Deciding not to use the system I built — an update on the photo-consent system

2026-06-11 ／ Vol 18 ／ draft at the time of publishing

In Vol 5, "Building "a system that listens properly" with Claude", and Vol 6, "going one layer deeper into the photo consent system", I wrote about building a system that asks each person, individually, for consent before any photo from karaha.org's activities gets published. Google Sheets, Apps Script, and Cloudflare Pages combined; a working system in 2 days, refined the same day, and taken as far as a test run with core members.

Today is the follow-up. Conclusion first: I decided not to use this system (2026-05-31).

I tested as planned — and got an answer I hadn't planned for

The original plan was simple. Try it with core members first. If nothing goes wrong, roll it out to all participants.

The test itself went exactly as planned. A personal URL arrives by email; you choose a usage scope for each photo and submit. The system from Vol 5 worked as designed. This was not a shutdown over a technical failure.

What actually trying it revealed was something separate from how well the system worked. This system does not fit this community. That was the answer the test run handed me.

Why it doesn't fit

What karaha.org runs is a gathering place where people involved in mental health — peers, supporters, families, neighbors — meet on flat ground, beyond their roles. Not drawing a line between "those who support" and "those who are supported," being able to show up without bracing yourself: that is what the place values most.

I won't write here what actually happened during the test. As I drew the line back in Vol 5, those records belong to the people who took part. What I can write is only the outline of my own judgment as the operator.

Asking someone to pick a usage scope for every single photo is, as a system, precise and honest — I still think so. But placed inside the air of this particular gathering, the weight of the procedure stood out before the care did. When I designed it, I believed "being able to choose is respect." Looking back, I had only seen the burden of choosing from the screen's side of the glass. What the consent system was trying to protect and what this place has been quietly growing were not pointing in the same direction. That was invisible in the code. It only became visible by trying it together with the people of the place.

Fortunately, this was the lightest possible stage to stop at. The test collected only a small amount of data from core members; no data from outside participants ever entered the system. Had I stopped after full rollout, withdrawal requests and inquiries would have been left hanging. Designing the rollout as "try small before going wide" turned out to be the right call in itself.

Using principle no. 3 for real, for the first time

The third of this site's three operating principles is "the courage to overrule your past self (when a policy changes, that change itself becomes an article)." I wrote those words on launch day, but I didn't expect such an unmistakable occasion to use them.

Vol 5 and Vol 6 were articles I put real effort into — the design, the pitfalls, the operating rules. Having their sequel be "we're not using it" takes a bit of courage on the writing side. But leaving those articles frozen as-is after the policy changed would be far more dishonest for a build-in-public site. So I'm writing this.

I don't think the time spent building was wasted. Because I tried it, "it doesn't fit" became a conviction instead of a guess. If I had decided in my head alone — "it probably won't fit, let's not" — I'd likely still be dragging around a "but wouldn't it have been nice." To let it go, I first had to give it a shape.

Honestly: the cleanup isn't done yet

To be honest about it, what I decided is only "we won't use it." The teardown work hasn't been touched.

Three things remain. Rewriting the public photo policy so it no longer routes people to the consent page, consolidating everything into a single email contact. Taking down the consent and registration pages themselves. Deleting the test data and shutting off the backend deployment.

"Deciding not to use it" and "cleaning it up" are separate tasks. The decision landed at the end of May; the cleanup needs hands-on hours. It will happen later. Writing it down here is how I assign myself the homework.

When building gets cheap, "build first, decide later" becomes an option

The one lesson I most want to carry out of this:

As AI drives down the cost of building, "build it, then judge it" becomes a realistic option.

Before, a system of this size meant outsourcing it or spending weeks building it myself. After pouring that much in, "it doesn't fit, let's stop" is very hard to say. The size of the sunk cost bends the judgment itself.

This time was different. Built with Claude in 2 days, tested, answered in under a month. Because the sunk cost is small, I can say "it doesn't fit" exactly as it is. Rather than debating everything to exhaustion before building, trying the real thing and then deciding was faster — and more certain.

The homework of "asking properly" for consent has not disappeared. I'm letting go of the system; the question stays. I'll keep looking for a way that fits this place. And that, too, will probably become an article someday.