correction + legend placement

crates/harness/plot/Cargo.toml

@@ -7,10 +7,9 @@ publish = false
 [dependencies]
 tlsn-harness-core = { workspace = true }
 # tlsn-server-fixture = { workspace = true }
-charming = { version = "0.5.1", features = ["ssr"] }
-csv = "1.3.0"
+charming = { version = "0.6.0", features = ["ssr"] }
 clap = { workspace = true, features = ["derive", "env"] }
-itertools = "0.14.0"
+polars = { version = "0.44", features = ["csv", "lazy"] }
 toml = { workspace = true }
 
 

crates/harness/plot/README.md

@@ -0,0 +1,111 @@
+# TLSNotary Benchmark Plot Tool
+
+Generates interactive HTML and SVG plots from TLSNotary benchmark results. Supports comparing multiple benchmark runs (e.g., before/after optimization, native vs browser).
+
+## Usage
+
+```bash
+tlsn-harness-plot <TOML> <CSV>... [OPTIONS]
+```
+
+### Arguments
+
+- `<TOML>` - Path to Bench.toml file defining benchmark structure
+- `<CSV>...` - One or more CSV files with benchmark results
+
+### Options
+
+- `-l, --labels <LABEL>...` - Labels for each dataset (optional)
+  - If omitted, datasets are labeled "Dataset 1", "Dataset 2", etc.
+  - Number of labels must match number of CSV files
+- `--min-max-band` - Add min/max bands to plots showing variance
+- `-h, --help` - Print help information
+
+## Examples
+
+### Single Dataset
+
+```bash
+tlsn-harness-plot bench.toml results.csv
+```
+
+Generates plots from a single benchmark run.
+
+### Compare Two Runs
+
+```bash
+tlsn-harness-plot bench.toml before.csv after.csv \
+  --labels "Before Optimization" "After Optimization"
+```
+
+Overlays two datasets to compare performance improvements.
+
+### Multiple Datasets
+
+```bash
+tlsn-harness-plot bench.toml native.csv browser.csv wasm.csv \
+  --labels "Native" "Browser" "WASM"
+```
+
+Compare three different runtime environments.
+
+### With Min/Max Bands
+
+```bash
+tlsn-harness-plot bench.toml run1.csv run2.csv \
+  --labels "Config A" "Config B" \
+  --min-max-band
+```
+
+Shows variance ranges for each dataset.
+
+## Output Files
+
+The tool generates two files per benchmark group:
+
+- `<output>.html` - Interactive HTML chart (zoomable, hoverable)
+- `<output>.svg` - Static SVG image for documentation
+
+Default output filenames:
+- `runtime_vs_bandwidth.{html,svg}` - When `protocol_latency` is defined in group
+- `runtime_vs_latency.{html,svg}` - When `bandwidth` is defined in group
+
+## Plot Format
+
+Each dataset displays:
+- **Solid line** - Total runtime (preprocessing + online phase)
+- **Dashed line** - Online phase only
+- **Shaded area** (optional) - Min/max variance bands
+
+Different datasets automatically use distinct colors for easy comparison.
+
+## CSV Format
+
+Expected columns in each CSV file:
+- `group` - Benchmark group name (must match TOML)
+- `bandwidth` - Network bandwidth in Kbps (for bandwidth plots)
+- `latency` - Network latency in ms (for latency plots)
+- `time_preprocess` - Preprocessing time in ms
+- `time_online` - Online phase time in ms
+- `time_total` - Total runtime in ms
+
+## TOML Format
+
+The benchmark TOML file defines groups with either:
+
+```toml
+[[group]]
+name = "my_benchmark"
+protocol_latency = 50  # Fixed latency for bandwidth plots
+# OR
+bandwidth = 10000      # Fixed bandwidth for latency plots
+```
+
+All datasets must use the same TOML file to ensure consistent benchmark structure.
+
+## Tips
+
+- Use descriptive labels to make plots self-documenting
+- Keep CSV files from the same benchmark configuration for valid comparisons
+- Min/max bands are useful for showing stability but can clutter plots with many datasets
+- Interactive HTML plots support zooming and hovering for detailed values

crates/harness/plot/bin/plot.rs

@@ -1,17 +1,18 @@
 use std::f32;
 
 use charming::{
-    Chart, HtmlRenderer,
+    Chart, HtmlRenderer, ImageRenderer,
     component::{Axis, Legend, Title},
-    element::{AreaStyle, LineStyle, NameLocation, Orient, TextStyle, Tooltip, Trigger},
+    element::{
+        AreaStyle, ItemStyle, LineStyle, LineStyleType, NameLocation, Orient, TextStyle, Tooltip,
+        Trigger,
+    },
     series::Line,
     theme::Theme,
 };
 use clap::Parser;
-use harness_core::bench::{BenchItems, Measurement};
-use itertools::Itertools;
-
-const THEME: Theme = Theme::Default;
+use harness_core::bench::BenchItems;
+use polars::prelude::*;
 
 #[derive(Parser, Debug)]
 #[command(author, version, about)]
@@ -19,72 +20,131 @@ struct Cli {
     /// Path to the Bench.toml file with benchmark spec
     toml: String,
 
-    /// Path to the CSV file with benchmark results
-    csv: String,
+    /// Paths to CSV files with benchmark results (one or more)
+    csv: Vec<String>,
 
-    /// Prover kind: native or browser
-    #[arg(short, long, value_enum, default_value = "native")]
-    prover_kind: ProverKind,
+    /// Labels for each dataset (optional, defaults to "Dataset 1", "Dataset 2", etc.)
+    #[arg(short, long, num_args = 0..)]
+    labels: Vec<String>,
 
     /// Add min/max bands to plots
     #[arg(long, default_value_t = false)]
     min_max_band: bool,
 }
 
-#[derive(Debug, Clone, Copy, PartialEq, Eq, clap::ValueEnum)]
-enum ProverKind {
-    Native,
-    Browser,
-}
-
-impl std::fmt::Display for ProverKind {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        match self {
-            ProverKind::Native => write!(f, "Native"),
-            ProverKind::Browser => write!(f, "Browser"),
-        }
-    }
-}
-
 fn main() -> Result<(), Box<dyn std::error::Error>> {
     let cli = Cli::parse();
 
-    let mut rdr = csv::Reader::from_path(&cli.csv)?;
+    if cli.csv.is_empty() {
+        return Err("At least one CSV file must be provided".into());
+    }
+
+    // Generate labels if not provided
+    let labels: Vec<String> = if cli.labels.is_empty() {
+        cli.csv
+            .iter()
+            .enumerate()
+            .map(|(i, _)| format!("Dataset {}", i + 1))
+            .collect()
+    } else if cli.labels.len() != cli.csv.len() {
+        return Err(format!(
+            "Number of labels ({}) must match number of CSV files ({})",
+            cli.labels.len(),
+            cli.csv.len()
+        )
+        .into());
+    } else {
+        cli.labels.clone()
+    };
+
+    // Load all CSVs and add dataset label
+    let mut dfs = Vec::new();
+    for (csv_path, label) in cli.csv.iter().zip(labels.iter()) {
+        let mut df = CsvReadOptions::default()
+            .try_into_reader_with_file_path(Some(csv_path.clone().into()))?
+            .finish()?;
+
+        let label_series = Series::new("dataset_label".into(), vec![label.as_str(); df.height()]);
+        df.with_column(label_series)?;
+        dfs.push(df);
+    }
+
+    // Combine all dataframes
+    let df = dfs
+        .into_iter()
+        .reduce(|acc, df| acc.vstack(&df).unwrap())
+        .unwrap();
 
     let items: BenchItems = toml::from_str(&std::fs::read_to_string(&cli.toml)?)?;
     let groups = items.group;
 
-    // Prepare data for plotting.
-    let all_data: Vec<Measurement> = rdr
-        .deserialize::<Measurement>()
-        .collect::<Result<Vec<_>, _>>()?;
-
     for group in groups {
-        if group.protocol_latency.is_some() {
-            let latency = group.protocol_latency.unwrap();
-            plot_runtime_vs(
-                &all_data,
-                cli.min_max_band,
-                &group.name,
-                |r| r.bandwidth as f32 / 1000.0, // Kbps to Mbps
-                "Runtime vs Bandwidth",
-                format!("{} ms Latency, {} mode", latency, cli.prover_kind),
-                "runtime_vs_bandwidth.html",
-                "Bandwidth (Mbps)",
-            )?;
+        // Determine which field varies in benches for this group
+        let benches_in_group: Vec<_> = items
+            .bench
+            .iter()
+            .filter(|b| b.group.as_deref() == Some(&group.name))
+            .collect();
+
+        if benches_in_group.is_empty() {
+            continue;
         }
 
-        if group.bandwidth.is_some() {
-            let bandwidth = group.bandwidth.unwrap();
+        // Check which field has varying values
+        let bandwidth_varies = benches_in_group
+            .windows(2)
+            .any(|w| w[0].bandwidth != w[1].bandwidth);
+        let latency_varies = benches_in_group
+            .windows(2)
+            .any(|w| w[0].protocol_latency != w[1].protocol_latency);
+        let download_size_varies = benches_in_group
+            .windows(2)
+            .any(|w| w[0].download_size != w[1].download_size);
+
+        if download_size_varies {
+            let upload_size = group.upload_size.unwrap_or(1024);
             plot_runtime_vs(
-                &all_data,
+                &df,
+                &labels,
                 cli.min_max_band,
                 &group.name,
-                |r| r.latency as f32,
+                "download_size",
+                1.0 / 1024.0, // bytes to KB
+                "Runtime vs Response Size",
+                format!("{} bytes upload size", upload_size),
+                "runtime_vs_download_size",
+                "Response Size (KB)",
+                true, // legend on left
+            )?;
+        } else if bandwidth_varies {
+            let latency = group.protocol_latency.unwrap_or(50);
+            plot_runtime_vs(
+                &df,
+                &labels,
+                cli.min_max_band,
+                &group.name,
+                "bandwidth",
+                1.0 / 1000.0, // Kbps to Mbps
+                "Runtime vs Bandwidth",
+                format!("{} ms Latency", latency),
+                "runtime_vs_bandwidth",
+                "Bandwidth (Mbps)",
+                false, // legend on right
+            )?;
+        } else if latency_varies {
+            let bandwidth = group.bandwidth.unwrap_or(1000);
+            plot_runtime_vs(
+                &df,
+                &labels,
+                cli.min_max_band,
+                &group.name,
+                "latency",
+                1.0,
                 "Runtime vs Latency",
-                format!("{} bps bandwidth, {} mode", bandwidth, cli.prover_kind),
-                "runtime_vs_latency.html",
+                format!("{} bps bandwidth", bandwidth),
+                "runtime_vs_latency",
                 "Latency (ms)",
+                true, // legend on left
             )?;
         }
     }
@@ -92,83 +152,51 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
     Ok(())
 }
 
-struct DataPoint {
-    min: f32,
-    mean: f32,
-    max: f32,
-}
-
-struct Points {
-    preprocess: DataPoint,
-    online: DataPoint,
-    total: DataPoint,
-}
-
 #[allow(clippy::too_many_arguments)]
-fn plot_runtime_vs<Fx>(
-    all_data: &[Measurement],
+fn plot_runtime_vs(
+    df: &DataFrame,
+    labels: &[String],
     show_min_max: bool,
     group: &str,
-    x_value: Fx,
+    x_col: &str,
+    x_scale: f32,
     title: &str,
     subtitle: String,
     output_file: &str,
     x_axis_label: &str,
-) -> Result<Chart, Box<dyn std::error::Error>>
-where
-    Fx: Fn(&Measurement) -> f32,
-{
-    fn data_point(values: &[f32]) -> DataPoint {
-        let mean = values.iter().copied().sum::<f32>() / values.len() as f32;
-        let max = values.iter().copied().reduce(f32::max).unwrap_or_default();
-        let min = values.iter().copied().reduce(f32::min).unwrap_or_default();
-        DataPoint { min, mean, max }
-    }
+    legend_left: bool,
+) -> Result<Chart, Box<dyn std::error::Error>> {
+    let stats_df = df
+        .clone()
+        .lazy()
+        .filter(col("group").eq(lit(group)))
+        .with_column((col(x_col).cast(DataType::Float32) * lit(x_scale)).alias("x"))
+        .with_columns([
+            (col("time_preprocess").cast(DataType::Float32) / lit(1000.0)).alias("preprocess"),
+            (col("time_online").cast(DataType::Float32) / lit(1000.0)).alias("online"),
+            (col("time_total").cast(DataType::Float32) / lit(1000.0)).alias("total"),
+        ])
+        .group_by([col("x"), col("dataset_label")])
+        .agg([
+            col("preprocess").min().alias("preprocess_min"),
+            col("preprocess").mean().alias("preprocess_mean"),
+            col("preprocess").max().alias("preprocess_max"),
+            col("online").min().alias("online_min"),
+            col("online").mean().alias("online_mean"),
+            col("online").max().alias("online_max"),
+            col("total").min().alias("total_min"),
+            col("total").mean().alias("total_mean"),
+            col("total").max().alias("total_max"),
+        ])
+        .sort(["dataset_label", "x"], Default::default())
+        .collect()?;
 
-    let stats: Vec<(f32, Points)> = all_data
-        .iter()
-        .filter(|r| r.group.as_deref() == Some(group))
-        .map(|r| {
-            (
-                x_value(r),
-                r.time_preprocess as f32 / 1000.0, // ms to s
-                r.time_online as f32 / 1000.0,
-                r.time_total as f32 / 1000.0,
-            )
-        })
-        .sorted_by(|a, b| a.0.partial_cmp(&b.0).unwrap())
-        .chunk_by(|entry| entry.0)
-        .into_iter()
-        .map(|(x, group)| {
-            let group_vec: Vec<_> = group.collect();
-            let preprocess = data_point(
-                &group_vec
-                    .iter()
-                    .map(|(_, t, _, _)| *t)
-                    .collect::<Vec<f32>>(),
-            );
-            let online = data_point(
-                &group_vec
-                    .iter()
-                    .map(|(_, _, t, _)| *t)
-                    .collect::<Vec<f32>>(),
-            );
-            let total = data_point(
-                &group_vec
-                    .iter()
-                    .map(|(_, _, _, t)| *t)
-                    .collect::<Vec<f32>>(),
-            );
-            (
-                x,
-                Points {
-                    preprocess,
-                    online,
-                    total,
-                },
-            )
-        })
-        .collect();
+    // Build legend entries
+    let mut legend_data = Vec::new();
+    for label in labels {
+        legend_data.push(format!("Total Mean ({})", label));
+        legend_data.push(format!("Online Mean ({})", label));
+    }
 
     let mut chart = Chart::new()
         .title(
@@ -179,14 +207,6 @@ where
                 .subtext_style(TextStyle::new().font_size(16)),
         )
         .tooltip(Tooltip::new().trigger(Trigger::Axis))
-        .legend(
-            Legend::new()
-                .data(vec!["Preprocess Mean", "Online Mean", "Total Mean"])
-                .top("80")
-                .right("110")
-                .orient(Orient::Vertical)
-                .item_gap(10),
-        )
         .x_axis(
             Axis::new()
                 .name(x_axis_label)
@@ -205,73 +225,156 @@ where
                 .name_text_style(TextStyle::new().font_size(21)),
         );
 
-    chart = add_mean_series(chart, &stats, "Preprocess Mean", |p| p.preprocess.mean);
-    chart = add_mean_series(chart, &stats, "Online Mean", |p| p.online.mean);
-    chart = add_mean_series(chart, &stats, "Total Mean", |p| p.total.mean);
+    // Add legend with conditional positioning
+    let legend = Legend::new()
+        .data(legend_data)
+        .top("80")
+        .orient(Orient::Vertical)
+        .item_gap(10);
 
-    if show_min_max {
-        chart = add_min_max_band(
-            chart,
-            &stats,
-            "Preprocess Min/Max",
-            |p| &p.preprocess,
-            "#ccc",
-        );
-        chart = add_min_max_band(chart, &stats, "Online Min/Max", |p| &p.online, "#ccc");
-        chart = add_min_max_band(chart, &stats, "Total Min/Max", |p| &p.total, "#ccc");
+    let legend = if legend_left {
+        legend.left("110")
+    } else {
+        legend.right("110")
+    };
+
+    chart = chart.legend(legend);
+
+    // Define colors for each dataset
+    let colors = vec![
+        "#5470c6", "#91cc75", "#fac858", "#ee6666", "#73c0de", "#3ba272", "#fc8452", "#9a60b4",
+    ];
+
+    for (idx, label) in labels.iter().enumerate() {
+        let color = colors.get(idx % colors.len()).unwrap();
+
+        // Total time - solid line
+        chart = add_dataset_series(
+            &chart,
+            &stats_df,
+            label,
+            &format!("Total Mean ({})", label),
+            "total_mean",
+            false,
+            color,
+        )?;
+
+        // Online time - dashed line (same color as total)
+        chart = add_dataset_series(
+            &chart,
+            &stats_df,
+            label,
+            &format!("Online Mean ({})", label),
+            "online_mean",
+            true,
+            color,
+        )?;
+
+        if show_min_max {
+            chart = add_dataset_min_max_band(
+                &chart,
+                &stats_df,
+                label,
+                &format!("Total Min/Max ({})", label),
+                "total",
+                color,
+            )?;
+        }
     }
-    // Save the chart as HTML file.
+    // Save the chart as HTML file (no theme)
     HtmlRenderer::new(title, 1000, 800)
-        .theme(THEME)
-        .save(&chart, output_file)
+        .save(&chart, &format!("{}.html", output_file))
+        .unwrap();
+
+    // Save SVG with default theme
+    ImageRenderer::new(1000, 800)
+        .theme(Theme::Default)
+        .save(&chart, &format!("{}.svg", output_file))
+        .unwrap();
+
+    // Save SVG with dark theme
+    ImageRenderer::new(1000, 800)
+        .theme(Theme::Dark)
+        .save(&chart, &format!("{}_dark.svg", output_file))
         .unwrap();
 
     Ok(chart)
 }
 
-fn add_mean_series(
-    chart: Chart,
-    stats: &[(f32, Points)],
-    name: &str,
-    extract: impl Fn(&Points) -> f32,
-) -> Chart {
-    chart.series(
-        Line::new()
-            .name(name)
-            .data(
-                stats
-                    .iter()
-                    .map(|(x, points)| vec![*x, extract(points)])
-                    .collect(),
-            )
-            .symbol_size(6),
-    )
+fn add_dataset_series(
+    chart: &Chart,
+    df: &DataFrame,
+    dataset_label: &str,
+    series_name: &str,
+    col_name: &str,
+    dashed: bool,
+    color: &str,
+) -> Result<Chart, Box<dyn std::error::Error>> {
+    // Filter for specific dataset
+    let mask = df.column("dataset_label")?.str()?.equal(dataset_label);
+    let filtered = df.filter(&mask)?;
+
+    let x = filtered.column("x")?.f32()?;
+    let y = filtered.column(col_name)?.f32()?;
+
+    let data: Vec<Vec<f32>> = x
+        .into_iter()
+        .zip(y.into_iter())
+        .filter_map(|(x, y)| Some(vec![x?, y?]))
+        .collect();
+
+    let mut line = Line::new()
+        .name(series_name)
+        .data(data)
+        .symbol_size(6)
+        .item_style(ItemStyle::new().color(color));
+
+    let mut line_style = LineStyle::new();
+    if dashed {
+        line_style = line_style.type_(LineStyleType::Dashed);
+    }
+    line = line.line_style(line_style.color(color));
+
+    Ok(chart.clone().series(line))
 }
 
-fn add_min_max_band(
-    chart: Chart,
-    stats: &[(f32, Points)],
+fn add_dataset_min_max_band(
+    chart: &Chart,
+    df: &DataFrame,
+    dataset_label: &str,
     name: &str,
-    extract: impl Fn(&Points) -> &DataPoint,
+    col_prefix: &str,
     color: &str,
-) -> Chart {
-    chart.series(
+) -> Result<Chart, Box<dyn std::error::Error>> {
+    // Filter for specific dataset
+    let mask = df.column("dataset_label")?.str()?.equal(dataset_label);
+    let filtered = df.filter(&mask)?;
+
+    let x = filtered.column("x")?.f32()?;
+    let min_col = filtered.column(&format!("{}_min", col_prefix))?.f32()?;
+    let max_col = filtered.column(&format!("{}_max", col_prefix))?.f32()?;
+
+    let max_data: Vec<Vec<f32>> = x
+        .into_iter()
+        .zip(max_col.into_iter())
+        .filter_map(|(x, y)| Some(vec![x?, y?]))
+        .collect();
+
+    let min_data: Vec<Vec<f32>> = x
+        .into_iter()
+        .zip(min_col.into_iter())
+        .filter_map(|(x, y)| Some(vec![x?, y?]))
+        .rev()
+        .collect();
+
+    let data: Vec<Vec<f32>> = max_data.into_iter().chain(min_data).collect();
+
+    Ok(chart.clone().series(
         Line::new()
             .name(name)
-            .data(
-                stats
-                    .iter()
-                    .map(|(x, points)| vec![*x, extract(points).max])
-                    .chain(
-                        stats
-                            .iter()
-                            .rev()
-                            .map(|(x, points)| vec![*x, extract(points).min]),
-                    )
-                    .collect(),
-            )
+            .data(data)
             .show_symbol(false)
             .line_style(LineStyle::new().opacity(0.0))
             .area_style(AreaStyle::new().opacity(0.3).color(color)),
-    )
+    ))
 }

crates/harness/toml/bandwidth.toml

@@ -0,0 +1,25 @@
+#### Bandwidth ####
+
+[[group]]
+name = "bandwidth"
+protocol_latency = 25
+
+[[bench]]
+group = "bandwidth"
+bandwidth = 10
+
+[[bench]]
+group = "bandwidth"
+bandwidth = 50
+
+[[bench]]
+group = "bandwidth"
+bandwidth = 100
+
+[[bench]]
+group = "bandwidth"
+bandwidth = 250
+
+[[bench]]
+group = "bandwidth"
+bandwidth = 1000

crates/harness/toml/download.toml

@@ -0,0 +1,37 @@
+[[group]]
+name = "download_size"
+protocol_latency = 10
+bandwidth = 200
+upload-size = 2048
+
+[[bench]]
+group = "download_size"
+download-size = 1024
+
+[[bench]]
+group = "download_size"
+download-size = 2048
+
+[[bench]]
+group = "download_size"
+download-size = 4096
+
+[[bench]]
+group = "download_size"
+download-size = 8192
+
+[[bench]]
+group = "download_size"
+download-size = 16384
+
+[[bench]]
+group = "download_size"
+download-size = 32768
+
+[[bench]]
+group = "download_size"
+download-size = 65536
+
+[[bench]]
+group = "download_size"
+download-size = 131072

crates/harness/toml/latency.toml

@@ -0,0 +1,25 @@
+#### Latency ####
+
+[[group]]
+name = "latency"
+bandwidth = 1000
+
+[[bench]]
+group = "latency"
+protocol_latency = 10
+
+[[bench]]
+group = "latency"
+protocol_latency = 25
+
+[[bench]]
+group = "latency"
+protocol_latency = 50
+
+[[bench]]
+group = "latency"
+protocol_latency = 100
+
+[[bench]]
+group = "latency"
+protocol_latency = 200