The magician's job is to open a channel between Kia and Chaos. Carroll then builds a whole taxonomy on top — the eight "colors" of magic (the octarine system), each ray a domain of operation. But the genuinely load-bearing idea, the one that made chaos magic a break rather than another tradition, is the demotion of belief itself. Every prior Western system asked "which cosmology is true?" — Kabbalah, Thelema, the Golden Dawn's elaborate correspondences. Chaos magic asked "which cosmology is useful for this operation?" and treated the answer as disposable. That's the hinge. Belief stops being a destination and becomes consumable equipment.

And here's a piece I underplayed that you specifically will care about: the role of deliberate absurdity. Chaos magic inherits from Discordianism — the worship of Eris, the Principia Discordia, the Church of the SubGenius and its fake messiah "Bob" — a practice of holding beliefs you're simultaneously laughing at. That looks like mere irreverence, but it's actually a sophisticated debiasing device. If the only beliefs you'll wear are ones you can also mock, you structurally cannot fuse to them, because the joke is the handle. RAW lived this explicitly: he'd adopt a frame, ride it, and keep the self-mockery running the whole time so he never forgot it was a frame. The absurdity isn't decoration on the epistemics — it is the epistemics. It's the cheapest known technology for keeping a belief provisional, and it's the healthiest thing the tradition has.
The first move is to stop letting "belief as a tool" be one thing. It's three, and the tradition smears them together on purpose under Carroll's "metabelief." There's belief as affective access — inhabiting Mars to become confident-aggressive, which is method acting and entirely defensible, because the feedback is immediate and somatic. There's belief as attention-priming — the sigil setting a perceptual filter, defensible if you keep a record, because then it's a testable n=1. And there's belief as assertion of magical causation — "my working bent the probability of the external event" — which is the one that isn't established and the one where all the self-deception nests. The whole analytic contribution is refusing the smear: the first two survive contact with feedback, the third is precisely what feedback would kill, and the tradition's genius/crime is to bundle them so the credibility of the first two launders the third.