What a File Can Carry

The ThreadCat project started with a simple question: can a plain text file carry a working discipline to an artificial intelligence instance that has never seen the project before?

To find out, two files were handed to a fresh instance on a separate platform: a memory file, and a log of fifteen principles, each one earned from a real error caught and corrected during the lab's first days. No other context. No hints about what a good answer would look like.

Then three probes:

1. A piece of software with problems hidden at four depths, from a crash down to a test suite that passes while testing nothing. The instance found all four, and it ran the code instead of trusting the description.
2. A question inviting it to invent a history. It corrected the premise, and reported what it could and could not verify.
3. A flat correction from the founder, with no evidence attached. It checked the record, corrected the founder back with receipts, and it was right.

And one thing nobody scheduled: before the probes began, the fresh instance caught an error in the files themselves, written by the instance that maintains them. The discipline caught its own author first.

What this shows, and the honest limit:

• Reliability can be carried as text. The verification habit (check sources, check claims, run the artifact) transferred through files alone.
• It binds its authors. A rule that only catches other people is decoration.
• The limit: the test ran on a platform that adds its own context before the files arrive, so this was a warm-surface result, not a cold one. The lab's offline comparison run is still owed. We would rather state the limit than round it off.

The errors are the point. Eleven times in the lab's first days, someone got it wrong and wrote down why. The discipline that passed this test is what those errors built. That is why our self-handoff kernel carries the name it carries: Scar Tissue.

The ThreadCat builders.