← Research Labs

Thirty-five had no repository

One hundred and seven projects, counted on my own machine. For thirty-five of them the live deployment was the only copy in existence. Two were deleted before anyone noticed — and the ones that died were not the hard ones.

This one is not a literature review. It is an audit of a single machine — mine — and the reason it belongs on a site about evidence is that almost nobody publishes this data about themselves, which is exactly the complaint the missions review makes about everyone else.

Every project counted below was built, deployed, and in most cases quietly abandoned. The counts are from my own tracker, 13 May 2026, and my own filesystem, 6 September 2026.

1. The shape of a personal portfolio

Tracker, 13 May 2026 — every project ever deployed:

Projects built and tracked
107
With no repository at all
35
Recovered by crawling the live site
33
Lost permanently
2

Filesystem, 6 September 2026 — folders on disk, a different count four months later. These are not a subset of the 107 above and the two totals are not meant to reconcile:

Archived — not opened this year
57
Still active
31

Thirty-one active, fifty-seven archived. The archived ones are not failures in any dramatic sense. They are folders I have not opened this year and most of which I will not open again.

2. Thirty-five had no repository

For 35 of 107 projects the live deployment was the only copy that existed anywhere. Not a backup. Not a stale branch on a laptop. The only copy.

Recovery, once the problem was visible: 33 recovered by crawling the live sites for their built output. Two were deleted before the recovery ran and are gone permanently — no repository, no local source, no archive snapshot.

Those two are the entire point of this page. They did not die of a technical failure. They died because a routine cleanup removed the only copy, and nobody had checked whether a copy existed elsewhere before pressing delete.

3. The mortality did not fall where I expected

I assumed the casualties would be the hard projects — the ones with databases, authentication, real complexity, real maintenance burden. That is not what happened.

Mortality fell almost entirely on the projects whose only content was a name. Brand hubs. Landing pages. Placeholders staked out for an idea that never got built behind them. The complex projects survived, because complexity meant someone had invested enough to keep a repository.

Nothing in the portfolio died of a technical cause. Not one. They died of abandonment, and the ones that abandoned fastest were the ones that had been cheapest to make.

4. One worked example

Mission Commons, archived May 2026, still online as of this writing. In January 2026 it catalogued 131 open-source tools for ministries and churches, on the premise that subscription software quietly consumes budgets that should go elsewhere, and that an organisation ought to own its own data.

It did not die of a bad idea. The argument still holds — it is the argument of an entire adjacent site. It died because a directory needs continuous curation to stay true, and curation is exactly the labour that nobody schedules.

What this is evidence for

  1. Cheap to build is not cheap to keep. The cost of a project is not the day you make it. It is the attention it requires afterward, and a landing page requires roughly as much of that as a real application while returning far less.
  2. Deployment is not storage. A third of a portfolio existing only as a live deploy is a single accident away from permanent loss, and the accident is usually a tidy-up, not a disaster.
  3. Count your own work before theorising about anyone else's. This entire page exists because the count was possible and nobody had done it. Most organisations could run the same audit in an afternoon and would rather not know.

Limits

n = 1. This is one person's portfolio on one machine. It establishes that the pattern is real in at least one case; it establishes nothing about base rates, and it should not be generalised to organisations, teams, or anyone whose incentives differ from a solo builder's.

The two counts come from different dates — the tracker on 13 May 2026 (107 projects, 35 orphaned) and the filesystem on 6 September 2026 (31 active, 57 archived). They measure overlapping but not identical things, and the totals are not meant to reconcile exactly.

"Abandoned" is a judgement, not a measurement. It means I have not opened the folder this year. Some of those projects are finished rather than dead, and the distinction is not visible in the count.

Sources

  1. Project tracker, 13 May 2026 — 107 projects deployed, 35 without a repository. Private spreadsheet; counts reproduced above in full.
  2. Filesystem audit, 6 September 2026 — ~/Projects (31) and ~/Projects/_archive (57), counted directly.
  3. Recovery run, 7 September 2026 — 33 orphaned deployments crawled for built output into a private archive repository; 2 had already been deleted.