Running the finished thing, and finding the last missing field

With the writer done, the obvious next step was to drive the whole chain by hand rather than through a test: link a module with the script here, build its vendor segment, wrap it in a container, and read it back.

It came out as an ordinary shared object with a SysV ABI. Every unit test passed; the end-to-end test passed. The reader had been saying so on its own third line the entire time:

type       e_type 0x0003
osabi      NOT FreeBSD - a loader refuses this before anything else

Three header fields no linker sets, because no linker knows about either console, and a loader refuses the file on any of them before it reads a single other byte. They were the last piece of format knowledge still living only in obSCEne. identity::stamp now writes them, reporting each change, and the end-to-end test asserts the module is an executable with the right ABI before it reads anything else. (D028)

Two things worth keeping from how it was found

Unit tests all passed. Each piece was correct about its own job. What was missing was a step nobody owned, and only running the thing end to end and reading the output surfaced it.

The reader already knew. It had a specific, correct diagnostic - written by someone who had clearly been bitten - and the writing side simply never asked it. Which is an argument for principle 4 that is easy to state and hard to remember: a reader and a writer in one crate is worth nothing if only the reader is ever pointed at the output.