Skip to main content

Performance evidence calibration

  • PE1: correct row-major indexing and declared model assumptions; distinguish ISA guarantees from cache/CPU implementation details.
  • PE2: independent expected checksum, a rejected wrong traversal, and restored correctness. Equal timings or mutually matching wrong sums do not pass.
  • PE3: raw repeated samples with alternating order, initialization excluded, warmed paths, explicit process CPU clock, compiler/target and resolution notes. No deletion of inconvenient samples without a documented reason.
  • PE4: actual assembly evidence from this target; do not copy an x86 explanation onto ARM output. Unchanged source does not guarantee unchanged generated instructions.
  • PE5: a conclusion limited to measured cases, spread and alternatives. Dimension changes affect several variables. No required performance ratio or reversal; inconclusive is acceptable when justified.
  • PE6: fresh prediction and honest observed outcome after a delay, with help record. Unknown hardware details stay unknown, and unavailable counters are not invented.

The harness establishes correctness and reproducible collection, not a universal causal performance claim. Manual review is required for the interpretation.