Report Detail Levels

NBenchmark supports three report detail levels that control how much statistical information reporters display. Simple is the default; Standard adds the full multi-section output; Advanced adds a per-benchmark stats block with the full distribution summary.

Simple mode (default)

Simple mode shows a compact table with the essential information an average developer needs to know whether their code performs well or how it compares to other implementations:

Column	Description
Benchmark	Benchmark name.
Median	Median timing.
Ops/s	Mean operations per second (`1e9 / Mean` when timing is in nanoseconds).
Ratio	Visual bar plus ratio relative to the baseline.
Sig	✓ = significant, ✗ = not significant, - = not applicable.
Alloc/op	Mean bytes allocated per iteration, or - if not measured.

A one-line footer shows the benchmark count, total duration, and confidence level. No statistical jargon, no auxiliary tables.

Standard mode

Standard mode shows the same comparison table with additional columns (Mean, Mag, Description) plus several auxiliary sections:

Precision & Tail Latency table: Error (±CI), StdDev, CV, and upper-tail percentiles (P95, P99, etc.).
Diagnostics table (when diagnostics are enabled): GC Gen0/Gen1/Gen2 collection counts, heap info, CPU/wall ratio, and exceptions per op. See Diagnostics.
Launch Aggregation table (when LaunchCount > 1): cross-launch mean, stddev, median, and CI.
Interpretation block: omnibus verdict, significance test name, outlier detector, effect metric summary, and measurement profile.
Auto-tune summary lines: resolved warmup, sample count, ops-per-sample, and achieved CI half-width.
Warnings (when present).

This is the level for practitioners who want to understand variability and the statistical rigour behind the results.

Advanced mode

Advanced mode shows everything in Standard plus a per-benchmark stats block. The console reporter prints each stats block below its row; the Markdown reporter emits a dedicated details section after the table. The stats block includes:

Outliers: count of removed samples and the trimming method.
Range: Min to Max spread.
Quartiles: Q1, Q3, and IQR.
Fences: Lower and upper fences (only for IqrFence mode).
Iterations: pre-trim and post-trim sample counts and warmup count.
Confidence interval: full CI bounds and margin percent of mean.
CV: coefficient of variation as a percentage.
Skewness and Kurtosis: shape of the distribution.
MAD: median absolute deviation (scaled).
Percentiles: the full set of configured percentile values (e.g. P50, P95, P99, P99.9, Max).
N: post-trim sample count.
Allocation breakdown (when MeasureAllocations = true): median, P95, and max allocation per iteration.
Diagnostics breakdown (when diagnostics are enabled): GC collection counts, heap committed and fragmented bytes, CPU time and CPU/wall ratio, and exceptions per operation.

Setting the detail level

`WithDetail()` (Harness and Suite modes)

Both BenchmarkHarness and BenchmarkSuite expose a WithDetail(ReportDetail) method. The detail level is stamped onto all registered reporters, so calling WithDetail before or after WithReporter works in either order.

csharp

// Harness mode
var host = BenchmarkHarness.Create(args);
host.WithDetail(ReportDetail.Advanced)
    .WithReporter(new ConsoleReporter())
    .RunAsync();

// Suite mode
var suite = new BenchmarkSuite("MySuite");
suite.WithDetail(ReportDetail.Standard)
     .WithReporter(new ConsoleReporter())
     .RunAsync();

`--detail` flag (Harness mode)

bash

dotnet run -- --detail advanced
dotnet run -- --detail standard
dotnet run -- --detail simple

Value	Behaviour
`simple`	Compact table with the essential statistics. (default)
`standard`	Full comparison table plus Precision & Tail Latency, auto-tune, and Interpretation sections.
`advanced`	Same as standard plus a per-benchmark stats block with quartiles, fences, confidence interval, skewness, kurtosis, MAD, configured percentiles, and allocation breakdown.

The --detail flag affects all registered reporters. JSON always emits the full record regardless of detail level.

Single mode

Single mode (Benchmark.Run / Benchmark.RunAsync) always uses Simple detail and does not support WithDetail().

Reporter behaviour

Reporter	Simple	Standard	Advanced
Console	6-column table + counts footer	Full table + Precision & Tail Latency + Diagnostics + Interpretation + auto-tune	Standard + per-benchmark stats block (incl. diagnostics breakdown)
Markdown	6-column table + counts footer	Full table + Precision & Tail Latency + Diagnostics + Interpretation	Standard + dedicated details section (incl. diagnostics breakdown)
CSV	12 core columns (incl. GC counts)	25 core columns (incl. GC counts)	51 columns including quartiles, fences, shape stats, and full diagnostics
JSON	Full record (always)	Full record (always)	Full record (always)

Report Detail Levels ​

Simple mode (default) ​

Standard mode ​

Advanced mode ​

Setting the detail level ​

WithDetail() (Harness and Suite modes) ​

--detail flag (Harness mode) ​

Single mode ​

Reporter behaviour ​

See also ​