.gitperfconfig

[measurement."report-size"]
epoch = "fedee674"
min_relative_deviation = 10.0
unit = "bytes"
min_measurements = 10

[measurement."report"]
epoch = "fedee674"
unit = "ns"
min_measurements = 10

[measurement."test-measure2"]
epoch = "439374ca"
min_relative_deviation = 10.0
unit = "ns"

[measurement."release-binary-size"]
epoch = "3a1aff1"
unit = "bytes"
min_relative_deviation = 10.0
min_measurements = 10

[measurement."report-benchmark"]
epoch = "296bc41"
unit = "ns"
sigma = 3
aggregate_by = "min"
min_measurements = 10

[measurement."bench::add_measurements/add_measurement/50::median"]
epoch = "4493382"
unit = "ns"
dispersion_method = "mad"
sigma = 3.5
aggregate_by = "min"
min_measurements = 3
# Sequential cross-runner study (10 instances, max-parallel=1): CoV=8.0%, MAD%=4.6%.
# Parallel study gave 34.8% CoV due to runner resource contention — sequential is accurate.
# min_relative_deviation=12 (1.5× CoV), max_cov=16 (2× CoV) per study recommendation.
min_relative_deviation = 12
max_cov = 16

[measurement."add-benchmark"]
epoch = "296bc41"
unit = "ns"
dispersion_method = "mad"
sigma = 3.5
aggregate_by = "min"
min_measurements = 10
# MAD converges to ~228µs as measurements stabilize, making the ±3.5σ threshold
# only ±800µs — too tight for CI runner variability. Even when the z-score
# exceeds 3.5σ, the failure is suppressed if the actual head deviation is
# below 10% of the tail median.
min_relative_deviation = 10.0

[measurement."report-size-benchmark"]
epoch = "fe2d91f5"
unit = "bytes"
min_relative_deviation = 10.0
sigma = 3.5
aggregate_by = "median"
min_measurements = 10

[measurement."bench::report/report_generation/10::median"]
unit = "ns"
dispersion_method = "mad"
sigma = 3.5
aggregate_by = "median"
min_measurements = 5
min_relative_deviation = 10.0