its category.
Reference chunking requires genuine field expertise to execute accurately — the ability to judge where a paper sits in the landscape depends on already having a mental map of the landscape. It is most effective once a domain schema is established. When it works, it makes pulling high-quality citations for a paper much faster, because the structure of the literature is reflected in the reference collection itself.
The choice between methods depends on whether the references will be used once (synthesise as you go), repeatedly across projects (second brain), or in a field where deep expertise is being developed over years (reference chunking). Combinations are possible and often appropriate.
What It Should Feel Like
Good research processing feels like increasing command over a messy territory. The confusion changes character — not less work, but better-organized difficulty.
Good signs:
- Papers start falling into recognizable camps without forcing them.
- Author names and dates become memorable without brute-force repetition.
- Gaps become more specific — you can say exactly what is missing rather than “this is complicated.”
- Writing starts having a natural order; the argument builds without fighting each sentence.
- Questions become sharper; the research question narrows because the field demands it, not because you ran out of patience.
- The field feels arguable — you have a position — not just readable.
Warning signs:
- Every paper becomes a standalone summary filed without placement.
- Notes grow while the map stays vague.
- AI summaries feel clearer than your own understanding.
- The research question narrowed before the field was intelligible.
- More reading produces more confusion without new structure.
- Writing begins before the mental schema can explain itself without notes.
Open Questions
- What would it look like to treat your current literature pile as a jigsaw-placement problem rather than a reading queue? Which pieces do you actually know where to place?
- Where in your research process are you still working at the “understand and file” level when the task actually demands comparison, gap detection, or evaluation?
- How do you distinguish between productive overwhelm (first contact with an unorganized field) and unproductive overwhelm (no placement strategy, just accumulation)?
- Which part of your current literature map is still just a pile? What would it take to build provisional structure there?
- Where are you reaching for AI because you do not want to do the hard synthesis yourself — and what does that tell you about where the real gap is?