# 5 Common Mistakes When Linking Research Notes and How to Fix Them Most linking research notes mistakes share one root cause: the link gets clicked before the reason for the link gets written. A link is a pointer. The reason two notes belong together exists in your working memory at the moment you connect them, and that memory fades within days. Here is the first thing to check. Open any note you wrote more than two weeks ago, follow one of its links, and try to state in a single sentence why that note points to the other. If you cannot, you have found the most common failure, and the four mistakes that follow are variations on it. This article covers the five mistakes in order of how often they occur, explains the mechanism behind each failure, and gives a troubleshooting table for quick reference. It closes with a 15-minute diagnostic and the prevention practices that keep a [literature notes system](/blog/research-skills/the-ultimate-guide-to-building-a-literature-notes-system-for-research) intact for years. The solutions are writing rules, not new software, and each one takes seconds to apply. ## Why Links Fail Before Anything Else Does Research note taking is the practice of capturing what you read and what you conclude while a project is live. Three note types matter here. A literature note records what one specific source claimed. An idea note records what you concluded. An index note exists to list and annotate links on one topic. A link is an explicit connection between two notes. A backlink is the reverse pointer that most modern apps maintain automatically. The graph is the whole network of notes and links. A note system breaks at the links before it breaks anywhere else, for a simple reason. Writing a note is visible work, so you do it. Creating a relationship is one click, so it is easy to do without thinking, and an unthinking relationship carries no information. The link says the notes are connected. It does not say how, why, in which direction, or with what strength of evidence. The causes behind linking research notes mistakes are few, and they repeat at every scale. Large knowledge projects hit the same wall. The ABCkb knowledgebase, which connects plant science to human health outcomes, was created because results in those domains were sparse, siloed, and difficult to integrate, and its core design choice was to store every connection between concepts as an explicit relationship rather than an implied one. One fact about readers should guide every rule below. A qualitative study of patients reading their own medical visit notes identified four themes in what readers valued about note content: confirm and remember next steps, quicker access and results, positive emotions, and sharing information with care partners. Future you is that reader. Linked notes succeed when they confirm next steps and speed up access, not when they maximize connection count. ## Mistake 1: The Bare Link (Most Common) A bare link is a pointer with no sentence explaining it. It shows up in two places, link lists at the bottom of notes, and concepts dropped into sentences with no stated relationship. Bare: "The mechanism involves [[reactive oxygen species]] and [[Nrf2 signaling]]. See also [[gut barrier function]]." Linked with context: "A high-fat diet increases intestinal reactive oxygen species (molecules with unpaired electrons that damage tissue, see [[reactive oxygen species]]), which is expected to activate [[Nrf2 signaling]], the [transcription factor](/knowledge/molecular-biology/transcription-factor) that switches on antioxidant defenses, a mechanism proposed to explain the barrier changes in [[gut barrier function]]." The second version takes about thirty seconds longer to write. It is the only version that still makes sense in six months. **The mechanism behind the failure.** A link stores an address, not a reason. At the moment of linking, the reason exists only in working memory, which holds it for seconds to minutes. Once the reason is gone, the pointer is unreadable in both directions. You cannot reconstruct why note A pointed to note B, and a collaborator reading your notes cannot either. The graph gains an edge and gains nothing else, because the edge has no meaning. This is the most common of the five mistakes because the cost asymmetry favors it. The click is free. The sentence costs thirty seconds. Under deadline pressure the click wins every time, and the debt is collected later, during re-reading. **The fix: the linking sentence rule.** Never place a link without a sentence that states the relationship and its direction. The sentence answers three questions. What does note A claim? What does note B claim? How does one bear on the other? If you cannot write the sentence, you have not earned the link. Delete it, or park the raw connection in a questions note with an honest line such as, "Possible link between [[Nrf2 signaling]] and [[gut barrier function]], reason unclear, check glutathione literature." A parked link with a label beats a confident link with no content. A side effect of the rule is that your notes remain readable as prose, and prose survives app migrations, exports, and format changes. Link lists do not. ## Mistake 2: Notes With No Source Anchor A literature note is a claim about what a source said. When the note carries no citation, it stops being a claim about a source and becomes an assertion of unknown origin. Researchers describe the slow spread of such assertions as citation rot. **What it looks like.** You are drafting a manuscript paragraph, you open the linked note that supports your sentence, and the note says only, "Secondary bile acids suppress GLP-1 secretion." GLP-1 is a gut hormone that signals fullness. The note shows no authors, no year, no DOI (the digital object identifier that retrieves the exact paper), no page. You now choose between hunting the source from scratch and writing a sentence you cannot defend. **The mechanism behind the failure.** Notes get paraphrased and linked onward. Each hop away from the original source adds drift, because paraphrase smooths detail and drops qualifiers. After three hops, a hedged finding, "reduced secretion in one mouse strain," has become a flat claim, "suppresses secretion." The links still work. The content behind them no longer points to anything checkable. **The fix: source-first creation.** A literature note begins with metadata, never with the summary. A workable header holds the full citation, the DOI or PubMed ID (the numeric identifier that retrieves the paper's database record), the date you read it, and one sentence on what the paper claims. The body lists claims with the page, figure, or table where each appears, and quotes numbers exactly rather than rounding them. Your own interpretation goes in a separated section labeled as yours. Curated knowledgebases apply the same discipline under heavier load. ABCkb integrates 11 curated sources together with relationships mined from Medline abstracts (the indexed biomedical literature), and every connection in its graph traces back to those inputs. [1] That provenance is what makes the graph trustworthy enough to generate hypotheses from. Your vault is a small version of the same problem, and the rule holds at every scale. **The boundary rule.** Keep literature notes and idea notes visually distinct. When an idea note depends on three literature notes, it links to them and says so in a sentence: "This hypothesis combines the bile acid finding in [[secondary bile acids, liver signaling]] with the secretion data in [[GLP-1 secretion, murine]]." If every note is a literature note, you have no place to record your own thinking. If every note is an idea note, nothing in the vault can be verified. ## Mistake 3: Over-Linking and Tag Soup **What it looks like.** Every noun becomes a link. Notes carry twenty or thirty tags. The graph view looks impressive and retrieves nothing. Searches return dozens of hits, and browsing keeps landing on vague hub notes such as "Methods" or "Inflammation" that link everywhere and say nothing. **The mechanism behind the failure.** A link is a signal, and a signal works through contrast. When every note links to every other note, the presence of a link carries no information, the way a highlighter applied to every line of a page highlights nothing. Retrieval by browsing follows the notes with the most links, and those notes are usually the vaguest, because broad topics attract connections from every direction. Each excess link also carries a maintenance cost, since every link you create is a link you may someday have to prune, rename, or re-explain. **The fix: claim-level linking.** Link only when a specific claim in note A depends on, supports, or contradicts a specific claim in note B. A mention is not a claim. "This paper mentions bile acids" is not a reason to link. "This paper's dose-response curve supports the threshold assumed in my dose note" is. Apply the same standard to tags. A tag is a retrieval category, useful only if you actually search it. Keep three to five tags per note, drawn from a short list you control, and delete any tag you have never searched. Replace tag piles with curated index notes. A good index note holds five to fifteen links, each with a one-line description of what the linked note claims and why it matters. Ten curated index notes will retrieve more, faster, than a thousand raw tags, because they store judgment rather than category labels. The test for any link or tag is the same. Ask what breaks if you delete it. If the answer is nothing, delete it. ## Mistake 4: Ambiguous Note Identity and Broken Links **What it looks like.** Two notes titled "Inflammation." A file renamed in the operating system's file manager, leaving links that point to nothing. A search that returns near-duplicates, so you cite the thin note instead of the good one. Over months, the graph splits into islands that never rejoin, because nobody knows the duplicates exist. **The mechanism behind the failure.** A link is an address, and addresses fail in two ways. They collide, when two notes share a title and a link resolves to the wrong one. They break, when a note is renamed or moved and the address is not updated. Both failures share one root, titles written for the moment rather than for retrieval. "Bile acids" is a fine title on the night you read the paper. Six months later you cannot remember whether that note covers primary or secondary bile acids, receptor activity, or liver-to-gut recycling, so you create a second note, and the vault now disagrees with itself. **The fix: unique, descriptive, stable titles.** Title every note with the concept plus the qualifier that makes it distinct, for example, "Secondary bile acids as FXR agonists" rather than "Bile acids." Before creating any note, search for the concept. If a note exists, link to it and improve it instead of duplicating it. Rename only with the app's own rename command, which updates links automatically, never with the file manager. Where the app supports aliases, add the shorter name as an alias so old links and old search habits still resolve. Run a monthly orphan check. Filter for notes with zero inbound links. An orphan is either a note the system has abandoned or a duplicate you already own. Merge it, link it, or delete it. **A short safety note.** Before any bulk rename, merge, or scripted rewrite of links, export a full backup of the vault and store it outside the working folder. A faulty find-and-replace across hundreds of files can corrupt every link in a vault in a single action, and undo tools rarely reach across that many files cleanly. ## Mistake 5: Untyped Links Treated as Evidence (Rarest, Most Damaging) **What it looks like.** A chain of links reads like an argument. Note A links to B, B links to C, and months later you present the chain in a talk or a draft as though it were a demonstrated pathway. On inspection, the first hop was a replicated finding, the second was one study in one cell line, and the third was a hunch you never checked. The chain was never evidence. It was a wish with links attached. **The mechanism behind the failure.** An untyped link defaults to the weakest possible relationship, "related." Relatedness is not mechanism, and a path through a note graph is not a pathway in the biological sense. Confidence does not accumulate across hops. It decays, because each hop multiplies the uncertainty of the one before. Three linked maybes do not add up to a maybe. They multiply into a maybe of a maybe of a maybe. The graph view makes this worse, because a connected path looks like support. **The fix: type your links and label your confidence.** Give links a relationship drawn from a short vocabulary you will actually use. Five types cover most research work: supports, contradicts, proposes mechanism for, method from, and question raised by. Add a confidence label to any claim not anchored to a source: sourced (a specific paper says this, cite it), inferred (you reasoned to this from two or more sourced notes), speculative (a hypothesis with no anchored support yet). When a claim depends on a chain, write the chain out with the relationship at each step: "High-fat feeding raises secondary bile acids (sourced, see [[secondary bile acids, liver signaling]]). Secondary bile acids activate FXR, the bile acid-sensing nuclear receptor (sourced, same note). FXR activation suppresses GLP-1 release (inferred from two cell studies, unconfirmed in vivo). Suppressed GLP-1 would reduce satiety after high-fat meals (speculative, testable)." That sentence is longer than a link chain, and it is the only form that survives contact with a reviewer, because every hop shows its evidence class. ABCkb demonstrates the same discipline at database scale. It connects plants to human health through a stepwise path from plant to plant product to human gene to pathways to indication, with each step an explicit relationship stored in a graph database, a store built to hold nodes and their relationships as first-class data. [1] The stepwise typing is what lets the resource generate hypotheses without presenting them as findings. Your vault needs the same rule at a smaller scale. ## Troubleshooting Table: Symptom, Cause, Confirmation, Fix Scan this table when a specific symptom appears. The rows run from the most common cause to the rarest, matching the sections above. | Symptom | Likely cause | How to confirm | Fix | |--|--|--|--| | You open a linked note and cannot recall why the two connect | Bare link with no context | Read the sentence around the link. If no sentence states the relationship, confirmed | Write a linking sentence at the link site, delete links you still cannot explain | | You cannot verify a claim while drafting | Literature note lacks a source anchor | Check the note header for a DOI, PubMed ID, or page number | Re-find the source, add full citation metadata, mark unanchored claims as unverified | | Browsing keeps landing on vague hub notes | Over-linking and tag soup | Count outbound links on your ten most-linked notes, then ask whether any link cites a specific claim | Prune to claim-level links, replace tag piles with three to five curated index notes | | Search returns near-duplicate notes on one concept | Ambiguous note identity | Search your five most-used concepts and count near-duplicate titles | Merge duplicates, adopt unique descriptive titles, add aliases for old names | | Links point to nothing after a rename | Address broken by a manual rename | Run the app's broken-link or orphan report if it has one | Restore from backup, then rename only with the app's own command | | A linked chain reads like a proven mechanism | Untyped links treated as evidence | Trace the chain and try to label each hop as sourced, inferred, or speculative | Add relationship types, downgrade unsupported hops, re-anchor each hop to its source | | Notes pile up that nothing ever links to | No index or backlink review | Filter for notes with zero inbound links | Build or update an index note that places each orphan, or delete it | ## How to Fix Linking Research Notes Mistakes: A 15-Minute Diagnostic ![Decision path asking whether a link has a sentence, a source anchor, and a typed relationship, ending in a fix for each failure.](/article-images/illustrations/linking-research-notes-mistakes-link-triage.webp) The fixes above work best in a sequence. This one finds your dominant failure in about fifteen minutes, using a sample rather than a full audit. 1. Open ten notes at random. Most note apps have a random-note command. If yours does not, pick ten from different months. 2. Check each note against the five failure points in order. Does every link sit next to a sentence stating the relationship? Does the note carry a source anchor, or an explicit label marking it as your own idea? Are the outbound links claim-level rather than mention-level? Is the title unique and descriptive enough to find from memory? Where the note makes a multi-hop claim, are the hops labeled sourced, inferred, or speculative? 3. Score each note from 0 to 5, one point per check passed. 4. Average the scores. Vaults that have grown without rules usually land between 1 and 3, which maps to mistakes one through three. 5. Identify your dominant mistake, the check that failed most often across the ten notes, and fix that one pattern across the sample. Do not fix everything at once. 6. Re-run the sequence monthly. A rising average is the signal that the rules have become habits. Two cautions. Ten notes is a sample, so treat the result as a direction, not a verdict on the whole vault. And if your dominant failure is mistake four, back up the vault before you start merging anything, for the reason given in the safety note above. ## Prevention: Keeping a Literature Notes System Linked ![Five parallel habits: linking sentence first, source-first notes, claim-level links, stable titles and typed links with confidence labels.](/article-images/illustrations/linking-research-notes-mistakes-prevention-habits.webp) The five mistakes have five matching habits, plus two structural ones. Together they cost minutes per day. **Write the linking sentence first.** When you feel the urge to link two notes, type the sentence before you create the link. If the sentence will not come, the connection is not ready. Park it in a questions note. **Create notes source-first.** Every literature note opens with its citation, DOI or PubMed ID, and date read, and every number is quoted exactly with its page or figure. Idea notes are labeled as idea notes. This one habit prevents citation rot entirely, because drift cannot start from an anchored note. **Link at the claim level and keep tags few.** Link when one note's specific claim depends on another's. Keep three to five tags from a list you control, and curate index notes instead of accumulating categories. **Keep identities unique and stable.** Descriptive titles, search before create, app-native renames, aliases for old names, and a monthly orphan check. **Type links and label confidence.** The five-relationship vocabulary and the three confidence labels convert your graph from a hairball into an argument you can defend. **Build in a feedback channel.** Notes improve when someone reacts to them. In a pilot that attached a simple feedback tool to patients' visit notes, 98.5 percent of the 260 reports submitted by patients and care partners rated the tool as valuable, and accuracy was among the themes readers raised. Your version is smaller and cheaper. Flag notes that need verification, and once a quarter, re-read the linking sentences in your ten most-linked notes, deleting or downgrading the ones you can no longer support. **Keep the infrastructure boring.** Plain-text files in a folder you back up, stable titles, and a reference manager for citations will outlast any app trend. ABCkb ships as a queryable Docker container (a packaged, runnable copy of the software and its environment), and all of its build scripts are available on GitHub, a reminder that even a large curated graph depends on reproducible, unglamorous infrastructure. Your vault is the same kind of object at a smaller scale, and it deserves the same standard, portable formats, scripted backups, and no single point of silent failure. Test any link-rewriting script on a copy of the vault before running it on the live folder. ## Common Mistakes and Limitations A few failure modes sit outside the five, along with honest limits on what a linking system can do. Do not rebuild the archive in one weekend. Migrate forward. Apply the rules to every note you touch from today, and repair old notes when a project brings you back to them. A full retrofit of a thousand notes produces a thousand mechanical edits and no thinking, and it is the fastest way to abandon the system. Do not import someone else's relationship vocabulary wholesale. If a published system uses fifteen link types and you would use three, adopt three. A vocabulary you will not use is tag soup with better branding. A linked vault does not replace a reference manager or a reference list. The vault organizes your thinking about sources. The reference manager formats them. Manuscripts still need formal citations, and the vault's job is to make those citations fast to assemble, not to replace them. Linking is not analysis. A beautifully linked vault of wrong claims is efficiently wrong. The confidence labels limit the damage, but they do not remove it, because labels are only as honest as the person writing them. Finally, no linking system substitutes for reading the primary sources yourself, and high-stakes outputs such as theses, manuscripts, and grant applications deserve a second reader who will trace your evidence chains and challenge each hop. ## Frequently Asked Questions ### What is the most common mistake when linking research notes? The most common mistake is the bare link, a pointer created without a sentence explaining the relationship. Apply the linking sentence rule: no link without a sentence stating what each note claims and how one bears on the other. ### How do I fix linking research notes mistakes in an old, messy vault? Run the ten-note diagnostic, find your most frequent failure pattern, and fix that pattern across the sample. Migrate forward instead of rebuilding, and back up the vault before any bulk merge or rename. ### What is a literature notes system? A note system in which each note records what one specific source claimed, opens with that source's full citation and identifiers, and links to other notes through claim-level relationships, with your own ideas kept in separately labeled idea notes. ### Should I link every mention of a concept? No, link only where a specific claim in one note depends on, supports, or contradicts a specific claim in another. Mentions create noise, claims create retrievable structure. ### How do I stop links from breaking when I rename notes? Rename only with the app's own rename command, which updates links automatically, and add aliases for old names. Back up the vault before any bulk rename or scripted rewrite. ### What is the difference between a tag and a link? A tag is a retrieval category you search, while a link is a claim about a relationship between two specific notes. Tags answer which bucket, links answer which relationship and why. ### How many relationship types should my links have? Fewer than ten, and five covers most research work: supports, contradicts, proposes mechanism for, method from, and question raised by. ### How often should I audit my links? Once a quarter for a full pass over your most-linked notes, plus a monthly check for orphan notes with zero inbound links. ## Sources 1. [The Aliment to Bodily Condition knowledgebase (ABCkb): a database connecting plants and human health.](https://pubmed.ncbi.nlm.nih.gov/34838100/)