The 10 AI surfaces — Surface-Bound AI Principle.

A chat box is the worst possible AI interface for a trader. It lets you keep asking until you get an answer that flatters the trade you already wanted to take. Every "AI trading" product in 2026 ships a chat box because chat boxes feel useful and are incredibly hard to enforce discipline against. The Swing Deck framework explicitly doesn't ship one. Instead, it ships 10 surface-bound AI surfaces: each fires on a specific framework moment, produces structured output, and never engages in open-ended dialogue. The trader doesn't summon them; the framework fires them when the trigger condition matches. That single architectural choice defines the difference between an AI that talks you out of bad trades and one that helps you talk yourself into them.

The Surface-Bound AI Principle

Three rules:

1. Each AI surface has one trigger event. "User clicks Audit card" or "Position approaches earnings" or "Drawdown crosses 7%." Surfaces don't fire on user query — they fire on framework state changes.
2. Each surface has a structured I/O contract. Inputs are specific framework signals; outputs are constrained to predefined shapes (audit verdict, narration sentence, refusal reason). No free-form generation.
3. No conversation memory. Each surface fires fresh; previous outputs don't influence the next. The trader can't iteratively re-prompt to get a better answer.

The principle's net effect: AI helps you refuse trades you'd otherwise rationalize. A chat box helps you rationalize them.

The 10 surfaces

Why this beats a chat box

Chat boxes have three failure modes the framework's design specifically avoids:

• Re-prompting. Ask "should I buy NVDA?" The chat says no. Ask again, slightly differently. Iterate until the answer is yes. With surface-bound AI, there's no "ask" — the audit returns its verdict and that's the verdict. No second opinion to fish for.
• Free-form rationalization. Chat boxes will fluently explain why a bad trade is good if you frame the question that way. Surface-bound outputs are constrained to predefined shapes — they can't generate "compelling case for the trade you already wanted."
• Hidden state. A chat box's tone and framing depend on conversation history. Surfaces have no memory. The audit reads the same name in the same regime and returns the same verdict whether you asked yesterday or never asked.

The architectural commitment

This isn't an interface choice; it's a discipline choice baked into the product. The framework could ship a chat box tomorrow — the LLM infrastructure is the same. It explicitly doesn't, because the failure mode is documented and the alternative (surface-bound) actually works. The brand contract: no chat boxes in the dashboard, ever. Every release reaffirms it.

The cost: users who expect a chat box are confused. The benefit: users who internalize the architecture stop expecting one. The framework's approach to AI is the same as its approach to refusal — make the discipline structural, not optional.

What you can do that's chat-shaped

One concession: a "What's this surface telling me?" hover tooltip on each surface card. The tooltip explains the verdict in plain English (e.g., "Pillar 5 refused — R:R 1.7:1 is below the 2.0:1 floor"). It's not a conversation; it's static documentation surfaced contextually. The trader gets clarity without the failure modes of free-form chat.

The real lesson

Chat boxes let you talk yourself into trades. Surface-bound AI talks you out of them by firing on specific triggers, returning structured output, and never engaging in open-ended dialogue. The 10 surfaces cover every framework moment that warrants AI assistance — audits, refusals, ritual prompts, drawdown narration, catalyst playbooks. Together they replace what most retail products call a "trading copilot," and they replace it with something that consistently makes the harder discipline easier and the rationalization shortcuts impossible.

Related: L32 — pre-flight chain · Standalone — why no chat box