Performance And Scaling

Performance and scaling changes must preserve DAG correctness guarantees and not blur replay or diff attribution.

Visual Summary

flowchart LR
    baseline[collect baseline metrics] --> optimize[apply optimization]
    optimize --> validate[validate run correctness]
    validate --> replay[replay and diff check]
    replay --> adopt[adopt or revert]

Performance Levers

• execution parallelism and scheduling strategy
• artifact write/read batching and backend tuning
• graph partitioning and incremental execution hints

Required Guardrails

• track latency and throughput with fixed benchmark graphs
• verify output equivalence and replay fidelity after tuning
• reject optimization changes that hide classification precision

Code Anchors

• crates/bijux-dag-runtime/src/engine/
• crates/bijux-dag-runtime/src/replay/
• crates/bijux-dag-app/tests/replay_diff_hardening_contract.rs

Next Reads

• Invariants
• Change Validation
• Common Workflows