Egregora ABX Judge

This is the smallest node in the Egregora pack and arguably the most important discipline it enforces. It takes two things - the abx_meta DICT produced by ABX Prepare (which secretly knows whether the mystery sample X was really A or B) and your guess - and it tells you, flatly, whether you were right. Two inputs, one output, zero audio. It's the "tell me the truth" node that closes the loop on a blind listening test.

You can do this without the node, of course: peek at the abx_meta and compare. But that's exactly the failure mode this exists to prevent. The point of a double-blind ABX test is that the person judging doesn't see the answer until after the guess is locked in. Running your guess through this node means the judgment and the reveal are separate steps - and if you're honest about not peeking, you get a result your brain can't retroactively rationalize.

How it works

ABX Prepare records the truth in abx_meta as x_is: "A" or "B", along with the seed. This node reads that, compares it to your guess (a dropdown with just A and B - no cop-outs), and returns an abx_result DICT containing the actual x_is, your guess, and a correct boolean.

The workflow

ABX Prepare → (audio_A, audio_B, audio_X → preview)
            → abx_meta → ABX Judge → abx_result

Listen blind to A, B, and X. Commit to a guess. Wire it in, run, and read abx_result. Repeat across several trials with different seeds, and count your hits.

The inputs that matter

• abx_meta - the DICT from ABX Prepare. Note the tradeoff baked into the pack's design: the truth travels inside this dict, so the "blindness" is a social contract between you and yourself. If you open the meta before guessing, you've just run an open test and learned nothing.
• guess - A or B.

Output is a single abx_result DICT.

The take

A few honest notes on interpreting it. One correct guess is luck; a string of correct guesses across different seeds is evidence. And remember the null-test rule still applies upstream: gain-match A and B before you even build the trial, or you're testing loudness, not quality. Used properly, this node answers the question the whole pack keeps circling: after all the FlashSR and Fat Llama processing, can you actually hear the difference - or did you just want to?