Who decided on the half-width space between Japanese text and alphanumerics? Tracing it back, I found an AI had decided it with no grounds

Summary

I took stock of ten guideline documents covering prose and UI wording, and went through all 412 normative rules in them. Of those, 147 rules, or about 36 percent, do not say on whose instruction they were decided, and I judged 49 of them likely to be misapplying English-language writing conventions to Japanese text.

It started with a single question: who decided the rule that inserts a half-width space between Japanese text and alphanumerics, and on what grounds? When I looked, the rule was not one I had decided. A generative AI had taken a convention from its training data and settled it into the guidelines without writing down any grounds. So I withdrew the rule, deleted 2,109 spaces from 38 existing articles in one pass, and moved the spacing between Japanese text and alphanumerics to automatic control by the CSS property text-autospace.

In the body I record how it went from three angles: how web standards handle this spacing, how I designed the bulk replacement script to be safe, and how I audited the grounds behind the rules.

What you can take away

This article is for engineers who draft and revise operating rules together with AI, and it explains four things.

• Finding and withdrawing a rule a generative AI settled in on its own: the procedure, starting from one question, for identifying a rule whose author is unknown and withdrawing it
• Converting text in bulk safely: to replace 2,109 places without an accident, make dry run the default behavior and verify zero unintended differences and idempotence across every file
• Controlling the spacing between Japanese text and alphanumerics with a web standard: checking browser support for the CSS property text-autospace as of August 2026, and where it sits in JLREQ, the W3C requirements for Japanese text layout
• Auditing the grounds behind rules: attaching to all 412 rules a record of who decided what, when, and on what grounds, so the whole set can be audited again later

This article is a practical record based on the primary records of this site’s own operation and rework.