Ten essays on personal information architecture covering taxonomy, repository management, structural modeling, and AI-driven synthesis—for the technically literate professional who has outgrown the folder.
the-architected-self.pdf (1.9 MB)
14 Likes
Thanks for the effort and sharing this information. 
2 Likes
I’ll be happy if there are any tidbits that will be useful to others.
I’ll study this in more detail, but noticed you identified 3 tools
“DEVONthink as the document repository, ArchiMate as the structural model, Aeon Timeline as the temporal mirror”
For my use, I value Devonthink more than just the document repository
and within DT I implemented my version of the structural model and the temporal mirror
Thank you for sharing this, which I really enjoyed reading!
The point about relational information (typed relationships) I found most interesting. Naturally (in this forum), one wonders whether there is a way to encode typed relationships in DT as well.
Below is the mind map or outline that I created while reading your substack posts. I am always interested in ideas for tagging, naming and organizing things
3 Likes
Since sharing is caring, here other elements of my organizational structure:
I added your typed relationships (serving, realization, triggering, composition, assignment), although I am not yet sure how to make use of that. I have used “target” and “action” types.
I am not yet using any linking but find that the most interesting aspect of your approach and what DT can actually offer. I have a database/folder, group, file structure that is hierarchical only, so far, and look for ideas to make use of linking, in the sense of what you desribe with the architecture model.
2 Likes
For completeness, here is my database/folder and group structure and how I name things. Subgroups start their name with the two letters of their parent. I have opened/shown here only a few examples.
Taxes are not under responsibilities and finances since this is shared with my tax advisor.
3 Likes
I use Devonthink tags; minimal groups/folders
No hierarchy, but it is indicated in the tag names
1 Like
Excellent diagrams! Most informative.
1 Like
My “temporal mirror” in Aeon Timeline is just a reflection and visualization of the dispositive data in DEVONthink. The entire Aeon Timeline database could disappear, and it would be easy to reconstruct it. Not so with the Archimate model – the relationships among structural elements there exist only within the model, not in DEVONthink. Archimate is a UML-derived language with a grammar, and the model is “written” in that language.
Always good to meet a fellow shellback.
I coordinate a monthly “curated” bluegrass jam session at the local Fleet Reserve Association with experienced musicians there by invitation – a band that’s not a band. We call ourselves the Shellbacks because we’re all experienced, and we’ve all crossed one line or another in our lives. I, myself, am a shellback from the good old days when you crawled the length of the deck in your underwear, and they whipped you with shillelaghs made of pieces of fire hose.
2 Likes
Pray tell how you are using structural & temporal mirrors within DT4. It would be interesting to do this and have another way of looking at my data
Thanks. Ian
The project described in my essays started for me as an effort to normalize a multi-terabyte body of data, both professional and personal, accumulated over many years and many workstations. The highest priority, though, was reducing the friction and admin time required to deal with incoming data day-to-day, so it was like fixing the car while you’re driving down the highway.
My most recent long-term data management solution outside of filesystem folders was Evernote, which billed not by size but by daily ingestion, so I would just toss incoming documents into the Evernote bucket and retrieve with searches. Meanwhile, I had historical data and music in folders, and ebooks in Calibre, a highly-regarded ebook manager. These products survived my move from a Windows to a Mac environment.
Having used many data management systems over the years I wanted to ensure that my data was not a slave to proprietary storage and formats that were dependent on a single vendor – I’ve seen them come and go. Evernote’s precarious status a couple of years ago triggered a serious effort to enforce this requirement.
I had been using DT3 for just a couple of years; once I was comfortable with it I moved all the historical Evernote data into DT databases. It took quite a bit of experimentation to see what worked and what didn’t. The steepest learning curve was determining what data worked well in DT and what did not, and the second challenge was learning when to use separate databases.
I’ve settled on an SSD-based working database for daily work not to exceed 250GB containing financial and legal records that is my baseline always-open database. DT is not as strong in dealing with media files, so I keep most audio and video files in the filesystem indexed by NeoFinder. DT, though, works well for my music files, which I store in a 3TB DT database on a 4TB SSD. As an amateur musician, I can replicate a song from albums, recordings in the field, and backing tracks into a single group, making it easy to listen to variations in performance and to practice.
DT’s heavy lift, though, comes from the working database. After importing and de-duplicating my working data, I set up logical groups as most folks do: finance, legal, travel, medical, and so on. That was the state of my affairs at the beginning of my investigation into personal data management, and the structures and processes I describe in the essays are an overlay.
The primary requirement was to reduce friction in day-to-day work. Today I do a daily journal in Evernote, which is easily accessible from its workstation app, its phone app, and the web interface. I move these periodically into DT.
In DT now, I have ::structural, ::organizational, ::facility, ::actor, and ::temporal data bucket groups which contain only subgroups, not individual artifacts. A structural subgroup represents a “thing” like like “Apple Mac Studio 2025.” An organizational subgroups has that organization’s name, as does a facility subgroup. An actor subgroup is named after a person, such as “Smith, John.” These subgroups contain artifacts or replicants of artifacts, like “Bank of America 20260501 Statement.”
Temporal subgroups are named beginning with a date or date range, such as “20250602-20250609 Alaska Cruise” and contain artifacts pertaining to that event.
What this provides is an easy way to file incoming artifacts and access them later. I can go to the ::organizational bucket, containing over 1000 subgroup organizations, and quickly find a particular organization with a couple of keystrokes and move the artifact there. Or, I can go to the ::temporal bucket and file the artifact there if appropriate, like a scan of a playbill from a show or a set list from a music performance.
Temporal events can be replicated into structural or organizational subgroups if needed. An event like “202600523 Replaced washing machine seal” exists in the “A-1 Appliances” organizational subgroup representing the vendor with a replicant in the “LG Washing Machine” structural subgroup.
Once the structure is set up it enforces good data hygiene if used. I can scan an item with my ScanSnap scanner directly into DT’s Inbox group, name it “SDG&E 20260515 Statement,” and move it into the SDG&E organizational subgroup in 30-60 seconds. I can take an airline trip itinerary coming in via email, save it to my Downloads folder, and move it into the DT ::temporal bucket subgroup for that trip, replicating it to that airline’s ::organizational bucket subgroup, and have either the trip information, containing multiple vendors’ artifacts or a vendor’s information, containing records of interactions with that vendor, easily sorted and accessible in the future.
This is a long answer, but it’s a boots-on-the-ground description of how I use the architecture I describe in the essays. The modeling is another layer on top of that, but that is built on the structures and process I describe here.
1 Like
Structural
I use tags; minimal folders
and hyperlinks within notes
Devonthink has a graph view Inspector
Temporal (date/time)
I use DT’s tags and custom metadata
For example Creation Date; adjusted as required
I also added two custom metadata
. Due Date (date format)
. Timeline Date (decimal number format)
For a simple gantt timeline view; a script exports my task data to a spreadsheet
1 Like
thanks Robin - so much stuff in common and I really enjoyed working through your opus!
I’m a folder sort of guy, but will complement it with more use of tags and smart groups ::tem etc
I know that all my files etc are in DT but the trick is to find different ways of viewing them. For this reason, I set up links from books and articles in DT and place them in Craft - which has a pleasant interface and I can find my books in there - the book then opens In DT for reading.
1 Like
thanks - I have not really explored the graph view inspector in tools - it looks very interesting.
I might take things further with Aeon timeline - fun to explore
Now to also go down a rabbit hole with Calibre!
Manually?
Yes. I automate a number of processes with scripts and now the MCP server, but many artifacts are not as well-behaved or named. This manual curation is useful to understand what I’m ingesting.
This is a big reason I don’t use autopayment methods for my bills
1 Like