Why proof is per project
A single global proof page proves nothing to anyone. Here is what I did instead.
A week ago, this site had one proof page. It said: “35 tests, the MQTT chain works, the addresses are read back”. That is true, and it is useless.
The problem
A visitor buying the greenhouse kit does not care about the self-monitoring layer’s tests. A client who wants a self-hosting guide does not care about the humidity sensor’s CRC. Putting all three on one page asks the reader to sort them out — and an argument that needs sorting is not an argument.
The answer
Each project now has its own proof page, right where it is sold. And more importantly: that page is not written, it is computed.
A manifest, store/projets/<slug>/preuve.yml, says where a project’s evidence
comes from: the test file, the command to run, the SQL query, the files to
measure, the images to show, and the limits to declare. Every time the site is
built, the engine runs those commands and injects the result.
What that changes in practice
- If I delete a test, the number shown drops. I cannot forget it.
- If a delivered file disappears, the build fails. The page does not go out with a dead link.
- The recalculation date is displayed. A buyer knows whether the figure is three hours old or three months.
- The limits are a separate block at the bottom of the page. Not a footnote in small print.
What that does not solve
A system that computes its own evidence does not make honesty automatic. I could very well write a manifest that only measures what suits me.
That is why the manifest is public, in the repository, and why every command it runs is visible on the page. You can read what is measured, and above all what is not.
The site’s code is not public yet. It will be when it is clean enough to be reviewed — that is a limit, and it is displayed as one.