My Notes Have Outlived Four Apps

Since 2015 I have written 6,140 notes totaling 1.94 million words. In that time, four applications I used to write them have died, been acquired, ruined themselves with a redesign, or moved to a subscription model that held my own thoughts for ransom. I lost nothing in any of those transitions. Migration time across all four combined: under one hour.

This was not luck. It was one boring decision made in 2015, and today's entry is the argument for making it: the notes are plain text files in folders, and every app I have ever used was merely a viewer.

The decision, stated precisely

A note in my system is a Markdown file with a timestamped name in a folder hierarchy I control, synced between machines by tools that move files without understanding them. The current viewer is excellent. The previous viewers were excellent until they were not. The files do not care. When an app dies, I point a new app at the folder, and the eleven-year archive is simply there, the way furniture is still in the house when you change the locks.

Compare the alternative, which I also lived through briefly and count in the four: a notes app whose data lived in a proprietary sync service. Export produced 214 files of tag soup that took most of my one migration hour to clean. The lesson cost me 40 minutes and some formatting. Others have paid more. Every few years a beloved notes platform shuts down and its users discover that "your data" was a UI label, not a legal or technical description.

Local-first, defined by its measurements

The term of art for the good pattern is local-first software, and it has a checkable definition. Seven properties were proposed in the original 2019 essay by the Ink & Switch research lab, but four measurements capture the substance:

Test	My notes system	Typical cloud notes app
Works with the network cable pulled?	Yes, fully	Partially, at best
Time to first note after total vendor death	0 seconds	Undefined
Export completeness, tested	100% (the files ARE the export)	Varies, often lossy
Cost of switching viewers	~10 minutes	Hours to never

That second row deserves a pause. For a local-first system, vendor death is a non-event you might read about in the news. For a cloud-first system, vendor death is an extinction event with a grace period, typically 30 to 90 days, during which you and every other user simultaneously discover whether the export button was ever load-bearing.

The honest costs

Symmetry demands the downsides, measured. Real-time collaboration: worse. Plain files merged by sync tools occasionally conflict; I see roughly one conflict per 400 edits, each costing a minute to resolve. A cloud service resolves these invisibly. Search across 6,140 files requires a decent local tool; mine indexes the archive in 1.8 seconds, so this stopped being a real cost years ago. Mobile editing is adequate rather than delightful. And there is no one to call, though eleven years of data suggests there was never anyone to call at the cloud services either, only someone to email during the shutdown grace period.

Net assessment: I pay a small, continuous convenience tax and in exchange my written memory has the lifespan of the filesystem, which is to say the lifespan of my diligence about backups, which is audited and published on this site.

The generalization

The principle extends beyond notes, and I apply it as a filter for adopting any software: data outlives applications, so choose the data format first and the application second. Photos: standard formats in folders (viewer: replaceable). Finances: plain-text ledger files (my accounting survives any app's death). This site: Markdown in a Git repository, hostable by anyone, including a laptop in my office if it comes to that.

Ask of every tool that wants your daily writing: if this company vanishes at midnight, what do I have at 00:01? For my notes the answer is: everything, in a format readable by software from 1975 and, at current standards trajectory, 2075.

Confidence that plain text files remain readable in fifty years: 0.97. It is the safest bet in computing. Nothing else in this log comes close.