> ## Documentation Index > Fetch the complete documentation index at: https://docs.alphaengine.trade/llms.txt > Use this file to discover all available pages before exploring further. # Trust and Confidential Compute > The assumptions behind confidential evaluation. Confidential compute reduces strategy leakage during evaluation. It does not remove all trust. Evaluate private strategy material inside a restricted runtime. Scores, diagnostics, hashes, artifact references, and eligibility can leave the runtime. Operator configuration, keys, permissions, and side channels still matter. ## What confidential compute does It can let AlphaEngine evaluate private strategy inputs inside a restricted runtime while publishing evidence outside that runtime. That evidence can include: * scores, * diagnostics, * report hashes, * artifact references, * eligibility output. ## What it does not do Confidential compute does not make final public-market effects invisible. It also does not prove that every operator, implementation, or governance choice is correct by itself. ## Primitive choice AlphaEngine should choose the weakest sufficient privacy primitive: Use for private evaluation when attestation is sufficient. Use fully homomorphic computation when encrypted state or encrypted computation is required. Use zero-knowledge proofs when external parties need to verify policy compliance. Use threshold cryptography when no single runtime or operator should see plaintext. ## Operator assumptions The current operator-side modules are internal infrastructure. Public builders should not need to operate them to use the beta API. Trust still depends on: * runtime configuration, * operator permissions, * key handling, * audited adapter boundaries, * typed failure behavior, * reproducible artifacts. ## Related pages The full trust and leakage boundary. How protected inputs become evidence. Where operator modules live.