Run the comparison
Step 1 — assemble the context
Before any reasoning happens, the engine builds the background it is entitled to use. It assembles parallel dimensions — facts (what we know), metaphysics (what is real), and the relevant law — each built identically:
start with an EMPTY file
then concatenate the applicable layers, top-down:
Earth -> country -> state/province -> community -> area -> decision
each block preceded by a header naming the source file it came from
Starting empty is the load-bearing part. Nothing enters the reasoning unless a layer explicitly put it there, and every line traces back to a file you can open.
All dimensions share one jurisdictional scope, so they cannot end up describing different places.
Step 2 — the first-principles constraint
The assembled text goes to the model with one hard rule:
Reason strictly from the provided text, from first principles. No outside knowledge. No appeal to authority.
This is the core rule of the engine and the reason its output is worth anything.
A model answering from its training is an opinion with a large vocabulary and no audit trail. A model reasoning only over text you can read, whose assembly you can inspect, is doing something you can actually check — and disagree with, specifically.
Step 3 — the output
A structured record with a fixed shape: a summary, what is decided, on what basis, and what would change the answer.
That last field matters most. A conclusion that cannot say what would falsify it is not a conclusion.
Step 4 — the run directory
Every run writes its inputs, its assembled dimension files, and its output to a dated directory.
You can open it and see exactly what the model was given. A conclusion nobody can trace back to its inputs is not evidence, and the movement does not get an exemption from the standard it applies to everybody else.
What the check is good at, and what it is not
Good at: noticing that your conclusion is broader than your links support. This is the most common failure in political reasoning and the one people are worst at catching in themselves.
Not for: telling you whether the world is a certain way. It has no access to anything except the text it was handed, on purpose.
Runs are asynchronous — you start one and come back.