← Back to Blog

The Art of Naming Things in a Knowledge Base

A name looks like a label and behaves like an address. That distinction is the entire subject, and almost nobody notices it because naming happens in a second, at the moment of creating something, when the thing being named is fresh and obvious and the person doing it cannot imagine ever forgetting what it is. The name feels temporary. It is, in fact, the most durable decision made about that item, since the content will be revised, the structure will be rearranged, and the name will sit there unchanged for years, determining whether anybody ever reaches what is underneath it.

The reason this matters more in a searchable system than in a filing cabinet is that titles carry weight. In any ranked retrieval, a match in a title counts for considerably more than a match in a body, which means the words somebody chooses for a name are not merely descriptive: they are the terms most likely to surface the item later. Somebody who names an entry for what it is called will find it when they remember what it is called. Somebody who names it for what they will later be looking for will find it whenever they need it, which is a different and much more useful property.

The Art of Naming Things in a Knowledge Base

The mistake almost everybody makes is naming descriptively rather than for retrieval, and the mistake is invisible for months. A researcher titles a paper entry with its authors and its own title, which is how a reference is conventionally written and which contains none of the terms they will actually search. A property manager names an inspection after the date, which is the one thing they will never remember. A minister names a note after what happened, which is exactly what has faded by the time they need it. In each case the name is accurate and useless.

There is a second failure, subtler and more expensive, which is inconsistency. A convention that is wrong but applied uniformly can be corrected in an afternoon, because the items sharing it can be found and adjusted together. A collection named inconsistently cannot be corrected at all, because there is nothing to search for: the whole problem is that similar things were named in dissimilar ways, so no query gathers them. Somebody who has titled things three different ways across eighteen months has produced three partial archives inside one system, and no amount of subsequent discipline recovers the first two.

What makes this genuinely difficult is that there is no universally correct convention, and the professionals who insist otherwise are describing their own field rather than a principle. A convention has to encode whatever a person will actually remember at the moment of searching, and what somebody remembers is a fact about their work rather than about naming. A researcher remembers authors. A property manager remembers addresses. A journalist protecting sources cannot put the most memorable thing in the title at all, since the title is searchable and searchable is the opposite of what that material requires.

The considerations pull against each other, which is why the subject produces such strong opinions. A name optimised for search wants the terms somebody will type. A name optimised for scanning wants to be short enough to read at a glance down a list of forty. A name optimised for privacy wants to reveal nothing. A name optimised for evocation, which matters for creative work, wants to be memorable rather than systematic. No single scheme serves all four, and choosing between them requires knowing which one the work actually demands.

What follows is a thread in a small private group where somebody raised the question and everybody answered, which produced twelve incompatible schemes, one genuine argument, and considerably more agreement about the underlying principle than anybody expected.

The post that started it

Tomas posted on a Tuesday evening and apologised in the first line, which several people told him afterward was becoming a habit.

The post read: I have just spent forty minutes failing to find something I wrote in March and I have worked out why. I have named my entries three different ways since I started, depending on what I was thinking about that week. The search is doing exactly what it should and my titles are giving it nothing to work with. So: what are your naming conventions, and how did you arrive at them?

There were twelve replies over three days and one of them started an argument.

Tomas Herrera, chemistry teacher, San Antonio

Since I asked, I will go first, and I will describe both the mess and the fix.

The mess was that I had three conventions running at once. Some entries were named after the topic, because that is what I was thinking about. Some were named after the lesson number, because that is what my planning document uses. And some were named after the activity, which is how the students refer to them, and which is why I have an entry called The Balloon Thing that I could not find in March because I had forgotten the balloon.

What I have moved to is a pattern with three parts, in a fixed order. The course code first, then the unit, then the topic. So an entry looks like AP Chemistry, unit three, and then whatever the thing actually is. Stoichiometry lab procedures. Enthalpy diagrams. The specific heat experiment.

The order matters and it took me a fortnight to work out why. Course first means my list sorts into course groupings without any filtering, which is how I think during a planning period. Unit second means everything for a unit is adjacent, which is how I think during a teaching week. Topic last means the distinguishing part is the part I read, and by the time my eye reaches it I have already established the context from the two things before it.

The thing I got wrong initially was putting the topic first because it seemed like the most important information. It is the most important information and it is the wrong thing to lead with, because a list of forty entries all beginning with different topics is a list you have to read rather than scan.

Two other rules I have adopted and would defend.

The first is that a label never duplicates a page name. My pages say where something belongs: which course, which unit. My labels say what kind of thing it is: lab, homework, exam preparation, safety. If I had a label called thermochemistry I would have two ways of expressing the same fact and I would use them inconsistently, which is exactly the problem I have just spent forty minutes paying for.

The second is that I write the terms I search for rather than the terms that are technically correct. My students and I refer to a particular experiment by a name that appears in no textbook. That name goes in the title, because that is what I will type at seven in the morning. The formal name goes in the body, where it belongs and where the search will still find it at a lower weight.

I would add one confession. Retitling my old entries is not something I am going to do comprehensively. I have about six hundred and I am doing the ones I actually reach for, which I identified by looking at what I have searched repeatedly, and it is roughly fifty. That will take me two evenings and it will fix the problem for everything I actually use.

Priya: The point about leading with the thing that groups rather than the thing that distinguishes is exactly right and I arrived at it independently for a completely different field.

Suki: Six hundred entries and you are fixing fifty. That is the correct ratio and almost nobody has the discipline to stop there.

Priya Sharma, doctoral candidate, London

Mine has two conventions because I have two kinds of thing and they are not alike.

Papers are named with the first author, then an abbreviation of the journal, then a short form of what the paper actually establishes. Not its own title. Its own title is written to sound significant to a reviewing committee and frequently contains none of the words anybody would search for.

The first author is doing the heaviest work in that pattern and I want to explain why, because it is the single most useful thing I have learned about naming in four years. In my field, the retrieval key is almost always a person. I do not remember what a paper was called. I remember that somebody’s group did the thing with the modified delivery mechanism, and I remember their name, because names are what circulate in conversation and at conferences and in every discussion I have ever had about anybody’s work.

Since a title match outranks a match anywhere else, putting the author name in the title means the entry surfaces first whenever I search a person, and searching a person is what I actually do.

Experiments are the other kind and they use a different pattern entirely. A sequential identifier first, then a brief description of what was being tested. The identifier is not decorative: my notebooks use it, my analysis scripts use it, and my supervisor refers to it, so it is the one string that appears in every context around that experiment.

Here is what I got wrong for three years and am now fixing. My experiment titles used to be the identifier and a date and nothing else, on the grounds that the identifier was unique and therefore sufficient. It is unique and it is not memorable, and I never search for an identifier because I do not know which one I want. I search for the variable I was investigating, and that word appeared nowhere in the title.

I am adding it to about a hundred and forty of them, which is the subset I go back to, which is the failures rather than the successes. A successful experiment produces a figure that ends up in a paper. A failed one produces knowledge that exists nowhere else, and those are the ones I reach for.

One more thing which is specific to a long project. My chapter labels do work my titles cannot, because a paper may bear on three chapters and its title cannot say so without becoming absurd. So the title is about the paper and the labels are about my argument, and those are genuinely separate axes rather than two ways of saying the same thing.

Claire: The observation that a document’s own title is written for a different purpose than retrieval is exactly true in my field as well and I had never articulated it.

Felix: A paper is titled to impress a committee. That sentence explains about thirty years of my irritation with my own discipline.

Claire Fontaine, patent attorney, Chicago

Mine is the most rigid scheme in this thread and I make no apology for it, because my titles are read by people other than me and occasionally by people who are not friendly.

Four components in a fixed order. A client code, then a matter number, then the document type, then a brief description of the specific document.

The client code is three or four letters and it is the same code the firm uses in its billing system, which is deliberate. A convention that agrees with the conventions already in use around it requires no translation, and translation is where errors enter. Anybody in the firm reading one of my titles recognises the client immediately without knowing anything about my personal system.

The matter number is sequential within each client rather than globally, which is a decision I made in my first month and which I would defend against anybody. Globally sequential numbers tell you when something started and nothing else. Client-sequential numbers tell you that this is the fourteenth matter for that client, which is information about a relationship, and I frequently want to know how deep a relationship runs before I want to know anything else.

The document type is a small controlled vocabulary. Application. Office action. Response. Prior art. Claim chart. Correspondence. Six or seven words, used exactly, never varied. This is the component that makes a matter scannable: a matter with thirty entries under it is unreadable if the types are described in free text and entirely readable if the same six words appear in the same position every time.

The description is the only free part and it is short. Enough to distinguish this response from the previous response, which usually means naming the claim or the ground it concerns.

Priya asked me about numbering and I will answer it properly because it is a real question. Sequential numbering has one genuine weakness, which is that it encodes the order things arrived and nothing about what they are. That is fine for a matter, because a matter is a proceeding and its order is meaningful. It would be wrong for anything where the sequence is arbitrary.

The other thing I want to say concerns version history and titles, because I see people confuse these.

A title identifies a document. The history identifies its states. Those are different jobs and mixing them produces the file-naming disease everybody in this thread has suffered from: response final, response final revised, response actually final. I have never once put a version marker in a title, because the entry is one document with a history inside it, and there is nothing to distinguish.

I would offer one warning that is specific to regulated work. Whatever appears in a title may be read by somebody who was not the intended audience. A colleague picking up a matter. Somebody in a discovery process. A person glancing at a screen. That is a reason for precision rather than for vagueness, but it is a genuine constraint and it rules out the kind of shorthand I would otherwise use.

Priya: Client-sequential rather than globally sequential is the answer to a question I have had for two years and never asked anyone.

Marcus: The point about matching the convention already in use around you is the one I would put at the top of this whole thread.

Marcus Webb, property investor, Denver

Mine is short because everything about my work rewards brevity.

Three components. An abbreviated address, then a document type, then a four-digit date.

The abbreviation is the part that matters and it is the part everybody questions. My properties are pages named with the full address written out, no shortenings, because the page name has to be unambiguous and sorts predictably. My entry titles use a compressed form: the street number and the first few letters of the street name, run together.

The reason is scanning. A unit page may hold sixty entries and I am reading down a list. Full addresses at the front of every title would occupy half the width and would be identical across every entry on that page, which is a considerable amount of screen space spent repeating what the page already told me.

The compressed form is short enough that the document type and the date are visible at the same time, and those are the parts I am actually distinguishing between.

The date is four digits, month and day, no year. The year is implied by ordering and by the property context, and adding it makes every title longer for information I have never once needed to see in a list. If I want the year I look at the dates column, which is right there.

The document type is a fixed vocabulary of about eight words. Inspection. Lease. Invoice. Notice. Certificate. Correspondence. Photos. Report. Never varied, never abbreviated, always the same word for the same kind of thing.

That last rule is the one I would put above everything else, and I learned it the expensive way. For my first year I used inspection, inspection report, and property inspection interchangeably, which are three names for one thing, and a query for any one of them returned a third of what existed.

One further thing, which came up in another conversation and which I have started applying. Anything arriving as a photograph of paper, rather than as a document, gets its key terms written into the title deliberately, because the extraction from a phone photograph depends on how somebody held the phone and I cannot control that. A contractor’s invoice photographed badly still gets found if I put the contractor’s name in the title myself.

I do not name most files. About one in fifteen, and only the ones that carry weight.

Nkechi: Compressed identifier first, then category, then date. That is maritime convention almost exactly and neither of us learned it from the other.

Amir: Eight words never varied. I am going to say something about this in my own post and you are all going to be unkind to me.

Suki Tanaka, security consultant, Tokyo and remote

Mine is three components and one of them exists for a reason nobody else in this thread has.

Client code, engagement type, phase.

The client code is short and it is not the client’s name. That is the deliberate part. A screen visible in a cafe, a laptop open on a train, somebody glancing over during a call: none of those should reveal which organisations I am currently working for, because the fact of an engagement is frequently more sensitive than anything inside it.

An organisation discovering that a competitor has retained a security consultant learns something valuable. An organisation discovering that I am in the middle of an engagement learns when they are least defended. Neither of those requires anybody to read a single finding.

So the codes are arbitrary, they are consistent, and the mapping lives in one encrypted entry that I open rarely.

The engagement type and the phase are both from small fixed vocabularies. Four types, six phases. Together they mean that any title tells me, at a glance, what kind of work this is and how far along it is, which is the same information the board columns carry and which I want available even when I am not looking at a board.

The thing I would add that is specific to my work concerns what goes inside entries rather than on them. My stored scripts and commands follow their own naming convention, separate from the entry titles, and I am strict about it because a snippet with an unhelpful name is a puzzle I have set for myself. The name of a stored procedure should state what it does and against what, in words, not in whatever I typed while tired.

On the general question Tomas asked, I would make one point that I think is underweighted in this thread so far.

A convention is a decision made once and applied thousands of times. That asymmetry is what makes the initial thinking worth doing properly and what makes changing it later so expensive. It also means the value of a convention is almost entirely in its consistency rather than in its cleverness. I have seen elaborate schemes fail because they required judgment at the moment of naming, and simple ones succeed for twenty years because they required none.

If your convention requires you to think, it will be applied inconsistently, and inconsistently applied is functionally the same as absent.

Lena: The point about the fact of an engagement being more sensitive than its contents is the correct framing and I use the identical logic for entirely different reasons.

David: I understand every word of this and none of it applies to me, which I think is the most useful thing about reading it.

Lena Johansson, investigative journalist, Stockholm

Suki has described the milder version of my rule and I want to state the strict one, because for me this is not a preference.

Three components: a source code, a document type, and a summary of what the content concerns.

No real name appears in any searchable field. Ever. Not in a title, not in a label, not in a hashtag, not in the name of an attached file. This applies to sources, to people mentioned by sources, and to organisations where naming them would identify who I am talking to.

The reason is not that somebody might read my screen, although they might. It is that a title is searchable, and searchable means that anybody who ever gains access to the material, by any route, has a fast index into it. A well-named archive is an efficient archive, and efficiency is neutral about who is using it.

So a source is a code. The mapping between codes and people exists in one encrypted entry with its own password, and that entry is the only place in my entire system where the two are connected.

The document type is a small vocabulary: memo, filing, correspondence, transcript, dataset, bundle.

The summary is where I put the retrieval terms, and it is deliberately about subject matter rather than about people. Executive compensation. Subsidiary structure. Regulatory correspondence. Those are the terms I search, they are meaningful to me, and they identify nobody.

There is a cost to this and I want to be honest about it, because everybody else in this thread is optimising for retrieval and I am not.

My titles are worse at retrieval than they could be. If I could put real names in them, my searches would be faster and my results would be more precise, because a person’s name is the most distinctive string available and it is exactly what I would type. I have given that up deliberately and I pay for it in small amounts every day.

What I have gained is that no single failure produces a map. Somebody who obtains my folder and cannot open the mapping entry has a set of documents about subjects, attributed to codes, which is considerably less useful than a set of documents attributed to names.

One practical consequence worth stating. Because my titles cannot carry the most memorable thing, I lean harder on everything else. My page structure does more work than most people’s, because a source’s page is where the context lives. My labels are more specific. And I use the search inside document contents constantly, since the real names are in the documents themselves and the documents are not what I control the naming of.

The general point I would offer to anybody whose material is sensitive: decide what your titles are allowed to contain before you name anything, because a title written once is a title that exists forever, and retroactively removing a name from four hundred entries is not a project anybody completes.

Suki: The line about efficiency being neutral about who is using it is going into my client presentations, attributed.

Meera: I have a considerably milder version of your problem and your last paragraph has just changed something about how I am going to handle new patients.

Raj Malakar, documentary filmmaker, Berlin and Mumbai

Mine is three parts and one of them is a number, which several of you will approve of and one of you will not.

Project abbreviation, scene or segment number, description.

The project abbreviation is three or four letters and exists because I run three projects simultaneously and always have. Without it, my entry list is a mixture of three films and I cannot tell at a glance which is which.

The number is the part that matters and it is not a sequential identifier in the way Priya’s is. It is a position. Scene fourteen is the fourteenth thing in the current cut, and when the cut changes, that number is wrong.

Which is exactly the objection somebody is about to raise, so I will raise it myself. Yes, my numbers go stale. Yes, I renumber. It happens perhaps three times per project, at points where the structure genuinely changes, and it takes an evening.

I do it anyway because the alternative is worse. Editing is sequential, my head is sequential during a cut, and a list sorted into the order of the film is the single most useful view I have. A list sorted alphabetically by description is a list I have to think about.

The description is short and it is concrete. Not what the scene is about. What is physically in it. Indigo dyeing process. The market at dawn. The interview in the hospital corridor. Because when I am searching at eleven at night I am remembering an image rather than a theme.

That is the distinction I would offer to anybody doing visual work. Name the thing you saw, not the thing it means. The meaning changes across a production and the image does not.

Hashtags are a separate convention and I keep them entirely apart from titles. Titles say what something is. Hashtags say what stage it is at, and those move constantly. Mixing them would mean editing a title every time something progressed, which is both tedious and a good way to lose the number.

One more thing, which is a confession. My first project used descriptive titles with no numbers and no project code, because I was being creative and thought codes were bureaucratic. That project is the one I cannot navigate, four years later, and it is the one I most frequently need to, because material from it keeps turning out to be relevant to what I am doing now.

Being creative about naming is being creative in the one place where creativity has no benefit and a permanent cost.

Nkechi: Renumbering three times a project sounds appalling to me and I have just realised it is because my numbers are dates and dates do not move.

Tomas: Name the thing you saw rather than the thing it means. I am going to use that with students about their lab notes.

Pastor David Achebe, Atlanta

I am going to be the dissenting voice here and I want to do it properly rather than apologetically.

I do not use codes. I do not use numbers. My entry titles are sentences a person would say.

Summer camp planning, volunteer coordination. Conversation with the family about the funeral arrangements. Elder board, the building question.

I have read everything above with genuine admiration and I want to explain why almost none of it would work for me.

The first reason is that my memory is verbal rather than systematic. I do not remember that something was the fourteenth of anything. I remember that it was the conversation in the car park after the service, and if my title says that, I find it. A code would be one more thing standing between me and the material, and it would be a thing I had to translate.

The second reason is that I am sixty-two and I have watched myself abandon four systems, and every one of them failed at the point where it required me to do something slightly artificial. Not difficult. Artificial. A scheme that asks me to produce a code before I can write a note is a scheme I will bypass at half past nine at night after a hospital visit, and once I have bypassed it twice it is finished.

The third reason is that my labels do the classification work that everybody else’s codes are doing. I have six of them and they are consistent and they carry the structure. The titles do not need to.

Now, the honest part.

My titles are worse at search than Priya’s would be. I accept that. What I get in exchange is that I actually write them, in a form I will recognise, at the moment I am least inclined to do anything careful.

The one place I do use a convention is pastoral material, and it is minimal for reasons that have nothing to do with retrieval. Those entries carry an initial rather than a name. Not because the entry is unprotected, since it is encrypted, but because the title appears in a list, and a list is visible at a glance in a room where somebody might be standing.

That is the only place where I let a system override what feels natural, and I let it override there because the stakes are somebody else’s.

Amir: Thank you. I was preparing to be the only person here without a scheme and now I am the second.

Meera: The distinction between difficult and artificial is precise and it explains almost every abandoned system I have watched a colleague give up on.

Captain Nkechi Obi, cargo ship officer, at sea

Mine is not a personal convention. It is a professional one that predates me by about a century, and I have simply applied it.

Vessel abbreviation, category, date.

The vessel comes first, always, without exception, and this is not negotiable in my industry. Every document produced aboard any ship identifies the ship first, because a document without a vessel is a document that cannot be filed by anybody ashore, and because I will serve on several vessels and my records need to remain sortable across all of them.

The category is from a short fixed list: inspection, drill, cargo, maintenance, handover, certificate.

The date is four digits, month and day. Never the year, for the same reason Marcus gave, which I read with some satisfaction because we arrived at it separately.

Raj’s comment about renumbering is the thing I want to respond to, because it clarified something for me.

My numbers are dates. A date is assigned by the world rather than by me, it never becomes wrong, and it never needs to be reissued. Which means my ordering is stable in a way his cannot be, and that is not a virtue of my system. It is a property of my work. Operations happen in time and films are assembled in an order somebody chooses.

What I would add for anybody working shifts or watches is a point about when the naming happens.

I capture during a watch in a chat-style log, one line at a time, with no title at all, because I have eleven seconds and titling is a decision. At the end of the watch I turn those lines into proper entries, and that is when they get named, when I am sitting down and can think.

Separating capture from naming is the thing that makes a convention survivable for anybody whose work interrupts them. A scheme requiring a well-formed title at the moment something happens is a scheme that will be abandoned during the first genuinely busy period, and busy periods are when the most important things happen.

One last thing. My certification entries carry the expiry date in the title, which breaks my own pattern, and I do it deliberately. Those are the entries where the date is the most important fact about the item, and a convention that hides the most important fact is a convention worth breaking once, in a documented and consistent way.

Marcus: Dates never becoming wrong is the cleanest argument for date-based naming I have heard and I had never thought of it as a contrast with anything.

Priya: Separating capture from naming is going into my laboratory practice this week.

Amir Hassan, bakery owner, Toronto

I have been reading this thread for two days and I want everybody to know that I have enjoyed it enormously and that I am about to disappoint all of you.

I do not have a convention.

My entries are called things like Mum’s baklava. That guy who delivers the flour. Tuesday was crazy. The dough that would not rise in February. Something I should ask Yasmin about.

Suki is going to be unkind and I have decided to go first.

Here is my defence, and I do think it is a defence rather than an excuse.

My collection is about nine hundred entries across four pages. Nine hundred is not a small number and it is not eleven hundred papers or eighty-one applications either. At my size, with my labels applied consistently, I have never once failed to find something.

I searched for that guy who delivers the flour last month. Not those words. I searched flour and the supplier’s name, which is in the entry because I wrote it in the body, and it came back first. The title being informal cost me nothing because the search reads the body as well.

My labels are consistent. Six of them, always the same, never varied. That is where my discipline lives, and I put it there because it is the place where I could sustain it.

And my titles are the way I actually think, which means when the widget hands me a random note at seven in the morning I know what it is from the title, immediately, without translating anything.

Now the honest part, because everybody else has been honest.

If I had eleven hundred papers I would be in trouble, and I know it. My approach works at my scale and would not work at Priya’s. If my business doubled I would have to change, and I would rather change then than adopt a scheme now for a problem I do not have.

And there is one place I have adopted a convention, which is my supplier entries. Those are named with the supplier and then what the thing is, because I have eleven suppliers and forty years of history with some of them, and that is the one part of my vault where I genuinely was losing things.

So my position is: have a convention where you need one, and not everywhere.

Suki: I was preparing something unkind and you defended it well enough that I have deleted it. Scale is a legitimate argument and I concede it.

Felix: That guy who delivers the flour is a better entry title than anything in my manuscript branch and I intend to steal the register if not the practice.

Dr. Felix Brennan, retired professor, Vermont

Mine is unlike everybody else’s because my titles have a second job.

Two components: a decade and a category, then a descriptive title that is doing something none of yours are.

The nineteen-nineties, teaching, and then something like the year I almost left. The department, and then the argument about the survey course. Correspondence, and then a name.

The decade and the category are ordinary. They group, they sort, they let me find a branch of my life without filtering.

The last part is where mine diverges. My descriptive titles are evocative rather than descriptive, deliberately, because a substantial number of them will become chapter headings and because I am writing something that has to be read rather than retrieved.

There is a practical argument for this and it is not sentimental.

I begin every working day with the widget that hands me a random entry from anywhere in the vault. It refreshes hourly and I read whatever it offers. And what determines whether that ritual produces anything is entirely the title, because the title is what I see first and what decides whether I open the thing.

An entry called Correspondence, box four, second folder is an entry I will not open. An entry called The letter I never answered is an entry I will open every time, and about one time in twenty opening it produces a chapter.

So my titles are optimised for a mechanism that most of you do not use in the way I do. They are written to make a version of me at seven in the morning curious.

I will grant Priya’s point that this costs me retrieval. My titles contain fewer searchable terms than they might and I find things slightly more slowly. I have decided that trade is correct for a book and would be wrong for a thesis, which is a distinction about the work rather than about naming.

One thing I would say to anybody in a long solitary project. Your titles are the only voice you have that reaches your future self, and your future self is the only reader you have for several years. Write them for that person. Nobody else is going to read them, and by the time anybody does, the titles will have already done their work.

David: Write them for the only reader you have for several years. That is going in my own notes and I do not care that it is a memoir principle.

Raj: Optimising titles for a random widget is the most specific piece of tooling advice in this entire thread and I am now wondering what mine are optimised for.

Dr. Meera Iyer, psychiatrist, Austin

Mine is the shortest scheme in this thread and it is short for a reason nobody else has.

Patient initials, then a sequential session number. That is the whole title.

Two characters and a number. Nothing descriptive, nothing about content, nothing about what happened.

The reason is that my entries are encrypted and the title is not. A title appears in a list. A list is visible on a screen. A screen is in a room, and my room has a door that people knock on and occasionally open, and the person opening it may be the next patient.

So the title carries the minimum required to identify the entry to me and nothing that would mean anything to anybody else. All of the clinical content is inside, behind a password.

That is the entire logic and it produces a scheme that would be inadequate for anybody else here.

There is a second benefit which I did not design for and now value. A list of forty entries, each two characters and a number, is scannable in a way that descriptive titles never are. I can find a specific session in a specific patient’s sequence in about a second, because there is nothing to read.

Lena’s post changed something for me and I want to acknowledge it publicly, because she said something I had half understood.

I use initials. Initials are not a name and they are also not nothing, particularly in a small practice where a person might reasonably guess. Her point about deciding what titles are allowed to contain before naming anything has made me think that for new patients I should use an arbitrary identifier rather than initials, with the mapping in one encrypted entry.

I am not going to retitle four hundred existing entries. I am going to do it going forward, which means I will have two schemes running for some years, which I am prepared to accept because the alternative is a project I will not finish.

The general principle I would offer, since everybody else has offered one: work out what your titles are exposed to before you decide what they should contain. Everybody in this thread has optimised their titles for retrieval. Two of us have optimised them for what happens when somebody else sees them, and that is a completely different constraint that produces a completely different answer.

Lena: Two schemes running for some years is the correct decision and I want to say so, because the instinct to make everything consistent retroactively is how people end up doing nothing.

Claire: Optimising for what happens when somebody else sees the list is a consideration I have only ever applied to content and never to titles.

The argument, which lasted about a day

The thread produced one genuine disagreement and it is worth recording because it did not resolve.

Priya raised it, addressing nobody in particular: should a title be optimised for search or for a human reading a list?

Her position was that search is the primary interface at scale, that a person reading a list is a person who already knows roughly where they are, and that the case where somebody genuinely does not know is the case where search is doing all the work. Therefore titles should be written for the query.

Marcus disagreed immediately and from experience. His argument was that scanning is what actually happens most of the time, because most retrieval is somebody looking at a page they have already navigated to, and that a title crammed with search terms is a title too long to scan. He pointed out that his own compressed addresses are worse for search and considerably better for reading a list of sixty.

David sided with Marcus for a different reason, which was that a title he cannot read at a glance is a title he will not write in the first place.

Suki sided with Priya on the grounds that scanning fails as a collection grows and search does not, so optimising for scanning is optimising for a phase.

Felix pointed out that both of them were describing a tradeoff rather than a principle and that the answer therefore depended on collection size, which nobody disputed and which did not end the argument.

Claire ended it, or rather stopped it, with an observation everybody accepted and which did not resolve anything.

She wrote: the two goals conflict only in the middle of a title. The first component is always for scanning, because it groups. The last component is always for search, because it distinguishes. What we are arguing about is how much space to give each, and that genuinely depends on how large the collection is and how often the person is navigating rather than searching.

Nobody was persuaded to change their scheme. Several people said afterward that the exchange was the most useful part of the thread.

The thing everybody had in common and nobody mentioned

Somebody, three days after the thread had apparently finished, posted an observation that produced a second round.

It was Meera, and she began by apologising for reopening it.

She wrote: I have read all twelve of these several times and I have just noticed something none of us said. Every single scheme here, including the two that claim not to be schemes, has exactly the same shape. There is something at the front that groups and something at the end that distinguishes. Nobody designed that. We all arrived at it.

Suki replied within the hour, which several people noted was faster than she had responded to anything in eleven months.

She wrote: That is correct and it is more interesting than it sounds. A name is doing two different jobs at once and they want opposite things. Grouping wants stability and repetition, because a group is only visible if the marker is identical across its members. Distinguishing wants variation, because a distinguisher that repeats has failed. Putting one at each end is the only arrangement that lets both work.

Priya added a point that Claire immediately took issue with.

She wrote: which means the middle is where the argument lives, and that is exactly what we spent a day disagreeing about.

Claire wrote: not quite. The middle is where the argument lives for people whose collections are large enough to need one. Amir has a front and a back and nothing between them, and his scheme works. So does David’s. The middle is optional.

Amir wrote: I have no idea what any of you are talking about and I have just looked at nine hundred entry titles and you are all correct about me.

Felix wrote something that ended it.

He wrote: I would put it differently and I think more usefully. The front of a name is written for the person navigating and the back is written for the person searching, and in every case both of those people are you, at different moments, in different states. The front is for the version of you who knows roughly where they are. The back is for the version who does not. A good name serves both of them and a bad one serves whichever you happened to be when you wrote it.

Nkechi wrote: that is the whole thread in four sentences and I am irritated that it took eleven of us three days to arrive at it.

Tomas wrote: I am not irritated at all. I would not have understood it if somebody had said it on the first day.

Consistency beats sophistication

Twelve people, twelve schemes, and no two alike. Read as a survey, that looks like an absence of any principle. Read carefully, it is the opposite: every one of those schemes is a correct answer to a different question, and the questions differ because the work differs.

Four distinct priorities appear across the thread and each one produces a different scheme. Retrieval optimisation puts the terms somebody will actually search into the highest-weighted field, which is why a researcher leads with an author name and a teacher with a course code. Scanability keeps titles short enough to read down a list of sixty, which is why a property manager compresses addresses and a clinician reduces a title to two characters and a number. Security determines what a title is allowed to contain at all, which for a journalist and a consultant rules out the single most memorable string available. And evocation matters where the title has a second job, as it does for somebody whose entries are candidate chapter headings and whose working day begins with a mechanism that offers a random one.

No scheme serves all four, which is why the argument in that thread did not resolve and should not have. What emerged instead was a structural observation: the first component of a title groups and the last component distinguishes, so the first is always doing scanning work and the last is always doing search work. Every scheme in the thread has that shape, including the two that reject conventions entirely.

The genuinely universal points are smaller and less interesting than any individual scheme, which is characteristic of things that are actually true.

Consistency outperforms sophistication by a wide margin. An elaborate convention applied unevenly produces a collection that cannot be corrected, because the items sharing a mistake cannot be gathered. A crude convention applied without exception can be fixed in an afternoon whenever somebody decides to. Every practitioner in that thread who described a failure described inconsistency rather than a wrong choice.

A convention that requires judgment at the moment of naming will not survive. This is why the most rigid schemes in the thread are also the ones with the smallest controlled vocabularies: six document types, four engagement types, eight categories, never varied. Judgment is what fails at the end of a long day, and a scheme designed for a well-rested user is a scheme designed for the wrong person.

Naming should match the conventions already in use around the work rather than inventing new ones. A client code that matches the billing system, a vessel abbreviation that matches a century of maritime practice, and a term that students actually use all require no translation, and translation is where errors enter.

And separating capture from naming is what makes any convention survivable for anybody whose work interrupts them. A scheme demanding a well-formed title at the moment something happens will be abandoned during the first genuinely busy week, which is precisely when the most important material is generated.

The two dissenters are worth taking seriously rather than treating as charming exceptions. One argued that his memory is verbal rather than systematic and that a scheme requiring something artificial, however easy, is a scheme he will bypass and then abandon. The other argued that his collection is small enough that consistent labels do the work titles would otherwise have to, and conceded frankly that this would fail at ten times the size. Both are describing a genuine fit between a practice and a context, and both had a convention exactly where they needed one and nowhere else.

Which is the actual conclusion. The right scheme is the one somebody will apply without deciding to, on a bad day, for years. Any convention meeting that standard outperforms a better one that does not, and the sophistication of the scheme turns out to matter considerably less than whether the person naming things has to think about it at all.

FAQs

Why do entry titles matter more than they appear to?

Because a title is an address rather than a description. In a ranked retrieval system a match in a title carries substantially more weight than a match anywhere else, so the words chosen at the moment of creation are the terms most likely to surface the item years later. Titles are also the most durable decision made about an item: content gets revised and structure gets rearranged, while the name sits unchanged, determining whether anybody reaches what is underneath it. Most people name in a second and pay for it for years. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field.

What is the most common naming mistake?

Naming descriptively rather than for retrieval, which is invisible for months. A researcher titles a paper with its own title, which was written to sound significant to a committee and contains none of the words anybody searches. A property manager names an inspection after the date, which is the one thing nobody remembers. A minister names a note after what happened, which is exactly what has faded by the time it is needed. In each case the name is accurate and useless. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

Why is inconsistency worse than a wrong convention?

Because a wrong convention applied uniformly is correctable and inconsistency is not. Items sharing a mistake can be found and adjusted together in an afternoon. Items named three different ways cannot be gathered at all, since the whole problem is that similar things were named dissimilarly and no query collects them. Practitioners describing naming failures almost always describe drift across months rather than a single bad decision, and drift is the version that cannot be undone. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Should the most important word come first in a title?

Usually not. The first component should group and the last should distinguish. Leading with the distinguishing detail produces a list somebody has to read rather than scan, because every title begins differently. Leading with the grouping element, such as a course code, a client code, or a vessel abbreviation, means the eye establishes context before reaching the part that differs. Practitioners who initially led with the topic and later changed describe the improvement as immediate. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

Is a title for search or for scanning?

Both, and the two goals conflict only in the middle. Because the first component groups and the last distinguishes, the front of a title is inherently doing scanning work and the end is doing search work. The genuine question is how much space to give each, and the answer depends on collection size and on whether somebody typically navigates to a page or searches from scratch. Larger collections favour search terms; heavy navigation within pages favours brevity. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

What makes a controlled vocabulary useful in titles?

It removes judgment at the moment of naming, which is where conventions fail. A fixed list of six or eight document types, used exactly and never varied, makes a page of thirty entries scannable because the same word appears in the same position every time. The alternative is free text, which produces three names for one thing and a query returning a third of what exists. Practitioners who have made this mistake describe it as the single most expensive naming error. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How should somebody choose which terms go in a title?

By identifying the retrieval key for their field, which is whatever they actually remember at the moment of searching. In academic work it is usually a person, since names circulate in conversation while paper titles do not. In property it is an address. In legal work it is a client and a matter. In creative work it is frequently a concrete image rather than a theme. The key differs by profession and is a fact about the work rather than about naming. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Should a document’s own title be used as the entry title?

Rarely, because documents are titled for purposes unrelated to retrieval. Academic papers are titled to signal significance, corporate reports to be defensible, and legal documents to satisfy formal requirements. None of those processes selects the words somebody will later search. A more useful pattern is the retrieval key first, then a short statement of what the item actually establishes or contains, with the formal title in the body where search will still reach it at lower weight. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

Do labels and pages change what titles need to carry?

Substantially. Where pages express location and labels express kind, titles are relieved of both jobs and can concentrate on identifying the specific item. A rule that labels never duplicate page names prevents the same fact being expressed two ways, which is where inconsistent usage begins. Practitioners with strong label discipline can afford looser titles, which is the explicit argument made by those in the thread who use natural-language names. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

How should dates appear in titles?

Compactly or not at all. A four-digit month and day is usually sufficient, since the year is implied by ordering and by context, and including it lengthens every title for information rarely needed while scanning. Where a date is the most important fact about an item, such as an expiry, putting it prominently is worth breaking a pattern for. The general principle is that a convention hiding the most important fact about a category deserves one documented exception. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

Why do date-based identifiers avoid a problem that sequential ones have?

Because dates are assigned by the world and never become wrong. A sequential number encoding position in a sequence somebody controls, such as a scene in an edit, goes stale whenever that sequence changes, which requires periodic renumbering. A sequential number encoding order of arrival, such as a matter number, remains valid because arrival order is fixed. Choosing between them means asking whether the sequence being encoded is imposed by the work or chosen by the person. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Should sequential numbering be global or scoped?

Scoped, in most cases. Numbering matters within a client, a project, or a subject conveys something meaningful, such as this being the fourteenth engagement in a relationship. Globally sequential numbers only encode when something started, which is information already available from dates. Scoped numbering also keeps identifiers short and makes them recognisable to colleagues who know the scope, which is a genuine benefit in shared professional settings. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

What are the security considerations in naming?

Titles are searchable, visible in lists, and readable at a glance by anybody near a screen, which makes them the wrong place for sensitive identifiers. Two distinct concerns arise: revealing who somebody is working with, which can be sensitive independently of any content, and providing an efficient index into material for anybody who gains access. Both are addressed by codes rather than names, with the mapping held in a single encrypted entry protected separately from everything else. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Does encryption remove the need for careful titles?

It changes the requirement rather than removing it. Encrypted content is protected, while the title generally appears in lists and is therefore visible. This argues for minimal titles on protected material: enough to identify the item to its owner and nothing meaningful to anybody else. Practitioners in clinical and pastoral settings converge on initials or arbitrary identifiers with a sequential number, which is scannable, sufficient, and reveals nothing to somebody glancing at a screen. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

Is it worth using arbitrary identifiers rather than initials?

For anybody whose material concerns identifiable individuals, arbitrary identifiers are stronger. Initials are not a name and are not nothing, particularly in a small practice where somebody could reasonably guess. Moving to an arbitrary code with a mapping held in one encrypted entry removes that inference entirely. Practitioners making this change generally apply it going forward rather than retitling everything, accepting two schemes running in parallel rather than attempting a project they will not finish. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Why do natural-language titles work for some people?

Because memory is verbal for many people and a code is one more thing standing between them and their material. Somebody who remembers a conversation as the one in the car park after the service will find it if the title says so, and will not find it under an identifier they never think in. The tradeoff is genuine: such titles retrieve less precisely. What they buy is that the person actually writes them, in a recognisable form, at the moment they are least inclined to be careful. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field.

What is the difference between a convention being difficult and being artificial?

A difficult convention is one that takes effort; an artificial one requires something that does not match how somebody thinks. Practitioners who abandon systems almost never do so because the system was hard. They do so at the point where it asked for something slightly unnatural, such as producing a code before writing a note, because that step gets bypassed once during a demanding evening and twice thereafter. Designing for how somebody actually thinks matters more than designing for elegance. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

When is having no convention defensible?

At small scale, with consistent labels, in a collection where search over full content reliably compensates. Somebody with several hundred entries across a handful of pages, whose labels are applied without variation, may genuinely never fail to find anything, because content search reaches the body regardless of what the title says. The honest version of this position acknowledges that it would fail at ten times the size, and adopts a convention specifically in the one area where things were being lost. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day.

Can somebody have a convention in some places and not others?

Yes, and it is a reasonable strategy. The useful question is where things are actually being lost, which is rarely everywhere. A practitioner may name most entries informally while applying a strict pattern to supplier records, client matters, or anything with many similar items. Adopting a convention only where the problem exists avoids imposing overhead on material that never needed it, which is one reason such partial systems survive while comprehensive ones sometimes do not. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Should naming happen at the moment of capture?

Not for anybody whose work interrupts them. Separating capture from naming is what makes conventions survivable: quick capture during a shift, a watch, or an experiment produces untitled fragments with timestamps, and the naming happens later when the person is sitting down and able to think. A scheme demanding a well-formed title at the moment something happens will be abandoned during the first genuinely busy period, which is exactly when the most important material is generated. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How much retroactive renaming is worth doing?

A deliberately chosen small subset rather than a comprehensive pass. The candidates are identifiable from query history: the entries somebody actually reaches for repeatedly. Fifty entries out of six hundred, or a hundred and forty out of three hundred, is typical, takes a couple of evenings, and fixes the problem for everything genuinely in use. Comprehensive retitling projects are announced frequently and completed rarely, and abandoning one halfway produces exactly the inconsistency it was meant to cure. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Which entries are most worth retitling?

The ones consulted repeatedly, which are frequently not the ones that look most important. In research this is often the failed experiments rather than the successful ones, since a success produces a published figure while a failure produces knowledge that exists nowhere else. In operational work it is the records that support disputes. The reliable way to identify them is to look at what has actually been searched for rather than to reason about what ought to matter. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

Should titles carry version information?

No. A title identifies a document while the history identifies its states, and mixing them produces the familiar file-naming failure where several items claim to be final. Where a system keeps versions attached to the entry itself, there is nothing to distinguish by name: one entry holds the document and its sequence of states. Practitioners who have made this transition describe the disappearance of version markers from titles as one of the clearest improvements. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Should naming conventions match those already used elsewhere in an organisation?

Wherever possible, since a convention agreeing with its surroundings requires no translation and translation is where errors enter. Using the same client code as a billing system, the same vessel abbreviation as the wider industry, or the same term students actually use makes titles immediately legible to colleagues and removes an entire category of mistake. Inventing a parallel scheme means maintaining a mapping in somebody’s head, which fails during exactly the periods when accuracy matters most. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How do hashtag conventions relate to title conventions?

They should be kept entirely separate, because they change at different rates. A title identifies what something is, which is stable. A hashtag typically encodes what stage something is at, which moves constantly. Combining them would mean editing a title every time work progressed, which is tedious and also destroys the stability that makes a title a reliable address. Practitioners running both describe the separation as essential rather than merely tidy. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Do evocative titles have any practical benefit?

For anybody using a mechanism that surfaces random material, considerably. A widget offering one entry at a time is only useful if the title makes somebody curious enough to open it, and an entry named as a box number and a year will not be opened, while one named for what it actually contained will. This matters most in long creative or reflective projects, where titles frequently become candidate headings and where the daily encounter with forgotten material is the point of the exercise. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field.

What is the universal principle underneath twelve different schemes?

That the right convention is the one somebody will apply without deciding to, on a bad day, for years. Consistency outperforms sophistication by a wide margin, controlled vocabularies beat free text because they remove judgment, and matching existing conventions beats inventing new ones. Every scheme in a survey of experienced practitioners has the same structure underneath: something that groups at the front, something that distinguishes at the end, and nothing in between that requires thinking about. The point is worth settling early, since naming decisions are cheap to make and expensive to revise across an accumulated collection. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

Why does every naming scheme have the same underlying shape?

Because a name performs two jobs that want opposite things. Grouping requires stability and repetition, since a group is only visible if its marker is identical across every member. Distinguishing requires variation, since a distinguisher that repeats has failed. Placing one at each end is the only arrangement allowing both to work simultaneously. Practitioners across a dozen unrelated fields arrive at this structure independently without designing it, which suggests it is a property of what names are for rather than a convention anybody invented. It is one of the few pieces of advice about knowledge systems that costs nothing to follow and produces a return every subsequent day.

Who is a title actually written for?

Two versions of the same person in different states. The front of a name serves whoever is navigating, who already knows roughly where they are and needs the grouping marker to confirm it. The back serves whoever is searching, who does not know where anything is and needs the distinguishing term. A good name serves both; a poor one serves whichever version the author happened to be at the moment of writing, which is usually the navigating one, since they had just created the item. Anybody uncertain whether this applies to their own work can test it by attempting to find something they wrote six months ago.

Is the middle of a title necessary?

Not always, and this is where collection size decides. Practitioners with large collections need intermediate components, such as a unit, a matter, or a phase, because grouping and distinguishing alone are insufficient to locate one item among thousands. Practitioners with a few hundred entries frequently have only a front and a back, or no formal structure at all, and their systems work. The middle is the part worth arguing about, and only for people whose collections have grown enough to require it. The failure mode is gradual rather than sudden, which is why it is usually noticed only after the correction has become expensive. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Why is a naming convention hard to appreciate before using one?

Because the failure it prevents is invisible for months and then arrives all at once, usually as a search that should have worked. Somebody told about conventions on their first day has no material, no failed retrieval, and no reason to believe any of it matters. The same person after a year of inconsistent naming understands immediately. This is why advice about naming lands so poorly in advance and so well in retrospect, and why most people learn it the expensive way. Practitioners across unrelated professions arrive at this independently, which is a reasonable sign that it reflects the problem rather than any particular field.

How should pages be named as distinct from entries?

Pages should be unambiguous and complete, since they are the stable structure everything else hangs from and there are far fewer of them. A property page carries the full address written out without abbreviation, so it sorts predictably and nobody has to guess whether a word was shortened. A course, a client, or a project page carries its full identifier. Entries within that page can then compress, because the page has already established the context and repeating it in every title wastes the width that distinguishing information needs. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

Should abbreviations be used in titles at all?

Within entries, frequently; within page names, rarely. An abbreviation in an entry title saves horizontal space on a list where the fuller form is already implied by the page. An abbreviation in a page name introduces ambiguity into the one place that should have none. The important discipline either way is that an abbreviation must be formed the same way every time, since a scheme producing two shortenings of the same word has recreated the inconsistency problem in miniature. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

How long should an entry title be?

Short enough to read at a glance in a list of sixty, which in practice means a handful of components rather than a sentence. Practitioners who compress deliberately report scanning as the dominant retrieval mode within a page, and a title occupying the full width forces reading rather than scanning. The exception is anybody whose titles serve a second purpose, such as prompting curiosity when material is surfaced at random, where evocation justifies additional length. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

What should be done about a collection already named inconsistently?

Fix a small chosen subset and change practice going forward. Comprehensive retitling is announced frequently and completed rarely, and an abandoned half-finished pass produces exactly the inconsistency it was meant to cure. The subset worth fixing is identifiable from search history: the items reached for repeatedly. Everything else remains findable through its content, and after a year the well-named portion will be visibly easier to navigate, which settles the question by demonstration. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Is it acceptable to run two naming schemes at once?

Yes, and it is frequently the only realistic option when a convention improves. Somebody adopting a stronger scheme for privacy or clarity generally applies it to new material rather than retitling several hundred existing items, accepting that two schemes will coexist for some years. The instinct to make everything consistent retroactively is precisely how people end up doing nothing at all, and a better convention applied going forward accumulates value immediately. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How do naming conventions interact with quick capture?

They should be separated entirely. Quick capture during an interruption produces untitled fragments, frequently timestamped, which are turned into properly named entries later when there is time to think. A scheme requiring a well-formed title at the moment of capture will be bypassed during the first genuinely busy period, and once bypassed twice it is finished. Practitioners in shift-based, clinical, and laboratory work all describe this separation as what makes their conventions survive. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

Does a naming convention help or hinder somebody else using the archive?

It helps considerably, provided the convention matches vocabulary already in use around the work. A colleague recognises a client code taken from the billing system, a vessel abbreviation from industry practice, or a document type from the profession’s own terminology, with no explanation required. A private scheme, however elegant, requires a mapping that exists only in one person’s head, which fails the moment somebody else needs to work with the material. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

What role do labels play if titles are well designed?

They carry the dimension titles cannot, which is anything cutting across the hierarchy. A title identifies one item and a page places it; a label states a property true of items in several different places. Where labels are consistent and few, titles can be looser, which is the explicit argument made by practitioners using natural-language names. Where labels are weak or absent, titles have to carry more, which is why disciplined labelling and relaxed titling frequently appear together. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Should a title say what something is about or what it contains?

For visual and observational work, what it contains, because that is what somebody remembers. A scene named for its meaning becomes wrong as the meaning shifts across a production, while a scene named for what is physically in it remains accurate permanently. The same applies to photographs, field notes, and inspection records. For analytical work the reverse frequently holds, since what a document establishes is more memorable than the material it happens to discuss. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How does naming affect a random-surfacing mechanism?

Decisively, since the title is the only thing visible before somebody decides whether to open an entry. A mechanism offering one random item at a time depends entirely on whether that item’s name makes the reader curious, and an entry named as a container reference will simply never be opened. Practitioners who rely on this for creative or reflective work deliberately name for evocation, accepting slightly worse retrieval in exchange for a daily encounter that actually produces something. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

What is the cost of an overly rigid convention?

Being bypassed. A scheme requiring judgment, translation, or a lookup before anything can be written will be skipped at the end of a demanding day, and a convention skipped twice is finished. This is why the most durable rigid schemes pair strict structure with tiny controlled vocabularies: the structure is fixed and the choices within it are trivial. Rigidity in the shape is sustainable; rigidity requiring thought is not. Anybody weighing this against other priorities should note that it costs nothing to apply and cannot be retrofitted cheaply, which places it firmly among the decisions worth settling before a collection grows.

What is the cost of no convention at all?

It depends almost entirely on scale, which is why practitioners disagree about it. At several hundred entries with consistent labels and full content search, informal naming genuinely works and costs very little. At several thousand, with many similar items, it produces a collection that can only be searched and never scanned, and where similar things cannot be gathered because nothing groups them. The honest position is that informality is a defensible choice with a known breaking point. Experienced users across several unrelated professions converge on the same answer here, which is a reasonable indication that it reflects something about retrieval rather than about any particular kind of work.

Should the convention be written down?

Yes, in the vault itself, as a short entry starred so it appears in the sidebar. Holding a convention in memory works until somebody is tired, in which case a variant appears, and variants are how drift begins. A written list of the controlled vocabulary and the component order takes two minutes to produce and prevents almost all subsequent inconsistency. It also serves as onboarding material if anybody else eventually works with the archive. Practitioners who have made this change describe the improvement as arriving immediately for everything named afterward, which is unusual among organisational adjustments and is why it is worth doing early rather than deferring it.

How often should a naming convention be reviewed?

Rarely, and only when something concrete prompts it. A convention that keeps changing produces exactly the inconsistency it exists to prevent, so stability is itself a virtue. The reasonable triggers are a genuine change in the work, such as taking on a new category of material, or repeated evidence that a particular kind of item cannot be found. Reviewing on a schedule, in the absence of a problem, tends to generate unnecessary revisions. The underlying reason is that naming is applied thousands of times from a decision made once, which is an asymmetry that rewards a modest amount of thinking at the outset far more than any subsequent effort.

What is the single most useful naming change most people could make?

Putting the term they actually search into the title rather than the term that is technically correct. Most people name accurately and unhelpfully, using the formal designation, the document’s own title, or the date, none of which is what they type when looking for something. Identifying the one word they would reach for, and placing it in the highest-weighted field, changes retrieval immediately for everything named afterward. It takes no additional time and requires only noticing what one actually searches for. Practitioners describe this as the adjustment they would make first if starting again, and several identify it as the one thing they wish somebody had told them in their first week rather than discovering after a year of accumulated material.

Want to build your second brain offline?
Try VaultBook and keep your library searchable and under your control.
Get VaultBook free