py -3 scripts/bench_stage_results.py \
  --baseline "F:/code-backup/patternia/build/bench/result.json" \
  --current "F:/code/patternia/build/bench/result.json"

This writes:

• bench/results/baseline/result.json
• bench/results/current/result.json

(If you pass a directory, the script auto-appends result.json.)

py -3 scripts/bench_compare.py

Default outputs:

• build/bench/compare/bench_compare.png
• build/bench/compare/bench_compare.md

Use this when one JSON already contains multiple implementations (Patternia / if-else / switch / std::visit) and you want per-scenario comparison.

py -3 scripts/bench_single_report.py `
  --input build/variant_all.json `
  --include "Variant|ProtocolRouter|CommandParser" `
  --outdir build/bench/single `
  --prefix single_impl

Outputs:

• build/bench/single/single_impl.png
• build/bench/single/single_impl.md
• build/bench/single/single_impl.csv

The PNG is a Patternia-focused gap map:

• each scenario is normalized to its fastest implementation (1.00x)
• the Patternia result is the diamond marker
• grey dots are alternative implementations
• rows are sorted by the largest Patternia gap first

The CSV includes filter-ready columns:

• is_patternia
• fastest_impl
• impl_vs_fastest_pct
• impl_vs_patternia_pct
• patternia_rank
• patternia_vs_fastest_pct
• patternia_status

You can also rebuild a report from an existing CSV:

py -3 scripts/bench_single_report.py `
  --input docs/assets/bench/latest.csv `
  --outdir docs/assets/bench `
  --prefix latest

py -3 scripts/bench_compare.py \
  --include "PacketMixedHeavyBind" \
  --label-baseline "backup" \
  --label-current "current" \
  --prefix "packet_heavy"

Variant benchmarks now include an in-code stable profile (ns unit, min time, repetitions, aggregate report). You can run the variant suite directly:

.\build\bench\ptn_bench.exe `
  --benchmark_filter="Variant(Mixed|AltHot)" `
  --benchmark_out=build/bench/result.json `
  --benchmark_out_format=json

py -3 scripts/bench_compare.py \
  --include "Variant(Mixed|AltHot)" \
  --fail-if-regress-pct 3.0 \
  --fail-if-mean-regress-pct 1.5

• --fail-if-regress-pct: fail when any selected benchmark regresses beyond threshold.
• --fail-if-mean-regress-pct: fail when mean regression across selected benchmarks exceeds threshold.

py -3 -m pip install matplotlib