Not treating my own README as evidence

Context

When I started writing up the works for this site, the READMEs in those repositories carried claims — encryption, a database-side implementation — that I could have copied across without ever checking them. Every one of them was prose I had written myself.

Options

• Publish the claims a README advertises exactly as they standRejected
• Trace each claim through the code and withdraw whatever cannot be backedChosen

What I chose

I traced each claim through its data path and withdrew the ones that were not in the implementation before publishing.

Why the rest were rejected

Publish the claims a README advertises exactly as they stand

In the repositories I looked at, the feature a README advertised as central was not in the implementation. Publishing as-is would mean claiming unverified work as my own record.

What I learned afterward

The withdrawn claims all had the same shape. None of them were "it turned out not to work"; they were all cases where the prose had run ahead of the implementation. A tunnel advertised as encrypted had no field for a public key anywhere in its shared-code layout — it was not even one wiring change away from working. A daily limit described as implemented in a database view turned out to reference that view from nowhere in the application. The only tools I needed were text search to follow the data path, and reading the struct definitions and the implementation itself.¹