From b80b37deba101e5bd65d3b7db327429b9d2ca3dd Mon Sep 17 00:00:00 2001
From: Haobin Ni <haobin.ni@gmail.com>
Date: Thu, 9 Apr 2026 14:10:38 -0700
Subject: [PATCH 01/31] Save

---
 src/poach.rs | 63 ++++++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 54 insertions(+), 9 deletions(-)

diff --git a/src/poach.rs b/src/poach.rs
index 2345986b9..5e153eb07 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -1,7 +1,8 @@
-use std::path::PathBuf;
+use poach::EGraph;
 
-use clap::{Args, Parser, Subcommand};
+use std::{fs::File, path::PathBuf};
 
+use clap::{Args, Parser, Subcommand};
 
 #[derive(Debug, Parser)]
 #[command(version, about)]
@@ -106,23 +107,67 @@ pub fn poach () {
             fine_tune(arg);
         }
         Commands::Test(arg) => {
-            println!("test({:?})", arg);
+            eprintln!("test({:?})", arg);
+            //TODO: run vanilla egglog tests
         }
     }
     // TODO handle report IO
 }
 
+/// VanillaEgglog's model is just unit
+/// Still, it would create an empty file
 fn train(arg : TrainArgs) {
-    println!("train({:?})", arg);
-    //TODO
+    let _ = File::create(arg.output_model_file.as_path());
 }
 
+/// VanillaEgglog
 fn serve(arg: ServeArgs) {
-    println!("serve({:?})", arg);
-    //TODO
+    match arg.serve_command {
+        None => {
+            let mut egraph = EGraph::default();
+
+        rayon::ThreadPoolBuilder::new()
+            .num_threads(1)
+            .build_global()
+            .unwrap();
+        egraph.repl(poach::RunMode::Normal);
+/*        
+    } else {
+        for input in &args.inputs {
+            let program = std::fs::read_to_string(input).unwrap_or_else(|_| {
+                let arg = input.to_string_lossy();
+                panic!("Failed to read file {arg}")
+            });
+
+            match run_commands(
+                &mut egraph,
+                Some(input.to_str().unwrap().into()),
+                &program,
+                io::stdout(),
+                args.mode,
+            ) {
+                Ok(None) => {}
+                _ => std::process::exit(1),
+            }
+*/
+        }
+        Some(cmd) => {
+            match cmd {
+                ServeCommands::Single{input_file: _} => {
+                    //TODO
+                    panic!("Single not implemented");
+                }
+                ServeCommands::Batch{input_dir:_, output_dir:_} => {
+                    //TODO
+                    panic!("Batch not implemented");
+                }
+            }
+        }
+    }
 }
 
+/// VanillaEgglog's model is just unit
+/// Still, it would create an empty file
 fn fine_tune(arg: FineTuneArgs) {
-    println!("fine_tune({:?})", arg);
-    //TODO
+    let _ = File::create(arg.output_model_file.as_path());
 }

From 0a49cf929fe8bf7f2b1c61c398670983363c782d Mon Sep 17 00:00:00 2001
From: Haobin Ni <haobin.ni@gmail.com>
Date: Thu, 9 Apr 2026 14:26:08 -0700
Subject: [PATCH 02/31] VanillaEgglog

---
 src/poach.rs | 63 +++++++++++++++++++++++++++++-----------------------
 1 file changed, 35 insertions(+), 28 deletions(-)

diff --git a/src/poach.rs b/src/poach.rs
index 5e153eb07..bdb9a44ab 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -1,6 +1,6 @@
-use poach::EGraph;
+use poach::{EGraph};
 
-use std::{fs::File, path::PathBuf};
+use std::{fs::File, path::PathBuf, process::exit};
 
 use clap::{Args, Parser, Subcommand};
 
@@ -126,36 +126,43 @@ fn serve(arg: ServeArgs) {
         None => {
             let mut egraph = EGraph::default();
 
-        rayon::ThreadPoolBuilder::new()
-            .num_threads(1)
-            .build_global()
-            .unwrap();
-        egraph.repl(poach::RunMode::Normal);
-/*        
-    } else {
-        for input in &args.inputs {
-            let program = std::fs::read_to_string(input).unwrap_or_else(|_| {
-                let arg = input.to_string_lossy();
-                panic!("Failed to read file {arg}")
-            });
-
-            match run_commands(
-                &mut egraph,
-                Some(input.to_str().unwrap().into()),
-                &program,
-                io::stdout(),
-                args.mode,
-            ) {
-                Ok(None) => {}
-                _ => std::process::exit(1),
+            rayon::ThreadPoolBuilder::new()
+                .num_threads(1)
+                .build_global()
+                .unwrap();
+
+            match egraph.repl(poach::RunMode::Normal) {
+                Ok(_) => {}
+                _ => {
+                    exit(-1);
+                }
             }
-*/
         }
         Some(cmd) => {
             match cmd {
-                ServeCommands::Single{input_file: _} => {
-                    //TODO
-                    panic!("Single not implemented");
+                ServeCommands::Single{input_file: input} => {
+                    let mut egraph = EGraph::default();
+
+                    rayon::ThreadPoolBuilder::new()
+                        .num_threads(1)
+                        .build_global()
+                        .unwrap();
+
+                    let program = std::fs::read_to_string(input.as_path()).unwrap_or_else(|_| {
+                        let arg = input.to_string_lossy();
+                        panic!("Failed to read file {arg}")
+                    });
+
+                    match egraph.parse_and_run_program(Some(input.to_str().unwrap().into()), &program) {
+                        Ok(msgs) => {
+                            for msg in msgs {
+                                print!("{msg}");
+                            }
+                        }
+                        _ => {
+                            exit(-1);
+                        }
+                    }
                 }
                 ServeCommands::Batch{input_dir:_, output_dir:_} => {
                     //TODO

From 8a9de86f7a6a399bb0418e5eba655fe7e9e07e6c Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 28 Apr 2026 14:41:09 -0700
Subject: [PATCH 03/31] fix build error

---
 src/poach.rs | 64 ++++++++++++++++++++++++----------------------------
 1 file changed, 29 insertions(+), 35 deletions(-)

diff --git a/src/poach.rs b/src/poach.rs
index 55e5b79a1..8d2d986a2 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -122,14 +122,14 @@ pub fn poach() {
 
 /// VanillaEgglog's model is just unit
 /// Still, it would create an empty file
-fn train(arg : TrainArgs) {
+fn train(arg: TrainArgs) {
     let _ = File::create(arg.output_model_file.as_path());
 }
 
 /// VanillaEgglog
 fn serve(arg: ServeArgs) {
-    match arg.serve_command {
-        None => {
+    match arg.mode {
+        ServeMode::Streaming => {
             let mut egraph = EGraph::default();
 
             rayon::ThreadPoolBuilder::new()
@@ -144,43 +144,37 @@ fn serve(arg: ServeArgs) {
                 }
             }
         }
-        Some(cmd) => {
-            match cmd {
-                ServeCommands::Single { input_file: input } => {
-                    let mut egraph = EGraph::default();
-
-                    rayon::ThreadPoolBuilder::new()
-                        .num_threads(1)
-                        .build_global()
-                        .unwrap();
-
-                    let program = std::fs::read_to_string(input.as_path()).unwrap_or_else(|_| {
-                        let arg = input.to_string_lossy();
-                        panic!("Failed to read file {arg}")
-                    });
-
-                    match egraph
-                        .parse_and_run_program(Some(input.to_str().unwrap().into()), &program)
-                    {
-                        Ok(msgs) => {
-                            for msg in msgs {
-                                print!("{msg}");
-                            }
-                        }
-                        _ => {
-                            exit(-1);
-                        }
+        ServeMode::Single { input_file: input } => {
+            let mut egraph = EGraph::default();
+
+            rayon::ThreadPoolBuilder::new()
+                .num_threads(1)
+                .build_global()
+                .unwrap();
+
+            let program = std::fs::read_to_string(input.as_path()).unwrap_or_else(|_| {
+                let arg = input.to_string_lossy();
+                panic!("Failed to read file {arg}")
+            });
+
+            match egraph.parse_and_run_program(Some(input.to_str().unwrap().into()), &program) {
+                Ok(msgs) => {
+                    for msg in msgs {
+                        print!("{msg}");
                     }
                 }
-                ServeCommands::Batch {
-                    input_dir: _,
-                    output_dir: _,
-                } => {
-                    //TODO
-                    panic!("Batch not implemented");
+                _ => {
+                    exit(-1);
                 }
             }
         }
+        ServeMode::Batch {
+            input_dir: _,
+            output_dir: _,
+        } => {
+            //TODO
+            panic!("Batch not implemented");
+        }
     }
 }
 

From 2b4cb7d253b57015e196b9642e02c7f77553371f Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 28 Apr 2026 15:36:48 -0700
Subject: [PATCH 04/31] nightly

---
 infra/nightly-resources/web/index.html | 24 ++++++++--
 infra/nightly-resources/web/index.js   | 61 +++++++++++++++++++++++---
 infra/nightly.py                       | 57 +++++++++++++++++-------
 src/poach.rs                           | 28 ++++++++++--
 tests/{passing => skipped}/include.egg |  0
 5 files changed, 142 insertions(+), 28 deletions(-)
 rename tests/{passing => skipped}/include.egg (100%)

diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index 2a56dbf46..789622937 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -11,13 +11,29 @@
       <h1>Poach Nightly</h1>
       <p id="status">Loading <code>data/data.json</code>...</p>
       <div id="contents">
-        <!-- Put tables/charts/etc here -->
-         <div id="summary"></div>
-
+        <div id="summary">
+          <h2>Summary</h2>
+          <p id="summary-text"></p>
+        </div>
+        <div id="benchmarks">
+          <h2>Reports</h2>
+          <table>
+            <thead>
+              <tr>
+                <th>Path</th>
+                <th>Wall Time</th>
+                <th>Rule Running</th>
+                <th>Extraction</th>
+                <th>Other</th>
+                <th># Timing Steps</th>
+              </tr>
+            </thead>
+            <tbody id="benchmarks-body"></tbody>
+          </table>
+        </div>
       </div>
     </main>
 
     <script type="module" src="./index.js"></script>
-    <script type="module" src="./util.js"></script>
   </body>
 </html>
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index c04090882..51030c86d 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -13,17 +13,66 @@ async function load() {
 
     const data = await response.json();
     statusNode.textContent = "Loaded data/data.json";
-    renderSummary(data.summary);
-    renderBenchmarks(data);
+    renderSummary(data);
+    renderBenchmarks(data.reports);
   } catch (error) {
     statusNode.textContent = `Failed to load data/data.json: ${error}`;
   }
 }
 
-function renderSummary(summary) {
-  // TODO
+function renderSummary(data) {
+  let ruleRunningMillis = 0;
+  let extractionMillis = 0;
+  let otherMillis = 0;
+
+  for (const { report } of data.reports) {
+    const totals = getTimingTotals(report);
+    ruleRunningMillis += totals.ruleRunningMillis;
+    extractionMillis += totals.extractionMillis;
+    otherMillis += totals.otherMillis;
+  }
+
+  document.querySelector("#summary-text").textContent =
+    `${data.reports.length} benchmarks | ` +
+    `Nightly time: ${data.summary.total_time_seconds.toFixed(1)} s | ` +
+    `Rule running: ${ruleRunningMillis} ms | ` +
+    `Extraction: ${extractionMillis} ms | ` +
+    `Other: ${otherMillis} ms`;
+}
+
+function renderBenchmarks(reports) {
+  document.querySelector("#benchmarks-body").innerHTML = reports
+    .map(({ path, time_seconds, report }) => {
+      const totals = getTimingTotals(report);
+
+      return `
+        <tr>
+          <td>${path}</td>
+          <td>${time_seconds.toFixed(3)} s</td>
+          <td>${formatMillis(totals.ruleRunningMillis)}</td>
+          <td>${formatMillis(totals.extractionMillis)}</td>
+          <td>${formatMillis(totals.otherMillis)}</td>
+          <td>${report.timings.length}</td>
+        </tr>
+      `;
+    })
+    .join("");
 }
 
-function renderBenchmarks(data) {
-  // TODO
+function getTimingTotals(report) {
+  let ruleRunningMillis = 0;
+  let extractionMillis = 0;
+  let otherMillis = 0;
+
+  for (const timing of report.timings) {
+    if (timing.tags.includes("running_rules")) {
+      ruleRunningMillis += timing.total;
+    } else if (timing.tags.includes("extraction")) {
+      extractionMillis += timing.total;
+    } else {
+      otherMillis += timing.total;
+    }
+  }
+
+  return { ruleRunningMillis, extractionMillis, otherMillis };
 }
diff --git a/infra/nightly.py b/infra/nightly.py
index 85ad4a1b9..a118a4678 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -19,6 +19,7 @@
 REPORT_OUTPUT_DIR = DATA_DIR / "reports"
 DATA_JSON_PATH = DATA_DIR / "data.json"
 POACH_BIN = REPO_ROOT / "target" / "release" / "poach"
+MODEL_FILE = REPO_ROOT / "empty.model"
 
 
 def main() -> None:
@@ -27,7 +28,7 @@ def main() -> None:
 
     benchmark_dir = (REPO_ROOT / sys.argv[1]).resolve()
 
-    benchmark_files = sorted(benchmark_dir.rglob("*.egg"))
+    benchmark_files = list(benchmark_dir.rglob("*.egg"))
     if not benchmark_files:
         raise SystemExit(
             f"No .egg benchmark files found under {benchmark_dir}."
@@ -40,10 +41,16 @@ def main() -> None:
     DATA_JSON_PATH.write_text(json.dumps(data, indent=2), encoding="utf-8")
     print(f"Wrote {DATA_JSON_PATH}")
 
-def run_command(command: list[str], *, cwd: Path) -> dict[str, Any]:
+def run_command(command: list[str], *, cwd: Path, report_path: Path) -> dict[str, Any]:
     started = time.perf_counter()
     try:
-        completed = subprocess.run(command, check=True, cwd=cwd)
+        completed = subprocess.run(
+            command,
+            check=True,
+            cwd=cwd,
+            capture_output=True,
+            text=True,
+        )
     except subprocess.CalledProcessError as err:
         time_seconds = time.perf_counter() - started
         print(
@@ -52,11 +59,13 @@ def run_command(command: list[str], *, cwd: Path) -> dict[str, Any]:
         )
         raise err
 
-    # TODO: collect stderr and stdout from completed
+    report_path.parent.mkdir(parents=True, exist_ok=True)
+    report_path.write_text(completed.stderr, encoding="utf-8")
 
     return {
         "argv": command,
         "cwd": str(cwd),
+        "report_path": str(report_path),
         "returncode": completed.returncode,
         "time_seconds": time.perf_counter() - started,
     }
@@ -67,34 +76,52 @@ def run_benchmarks(benchmark_dir: Path) -> list[dict[str, Any]]:
         shutil.rmtree(REPORT_OUTPUT_DIR)
     REPORT_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
-    # TODO: poach serve streaming is the default mode
-
-    command = [
-        str(POACH_BIN),         # poach binary
-                                # fill in other args
-        str(benchmark_dir),     # input files
-    ]
-    print("Running benchmarks:", " ".join(command))
-    return [run_command(command, cwd=REPO_ROOT)]
+    results = []
+    for benchmark_file in benchmark_dir.rglob("*.egg"):
+        relative_benchmark = benchmark_file.relative_to(benchmark_dir)
+        report_path = REPORT_OUTPUT_DIR / relative_benchmark.with_suffix(".report.json")
+        command = [
+            str(POACH_BIN),
+            "serve",
+            "--debug",
+            str(MODEL_FILE),
+            "single",
+            str(benchmark_file),
+        ]
+        print("Running benchmark:", " ".join(command))
+        results.append(run_command(command, cwd=REPO_ROOT, report_path=report_path))
+    return results
 
 
 def aggregate_reports(
     benchmark_dir: Path, command_results: list[dict[str, Any]]
 ) -> dict[str, Any]:
-    report_files = sorted(REPORT_OUTPUT_DIR.rglob("*.report.json"))
+    report_files = list(REPORT_OUTPUT_DIR.rglob("*.report.json"))
     if not report_files:
         raise SystemExit(f"No report files were generated under {REPORT_OUTPUT_DIR}")
 
     return {
         "generated_at": datetime.now(timezone.utc).isoformat(),
         "suite": benchmark_dir.name,
-        "benchmark_root": str(benchmark_dir.relative_to(REPO_ROOT)),
+        "benchmark_root": (
+            str(benchmark_dir.relative_to(REPO_ROOT))
+            if benchmark_dir.is_absolute()
+            else str(benchmark_dir)
+        ),
         "summary": {
             "commands": command_results,
             "total_time_seconds": sum(
                 result["time_seconds"] for result in command_results
             ),
         },
+        "reports": [
+            {
+                "path": str(Path(result["report_path"]).relative_to(OUTPUT_DIR)),
+                "time_seconds": result["time_seconds"],
+                "report": json.loads(Path(result["report_path"]).read_text(encoding="utf-8")),
+            }
+            for result in command_results
+        ],
     }
 
 
diff --git a/src/poach.rs b/src/poach.rs
index 8d2d986a2..a53a5f20f 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -1,6 +1,6 @@
-use poach::EGraph;
+use poach::{EGraph, report::Reporter};
 
-use std::{fs::File, path::PathBuf, process::exit};
+use std::{fs::File, io::Write, path::PathBuf, process::exit};
 
 use clap::{Args, Parser, Subcommand};
 
@@ -157,7 +157,29 @@ fn serve(arg: ServeArgs) {
                 panic!("Failed to read file {arg}")
             });
 
-            match egraph.parse_and_run_program(Some(input.to_str().unwrap().into()), &program) {
+            let result: Result<_, poach::Error> = if arg.debug {
+                let filename = Some(input.to_str().unwrap().into());
+                match egraph.parser.get_program_from_string(filename, &program) {
+                    Ok(parsed) => {
+                        let mut reporter = Reporter::new();
+                        let result = egraph.run_program_with_reporter(parsed, &mut reporter);
+                        if result.is_ok() {
+                            let mut stderr = std::io::stderr().lock();
+                            serde_json::to_writer(&mut stderr, &reporter.build_report())
+                                .expect("failed to serialize debug report");
+                            writeln!(stderr).expect("failed to terminate debug report");
+                        }
+                        result
+                    }
+                    Err(err) => Err(err.into()),
+                }
+            } else {
+                egraph
+                    .parse_and_run_program(Some(input.to_str().unwrap().into()), &program)
+                    .map_err(Into::into)
+            };
+
+            match result {
                 Ok(msgs) => {
                     for msg in msgs {
                         print!("{msg}");
diff --git a/tests/passing/include.egg b/tests/skipped/include.egg
similarity index 100%
rename from tests/passing/include.egg
rename to tests/skipped/include.egg

From 43acac4dabf330752a450e3e02c32a9c5ff54298 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 28 Apr 2026 17:55:30 -0700
Subject: [PATCH 05/31] smaller data.json

---
 infra/nightly-resources/web/index.js | 39 +++++++---------------------
 infra/nightly.py                     | 28 ++++++++++++++++++--
 src/report.rs                        | 20 ++++++++++++--
 3 files changed, 53 insertions(+), 34 deletions(-)

diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 51030c86d..f664a725a 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -25,11 +25,10 @@ function renderSummary(data) {
   let extractionMillis = 0;
   let otherMillis = 0;
 
-  for (const { report } of data.reports) {
-    const totals = getTimingTotals(report);
-    ruleRunningMillis += totals.ruleRunningMillis;
-    extractionMillis += totals.extractionMillis;
-    otherMillis += totals.otherMillis;
+  for (const { timing_summary } of data.reports) {
+    ruleRunningMillis += timing_summary.rule_running_millis;
+    extractionMillis += timing_summary.extraction_millis;
+    otherMillis += timing_summary.other_millis;
   }
 
   document.querySelector("#summary-text").textContent =
@@ -42,37 +41,17 @@ function renderSummary(data) {
 
 function renderBenchmarks(reports) {
   document.querySelector("#benchmarks-body").innerHTML = reports
-    .map(({ path, time_seconds, report }) => {
-      const totals = getTimingTotals(report);
-
+    .map(({ path, time_seconds, timing_summary }) => {
       return `
         <tr>
           <td>${path}</td>
           <td>${time_seconds.toFixed(3)} s</td>
-          <td>${formatMillis(totals.ruleRunningMillis)}</td>
-          <td>${formatMillis(totals.extractionMillis)}</td>
-          <td>${formatMillis(totals.otherMillis)}</td>
-          <td>${report.timings.length}</td>
+          <td>${formatMillis(timing_summary.rule_running_millis)}</td>
+          <td>${formatMillis(timing_summary.extraction_millis)}</td>
+          <td>${formatMillis(timing_summary.other_millis)}</td>
+          <td>${timing_summary.timing_steps}</td>
         </tr>
       `;
     })
     .join("");
 }
-
-function getTimingTotals(report) {
-  let ruleRunningMillis = 0;
-  let extractionMillis = 0;
-  let otherMillis = 0;
-
-  for (const timing of report.timings) {
-    if (timing.tags.includes("running_rules")) {
-      ruleRunningMillis += timing.total;
-    } else if (timing.tags.includes("extraction")) {
-      extractionMillis += timing.total;
-    } else {
-      otherMillis += timing.total;
-    }
-  }
-
-  return { ruleRunningMillis, extractionMillis, otherMillis };
-}
diff --git a/infra/nightly.py b/infra/nightly.py
index a118a4678..3e7e31572 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -93,6 +93,27 @@ def run_benchmarks(benchmark_dir: Path) -> list[dict[str, Any]]:
     return results
 
 
+def summarize_report(report: dict[str, Any]) -> dict[str, int]:
+    rule_running_millis = 0
+    extraction_millis = 0
+    other_millis = 0
+
+    for timing in report["timings"]:
+        if "running_rules" in timing["tags"]:
+            rule_running_millis += timing["total"]
+        elif "extraction" in timing["tags"]:
+            extraction_millis += timing["total"]
+        else:
+            other_millis += timing["total"]
+
+    return {
+        "rule_running_millis": rule_running_millis,
+        "extraction_millis": extraction_millis,
+        "other_millis": other_millis,
+        "timing_steps": len(report["timings"]),
+    }
+
+
 def aggregate_reports(
     benchmark_dir: Path, command_results: list[dict[str, Any]]
 ) -> dict[str, Any]:
@@ -109,7 +130,6 @@ def aggregate_reports(
             else str(benchmark_dir)
         ),
         "summary": {
-            "commands": command_results,
             "total_time_seconds": sum(
                 result["time_seconds"] for result in command_results
             ),
@@ -118,7 +138,11 @@ def aggregate_reports(
             {
                 "path": str(Path(result["report_path"]).relative_to(OUTPUT_DIR)),
                 "time_seconds": result["time_seconds"],
-                "report": json.loads(Path(result["report_path"]).read_text(encoding="utf-8")),
+                "timing_summary": summarize_report(
+                    json.loads(
+                        Path(result["report_path"]).read_text(encoding="utf-8")
+                    )
+                ),
             }
             for result in command_results
         ],
diff --git a/src/report.rs b/src/report.rs
index 3bf7c3b39..0d85dc6fd 100644
--- a/src/report.rs
+++ b/src/report.rs
@@ -1,7 +1,7 @@
 use std::fmt::{self, Display, Formatter};
 use std::time::{Duration, Instant};
 
-use serde::Serialize;
+use serde::{Serialize, Serializer};
 
 type TimerHandle = usize;
 
@@ -30,6 +30,7 @@ struct TimingStep {
     tags: Vec<String>,
     #[serde(with = "serde_millis")]
     total: Duration,
+    #[serde(serialize_with = "serialize_duration_breakdown")]
     breakdown: Vec<Duration>,
 }
 
@@ -109,11 +110,26 @@ impl Display for TimingStep {
         writeln!(f, "  name: {}", self.name)?;
         writeln!(f, "  tags: {:?}", self.tags)?;
         writeln!(f, "  total: {}", self.total.as_secs_f64())?;
-        writeln!(f, "  breakdown: {:?}", self.breakdown)?;
+        writeln!(
+            f,
+            "  breakdown: {:?}",
+            self.breakdown
+                .iter()
+                .map(Duration::as_millis)
+                .collect::<Vec<_>>()
+        )?;
         Ok(())
     }
 }
 
+fn serialize_duration_breakdown<S>(breakdown: &[Duration], serializer: S) -> Result<S::Ok, S::Error>
+where
+    S: Serializer,
+{
+    let breakdown_millis: Vec<_> = breakdown.iter().map(Duration::as_millis).collect();
+    breakdown_millis.serialize(serializer)
+}
+
 fn write_metric_section<T>(f: &mut Formatter<'_>, title: &str, metrics: &[T]) -> fmt::Result
 where
     T: Display,

From 98ab672b42a723d60b4279869b2aa403090f0d82 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Wed, 29 Apr 2026 09:31:55 -0700
Subject: [PATCH 06/31] nightlies with real benchmarks

---
 AGENTS.md                              |   1 +
 infra/nightly-resources/web/index.html |   2 +
 infra/nightly-resources/web/index.js   |  60 ++++++++++--
 infra/nightly-resources/web/style.css  |  38 ++++++++
 infra/nightly.py                       | 122 +++++++++++++++----------
 infra/nightly.sh                       |   9 +-
 6 files changed, 173 insertions(+), 59 deletions(-)

diff --git a/AGENTS.md b/AGENTS.md
index a9f87946d..8fb28664e 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -18,6 +18,7 @@
 ## Conventions + Style
 
 - Avoid defensive code that adds complexity without a clear, likely payoff.
+- Avoid introducing local variables that are only used once or twice when inlining keeps the code clear.
 
 ## Response format
 
diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index 789622937..20b84bb35 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -17,6 +17,8 @@ <h2>Summary</h2>
         </div>
         <div id="benchmarks">
           <h2>Reports</h2>
+          <div id="suite-tabs"></div>
+          <div id="active-suite-summary"></div>
           <table>
             <thead>
               <tr>
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index f664a725a..49bb4a6a0 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -1,10 +1,12 @@
 import { formatMillis } from "./util.js";
 
-const statusNode = document.querySelector("#status");
-
+let suites = [];
+let activeSuiteName = null;
 load();
 
 async function load() {
+  const statusNode = document.querySelector("#status");
+
   try {
     const response = await fetch("./data/data.json");
     if (!response.ok) {
@@ -12,9 +14,11 @@ async function load() {
     }
 
     const data = await response.json();
+    suites = data.suites;
+    activeSuiteName = suites[0]?.name ?? null;
     statusNode.textContent = "Loaded data/data.json";
     renderSummary(data);
-    renderBenchmarks(data.reports);
+    renderSuites();
   } catch (error) {
     statusNode.textContent = `Failed to load data/data.json: ${error}`;
   }
@@ -32,19 +36,59 @@ function renderSummary(data) {
   }
 
   document.querySelector("#summary-text").textContent =
-    `${data.reports.length} benchmarks | ` +
+    `${data.summary.benchmark_count} benchmarks across ${data.suites.length} suites | ` +
     `Nightly time: ${data.summary.total_time_seconds.toFixed(1)} s | ` +
     `Rule running: ${ruleRunningMillis} ms | ` +
     `Extraction: ${extractionMillis} ms | ` +
     `Other: ${otherMillis} ms`;
 }
 
-function renderBenchmarks(reports) {
-  document.querySelector("#benchmarks-body").innerHTML = reports
-    .map(({ path, time_seconds, timing_summary }) => {
+function renderSuites() {
+  document.querySelector("#suite-tabs").innerHTML = suites
+    .map((suite) => {
+      return `
+        <button
+          type="button"
+          class="suite-tab${suite.name === activeSuiteName ? " is-active" : ""}"
+          data-suite-name="${suite.name}"
+        >
+          ${suite.name}
+        </button>
+      `;
+    })
+    .join("");
+
+  for (const button of document.querySelectorAll(".suite-tab")) {
+    button.addEventListener("click", () => {
+      activeSuiteName = button.dataset.suiteName;
+      renderSuites();
+    });
+  }
+
+  const activeSuite = suites.find((suite) => suite.name === activeSuiteName);
+  if (!activeSuite) {
+    document.querySelector("#active-suite-summary").textContent = "";
+    document.querySelector("#benchmarks-body").innerHTML = "";
+    return;
+  }
+
+  document.querySelector("#active-suite-summary").innerHTML = `
+    <div class="suite-header">
+      <h3>${activeSuite.name}</h3>
+      <p>${activeSuite.reports.length} benchmarks | ${activeSuite.summary.total_time_seconds.toFixed(1)} s</p>
+    </div>
+  `;
+  document.querySelector("#benchmarks-body").innerHTML = renderRows(
+    activeSuite.reports,
+  );
+}
+
+function renderRows(reports) {
+  return reports
+    .map(({ benchmark_path, time_seconds, timing_summary }) => {
       return `
         <tr>
-          <td>${path}</td>
+          <td>${benchmark_path}</td>
           <td>${time_seconds.toFixed(3)} s</td>
           <td>${formatMillis(timing_summary.rule_running_millis)}</td>
           <td>${formatMillis(timing_summary.extraction_millis)}</td>
diff --git a/infra/nightly-resources/web/style.css b/infra/nightly-resources/web/style.css
index 99bb78f65..8eb1262d4 100644
--- a/infra/nightly-resources/web/style.css
+++ b/infra/nightly-resources/web/style.css
@@ -31,6 +31,44 @@ h1 {
   margin: 0 0 16px;
 }
 
+.suite {
+  margin: 0 0 24px;
+}
+
+.suite-tab {
+  appearance: none;
+  border: 1px solid #ccc;
+  background: #f0f0f0;
+  color: inherit;
+  padding: 8px 12px;
+  margin: 0 8px 12px 0;
+  font: inherit;
+  cursor: pointer;
+}
+
+.suite-tab.is-active {
+  background: #111;
+  border-color: #111;
+  color: #fff;
+}
+
+.suite-header {
+  display: flex;
+  align-items: baseline;
+  justify-content: space-between;
+  gap: 16px;
+  margin: 0 0 8px;
+}
+
+.suite-header p {
+  margin: 0;
+}
+
+h3 {
+  margin: 0;
+  font-size: 1rem;
+}
+
 h2 {
   margin: 0 0 8px;
   font-size: 1.1rem;
diff --git a/infra/nightly.py b/infra/nightly.py
index 3e7e31572..c696980f2 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -26,16 +26,17 @@ def main() -> None:
     if len(sys.argv) != 2:
         raise SystemExit(f"Usage: {Path(sys.argv[0]).name} <benchmark-dir>")
 
-    benchmark_dir = (REPO_ROOT / sys.argv[1]).resolve()
-
-    benchmark_files = list(benchmark_dir.rglob("*.egg"))
-    if not benchmark_files:
-        raise SystemExit(
-            f"No .egg benchmark files found under {benchmark_dir}."
-        )
-
-    command_results = run_benchmarks(benchmark_dir)
-    data = aggregate_reports(benchmark_dir, command_results)
+    benchmark_root = (REPO_ROOT / sys.argv[1]).resolve()
+    benchmark_dirs = list(
+        path
+        for path in benchmark_root.iterdir()
+        if path.is_dir() and any(path.rglob("*.egg"))
+    )
+    if not benchmark_dirs:
+        raise SystemExit(f"No benchmark suite directories found under {benchmark_root}.")
+
+    benchmark_results = run_benchmarks(benchmark_dirs)
+    data = aggregate_reports(benchmark_dirs, benchmark_results)
 
     DATA_DIR.mkdir(parents=True, exist_ok=True)
     DATA_JSON_PATH.write_text(json.dumps(data, indent=2), encoding="utf-8")
@@ -70,26 +71,32 @@ def run_command(command: list[str], *, cwd: Path, report_path: Path) -> dict[str
         "time_seconds": time.perf_counter() - started,
     }
 
-
-def run_benchmarks(benchmark_dir: Path) -> list[dict[str, Any]]:
+def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
     if REPORT_OUTPUT_DIR.exists():
         shutil.rmtree(REPORT_OUTPUT_DIR)
     REPORT_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
 
+    benchmark_root = benchmark_dirs[0].parent
     results = []
-    for benchmark_file in benchmark_dir.rglob("*.egg"):
-        relative_benchmark = benchmark_file.relative_to(benchmark_dir)
-        report_path = REPORT_OUTPUT_DIR / relative_benchmark.with_suffix(".report.json")
-        command = [
-            str(POACH_BIN),
-            "serve",
-            "--debug",
-            str(MODEL_FILE),
-            "single",
-            str(benchmark_file),
-        ]
-        print("Running benchmark:", " ".join(command))
-        results.append(run_command(command, cwd=REPO_ROOT, report_path=report_path))
+    for benchmark_dir in benchmark_dirs:
+        benchmark_files = list(benchmark_dir.rglob("*.egg"))
+        for benchmark_file in benchmark_files:
+            report_path = REPORT_OUTPUT_DIR / benchmark_file.relative_to(
+                benchmark_root
+            ).with_suffix(".report.json")
+            command = [
+                str(POACH_BIN),
+                "serve",
+                "--debug",
+                str(MODEL_FILE),
+                "single",
+                str(benchmark_file),
+            ]
+            print("Running benchmark:", " ".join(command))
+            result = run_command(command, cwd=REPO_ROOT, report_path=report_path)
+            result["suite"] = benchmark_dir.name
+            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir))
+            results.append(result)
     return results
 
 
@@ -115,37 +122,58 @@ def summarize_report(report: dict[str, Any]) -> dict[str, int]:
 
 
 def aggregate_reports(
-    benchmark_dir: Path, command_results: list[dict[str, Any]]
+    benchmark_dirs: list[Path], benchmark_results: list[dict[str, Any]]
 ) -> dict[str, Any]:
-    report_files = list(REPORT_OUTPUT_DIR.rglob("*.report.json"))
-    if not report_files:
+    if not benchmark_results:
         raise SystemExit(f"No report files were generated under {REPORT_OUTPUT_DIR}")
 
+    benchmark_root = benchmark_dirs[0].parent
+    suites = []
+    for benchmark_dir in benchmark_dirs:
+        suite_results = [
+            result for result in benchmark_results if result["suite"] == benchmark_dir.name
+        ]
+        suites.append(
+            {
+                "name": benchmark_dir.name,
+                "benchmark_root": str(benchmark_dir.relative_to(REPO_ROOT)),
+                "summary": {
+                    "total_time_seconds": sum(
+                        result["time_seconds"] for result in suite_results
+                    ),
+                },
+                "reports": [
+                    {
+                        "suite": benchmark_dir.name,
+                        "benchmark_path": result["benchmark_path"],
+                        "path": str(
+                            Path(result["report_path"]).relative_to(OUTPUT_DIR)
+                        ),
+                        "time_seconds": result["time_seconds"],
+                        "timing_summary": summarize_report(
+                            json.loads(
+                                Path(result["report_path"]).read_text(
+                                    encoding="utf-8"
+                                )
+                            )
+                        ),
+                    }
+                    for result in suite_results
+                ],
+            }
+        )
+
     return {
         "generated_at": datetime.now(timezone.utc).isoformat(),
-        "suite": benchmark_dir.name,
-        "benchmark_root": (
-            str(benchmark_dir.relative_to(REPO_ROOT))
-            if benchmark_dir.is_absolute()
-            else str(benchmark_dir)
-        ),
+        "benchmark_root": str(benchmark_root.relative_to(REPO_ROOT)),
         "summary": {
+            "benchmark_count": len(benchmark_results),
             "total_time_seconds": sum(
-                result["time_seconds"] for result in command_results
+                result["time_seconds"] for result in benchmark_results
             ),
         },
-        "reports": [
-            {
-                "path": str(Path(result["report_path"]).relative_to(OUTPUT_DIR)),
-                "time_seconds": result["time_seconds"],
-                "timing_summary": summarize_report(
-                    json.loads(
-                        Path(result["report_path"]).read_text(encoding="utf-8")
-                    )
-                ),
-            }
-            for result in command_results
-        ],
+        "suites": suites,
+        "reports": [report for suite in suites for report in suite["reports"]],
     }
 
 
diff --git a/infra/nightly.sh b/infra/nightly.sh
index fd581c521..2807f9bba 100644
--- a/infra/nightly.sh
+++ b/infra/nightly.sh
@@ -20,14 +20,15 @@ rm -rf nightly
 mkdir -p nightly/output
 mkdir -p nightly/tmp
 
-# TODO: use real benchmarks
-# git clone https://github.com/ajpal/poach-benchmarks.git
+BENCHMARKS_DIR="nightly/tmp/poach-benchmarks"
+
+git clone https://github.com/ajpal/poach-benchmarks.git "$BENCHMARKS_DIR"
 
 # Build in release mode before running nightly.py
 cargo build --release
 
 # This script runs all of the benchmarks/experiments
-python3 infra/nightly.py tests/passing
+python3 infra/nightly.py "$BENCHMARKS_DIR"
 
 # Abort if nightly.py failed to produce data.json. Without this check,
 # the nightly runner will report the nightly as successful even though the
@@ -40,4 +41,4 @@ fi
 cp infra/nightly-resources/web/* nightly/output
 
 # Uncomment for local development
-# cd nightly/output && python3 -m http.server 8002
+cd nightly/output && python3 -m http.server 8002

From 5bc67d3a12bec98fc73a0736569d33cfe32acbc9 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Wed, 29 Apr 2026 09:43:07 -0700
Subject: [PATCH 07/31] fail faster

---
 infra/nightly.sh | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/infra/nightly.sh b/infra/nightly.sh
index 2807f9bba..becf2026e 100644
--- a/infra/nightly.sh
+++ b/infra/nightly.sh
@@ -1,4 +1,6 @@
 #!/bin/bash
+set -euo pipefail
+
 REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 
 echo "Beginning POACH nightly script..."

From f1979b311440bee6e1e2371477967a768144dab6 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Wed, 29 Apr 2026 09:51:31 -0700
Subject: [PATCH 08/31] oops

---
 infra/nightly.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/infra/nightly.sh b/infra/nightly.sh
index becf2026e..795e73dd2 100644
--- a/infra/nightly.sh
+++ b/infra/nightly.sh
@@ -43,4 +43,4 @@ fi
 cp infra/nightly-resources/web/* nightly/output
 
 # Uncomment for local development
-cd nightly/output && python3 -m http.server 8002
+# cd nightly/output && python3 -m http.server 8002

From b861b1650720dd55ab5ae60c30652e9bfd5e7447 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Wed, 29 Apr 2026 10:53:22 -0700
Subject: [PATCH 09/31] update paths

---
 infra/nightly.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index c696980f2..e66944b8e 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -30,7 +30,7 @@ def main() -> None:
     benchmark_dirs = list(
         path
         for path in benchmark_root.iterdir()
-        if path.is_dir() and any(path.rglob("*.egg"))
+        if path.is_dir() and any((path / "train").glob("*.egg"))
     )
     if not benchmark_dirs:
         raise SystemExit(f"No benchmark suite directories found under {benchmark_root}.")
@@ -79,7 +79,7 @@ def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
     benchmark_root = benchmark_dirs[0].parent
     results = []
     for benchmark_dir in benchmark_dirs:
-        benchmark_files = list(benchmark_dir.rglob("*.egg"))
+        benchmark_files = list((benchmark_dir / "train").glob("*.egg"))
         for benchmark_file in benchmark_files:
             report_path = REPORT_OUTPUT_DIR / benchmark_file.relative_to(
                 benchmark_root

From d152534ecd099563175797b1abb933e5865ca149 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Wed, 29 Apr 2026 12:07:57 -0700
Subject: [PATCH 10/31] sortable tables

---
 infra/nightly-resources/web/index.html | 14 +++----
 infra/nightly-resources/web/index.js   | 51 +++++++++++++++++++++++++-
 2 files changed, 56 insertions(+), 9 deletions(-)

diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index 20b84bb35..ed9d726eb 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -21,13 +21,13 @@ <h2>Reports</h2>
           <div id="active-suite-summary"></div>
           <table>
             <thead>
-              <tr>
-                <th>Path</th>
-                <th>Wall Time</th>
-                <th>Rule Running</th>
-                <th>Extraction</th>
-                <th>Other</th>
-                <th># Timing Steps</th>
+              <tr id="benchmarks-header">
+                <th data-sort-key="benchmark_path">Path</th>
+                <th data-sort-key="time_seconds">Wall Time</th>
+                <th data-sort-key="timing_summary.rule_running_millis">Rule Running</th>
+                <th data-sort-key="timing_summary.extraction_millis">Extraction</th>
+                <th data-sort-key="timing_summary.other_millis">Other</th>
+                <th data-sort-key="timing_summary.timing_steps"># Timing Steps</th>
               </tr>
             </thead>
             <tbody id="benchmarks-body"></tbody>
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 49bb4a6a0..51c4d3238 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -2,7 +2,10 @@ import { formatMillis } from "./util.js";
 
 let suites = [];
 let activeSuiteName = null;
+let sortKey = "benchmark_path";
+let sortDir = "asc";
 load();
+installHeaderSortHandlers();
 
 async function load() {
   const statusNode = document.querySelector("#status");
@@ -14,7 +17,7 @@ async function load() {
     }
 
     const data = await response.json();
-    suites = data.suites;
+    suites = [...data.suites].sort((a, b) => a.name.localeCompare(b.name));
     activeSuiteName = suites[0]?.name ?? null;
     statusNode.textContent = "Loaded data/data.json";
     renderSummary(data);
@@ -79,8 +82,52 @@ function renderSuites() {
     </div>
   `;
   document.querySelector("#benchmarks-body").innerHTML = renderRows(
-    activeSuite.reports,
+    sortReports(activeSuite.reports),
   );
+  updateHeaderIndicators();
+}
+
+function getSortValue(report, key) {
+  return key.split(".").reduce((acc, part) => acc?.[part], report) ?? 0;
+}
+
+function sortReports(reports) {
+  const sorted = [...reports];
+  const dir = sortDir === "asc" ? 1 : -1;
+  sorted.sort((a, b) => {
+    const av = getSortValue(a, sortKey);
+    const bv = getSortValue(b, sortKey);
+    if (typeof av === "string" || typeof bv === "string") {
+      return String(av).localeCompare(String(bv)) * dir;
+    }
+    return (av - bv) * dir;
+  });
+  return sorted;
+}
+
+function installHeaderSortHandlers() {
+  for (const th of document.querySelectorAll("#benchmarks-header th")) {
+    th.style.cursor = "pointer";
+    th.addEventListener("click", () => {
+      const key = th.dataset.sortKey;
+      if (sortKey === key) {
+        sortDir = sortDir === "asc" ? "desc" : "asc";
+      } else {
+        sortKey = key;
+        sortDir = "asc";
+      }
+      renderSuites();
+    });
+  }
+}
+
+function updateHeaderIndicators() {
+  for (const th of document.querySelectorAll("#benchmarks-header th")) {
+    const label = th.textContent.replace(/[ ▲▼]+$/, "");
+    const arrow =
+      th.dataset.sortKey === sortKey ? (sortDir === "asc" ? " ▲" : " ▼") : "";
+    th.textContent = label + arrow;
+  }
 }
 
 function renderRows(reports) {

From 6440e81ebfd272171497e0baef266cf7e650b8fc Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Thu, 30 Apr 2026 10:47:19 -0700
Subject: [PATCH 11/31] Split nightly setup into its own script

Extract toolchain update + benchmarks clone into infra/setup.sh, and have
nightly.sh skip that setup when run by the combined-nightly orchestrator
(which sets POACH_NIGHTLY_COMBINED=1 and supplies POACH_BENCHMARKS_DIR).
Standalone behavior is unchanged.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 infra/nightly.sh | 24 +++++++++++++-----------
 infra/setup.sh   | 13 +++++++++++++
 2 files changed, 26 insertions(+), 11 deletions(-)
 mode change 100644 => 100755 infra/nightly.sh
 create mode 100755 infra/setup.sh

diff --git a/infra/nightly.sh b/infra/nightly.sh
old mode 100644
new mode 100755
index 795e73dd2..8a04ee749
--- a/infra/nightly.sh
+++ b/infra/nightly.sh
@@ -6,25 +6,27 @@ REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 echo "Beginning POACH nightly script..."
 
 ###############################################################################
-# This script generates the data for the nightly frontend 
+# This script generates the data for the nightly frontend
 ###############################################################################
 
 export PATH=~/.cargo/bin:$PATH
 
-rustup update
-
-cargo install rustfilt
-
 # Ensure we start from a clean slate
 rm -rf nightly
+mkdir -p nightly/output nightly/tmp
 
-# Set Up
-mkdir -p nightly/output
-mkdir -p nightly/tmp
-
-BENCHMARKS_DIR="nightly/tmp/poach-benchmarks"
+# Standalone runs do their own setup (toolchain + benchmarks clone). When
+# driven by the combined orchestrator, POACH_NIGHTLY_COMBINED=1 and the
+# benchmarks dir is supplied via POACH_BENCHMARKS_DIR.
+if [ -z "${POACH_NIGHTLY_COMBINED:-}" ]; then
+  bash infra/setup.sh
+fi
 
-git clone https://github.com/ajpal/poach-benchmarks.git "$BENCHMARKS_DIR"
+BENCHMARKS_DIR="${POACH_BENCHMARKS_DIR:-nightly/tmp/poach-benchmarks}"
+if [ ! -d "$BENCHMARKS_DIR" ]; then
+  echo "ERROR: benchmarks dir $BENCHMARKS_DIR not found" >&2
+  exit 1
+fi
 
 # Build in release mode before running nightly.py
 cargo build --release
diff --git a/infra/setup.sh b/infra/setup.sh
new file mode 100755
index 000000000..a55c5c13a
--- /dev/null
+++ b/infra/setup.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+# One-time setup for the nightly run: rust toolchain update + benchmarks clone.
+# Skipped when invoked by the combined-nightly orchestrator, which performs
+# these steps once at the top level before driving each branch's nightly.sh.
+set -euo pipefail
+
+export PATH=~/.cargo/bin:$PATH
+
+rustup update
+cargo install rustfilt
+
+mkdir -p nightly/tmp
+git clone https://github.com/ajpal/poach-benchmarks.git nightly/tmp/poach-benchmarks

From 90b12c16253b6b77e4f235031c50efeaae0c41e3 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Thu, 30 Apr 2026 11:01:33 -0700
Subject: [PATCH 12/31] Tolerate benchmarks dir outside REPO_ROOT in nightly.py

When the combined-nightly orchestrator runs this branch in a worktree, it
passes POACH_BENCHMARKS_DIR pointing at a shared clone above the worktree
(so we don't re-clone per branch). The benchmark_root.relative_to(REPO_ROOT)
call then raises ValueError. Add display_path() that falls back to the
absolute path when the input isn't a subpath of REPO_ROOT.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 infra/nightly.py | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index e66944b8e..bd184174b 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -100,6 +100,17 @@ def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
     return results
 
 
+def display_path(p: Path) -> str:
+    # Used for serializing benchmark roots into data.json. Falls back to the
+    # absolute path when the benchmarks dir lives outside REPO_ROOT (e.g. when
+    # the combined-nightly orchestrator points POACH_BENCHMARKS_DIR at a
+    # shared clone above the worktree).
+    try:
+        return str(p.relative_to(REPO_ROOT))
+    except ValueError:
+        return str(p)
+
+
 def summarize_report(report: dict[str, Any]) -> dict[str, int]:
     rule_running_millis = 0
     extraction_millis = 0
@@ -136,7 +147,7 @@ def aggregate_reports(
         suites.append(
             {
                 "name": benchmark_dir.name,
-                "benchmark_root": str(benchmark_dir.relative_to(REPO_ROOT)),
+                "benchmark_root": display_path(benchmark_dir),
                 "summary": {
                     "total_time_seconds": sum(
                         result["time_seconds"] for result in suite_results
@@ -165,7 +176,7 @@ def aggregate_reports(
 
     return {
         "generated_at": datetime.now(timezone.utc).isoformat(),
-        "benchmark_root": str(benchmark_root.relative_to(REPO_ROOT)),
+        "benchmark_root": display_path(benchmark_root),
         "summary": {
             "benchmark_count": len(benchmark_results),
             "total_time_seconds": sum(

From 14fecb744932355f2d62ed8530fb02abea43882f Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Thu, 30 Apr 2026 12:27:01 -0700
Subject: [PATCH 13/31] Drop train/ prefix from benchmark_path

Emit benchmark_path relative to the suite's train/ subdir (matching the
serialize branch's convention) so the combined-nightly comparison table
can join rows by benchmark_path without the orchestrator having to strip
the prefix.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 infra/nightly.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index bd184174b..dcd460039 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -95,7 +95,7 @@ def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
             print("Running benchmark:", " ".join(command))
             result = run_command(command, cwd=REPO_ROOT, report_path=report_path)
             result["suite"] = benchmark_dir.name
-            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir))
+            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir / "train"))
             results.append(result)
     return results
 

From 2296180d26dab9f96f120b01841af5e089cbd831 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Thu, 30 Apr 2026 12:28:52 -0700
Subject: [PATCH 14/31] Strip .egg suffix from benchmark_path

Match the serialize branch exactly so the combined-nightly merger can join
rows by benchmark_path with no normalization.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 infra/nightly.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index dcd460039..2ee209253 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -95,7 +95,9 @@ def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
             print("Running benchmark:", " ".join(command))
             result = run_command(command, cwd=REPO_ROOT, report_path=report_path)
             result["suite"] = benchmark_dir.name
-            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir / "train"))
+            result["benchmark_path"] = str(
+                benchmark_file.relative_to(benchmark_dir / "train").with_suffix("")
+            )
             results.append(result)
     return results
 

From 37c519a34aa568217f1cd31c4b0f7246230d8887 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Thu, 30 Apr 2026 13:32:25 -0700
Subject: [PATCH 15/31] Revert .egg stripping in benchmark_path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Serialize emits `<name>.egg`, not `<name>` — the previous edit over-corrected.
Drop the .with_suffix("") so vanilla and serialize both emit `<name>.egg`
and the merger joins rows correctly.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---
 infra/nightly.py | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index 2ee209253..dcd460039 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -95,9 +95,7 @@ def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
             print("Running benchmark:", " ".join(command))
             result = run_command(command, cwd=REPO_ROOT, report_path=report_path)
             result["suite"] = benchmark_dir.name
-            result["benchmark_path"] = str(
-                benchmark_file.relative_to(benchmark_dir / "train").with_suffix("")
-            )
+            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir / "train"))
             results.append(result)
     return results
 

From 85356380f32b7fcff4c3015b8268d5573b636cec Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Mon, 4 May 2026 09:52:38 -0700
Subject: [PATCH 16/31] no lock needed

---
 src/poach.rs | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/src/poach.rs b/src/poach.rs
index a53a5f20f..5de7cc439 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -164,10 +164,9 @@ fn serve(arg: ServeArgs) {
                         let mut reporter = Reporter::new();
                         let result = egraph.run_program_with_reporter(parsed, &mut reporter);
                         if result.is_ok() {
-                            let mut stderr = std::io::stderr().lock();
-                            serde_json::to_writer(&mut stderr, &reporter.build_report())
+                            serde_json::to_writer(&mut std::io::stderr(), &reporter.build_report())
                                 .expect("failed to serialize debug report");
-                            writeln!(stderr).expect("failed to terminate debug report");
+                            writeln!(std::io::stderr()).expect("failed to terminate debug report");
                         }
                         result
                     }

From 7899210c15a741922141aca99e30d93bc6c217e9 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 14:22:53 -0700
Subject: [PATCH 17/31] nightly script

---
 infra/nightly.py | 291 ++++++++++++++++++-----------------------------
 infra/nightly.sh |   2 +-
 2 files changed, 109 insertions(+), 184 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index dcd460039..9af258dba 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -1,196 +1,121 @@
-#!/usr/bin/env python3
-
-from __future__ import annotations
+#!/user/bin/env python3
 
 import json
-import shutil
 import subprocess
 import sys
 import time
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Any
-
-
-REPO_ROOT = Path(__file__).resolve().parent.parent
-NIGHTLY_DIR = REPO_ROOT / "nightly"
-OUTPUT_DIR = NIGHTLY_DIR / "output"
-DATA_DIR = OUTPUT_DIR / "data"
-REPORT_OUTPUT_DIR = DATA_DIR / "reports"
-DATA_JSON_PATH = DATA_DIR / "data.json"
-POACH_BIN = REPO_ROOT / "target" / "release" / "poach"
-MODEL_FILE = REPO_ROOT / "empty.model"
-
-
-def main() -> None:
-    if len(sys.argv) != 2:
-        raise SystemExit(f"Usage: {Path(sys.argv[0]).name} <benchmark-dir>")
-
-    benchmark_root = (REPO_ROOT / sys.argv[1]).resolve()
-    benchmark_dirs = list(
-        path
-        for path in benchmark_root.iterdir()
-        if path.is_dir() and any((path / "train").glob("*.egg"))
-    )
-    if not benchmark_dirs:
-        raise SystemExit(f"No benchmark suite directories found under {benchmark_root}.")
-
-    benchmark_results = run_benchmarks(benchmark_dirs)
-    data = aggregate_reports(benchmark_dirs, benchmark_results)
-
-    DATA_DIR.mkdir(parents=True, exist_ok=True)
-    DATA_JSON_PATH.write_text(json.dumps(data, indent=2), encoding="utf-8")
-    print(f"Wrote {DATA_JSON_PATH}")
-
-def run_command(command: list[str], *, cwd: Path, report_path: Path) -> dict[str, Any]:
-    started = time.perf_counter()
-    try:
-        completed = subprocess.run(
-            command,
-            check=True,
-            cwd=cwd,
-            capture_output=True,
-            text=True,
-        )
-    except subprocess.CalledProcessError as err:
-        time_seconds = time.perf_counter() - started
-        print(
-            f"Command failed after {time_seconds:.3f}s: {' '.join(command)}",
-            file=sys.stderr,
-        )
-        raise err
-
-    report_path.parent.mkdir(parents=True, exist_ok=True)
-    report_path.write_text(completed.stderr, encoding="utf-8")
-
-    return {
-        "argv": command,
-        "cwd": str(cwd),
-        "report_path": str(report_path),
-        "returncode": completed.returncode,
-        "time_seconds": time.perf_counter() - started,
-    }
-
-def run_benchmarks(benchmark_dirs: list[Path]) -> list[dict[str, Any]]:
-    if REPORT_OUTPUT_DIR.exists():
-        shutil.rmtree(REPORT_OUTPUT_DIR)
-    REPORT_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
-
-    benchmark_root = benchmark_dirs[0].parent
-    results = []
-    for benchmark_dir in benchmark_dirs:
-        benchmark_files = list((benchmark_dir / "train").glob("*.egg"))
-        for benchmark_file in benchmark_files:
-            report_path = REPORT_OUTPUT_DIR / benchmark_file.relative_to(
-                benchmark_root
-            ).with_suffix(".report.json")
-            command = [
-                str(POACH_BIN),
-                "serve",
-                "--debug",
-                str(MODEL_FILE),
-                "single",
-                str(benchmark_file),
-            ]
-            print("Running benchmark:", " ".join(command))
-            result = run_command(command, cwd=REPO_ROOT, report_path=report_path)
-            result["suite"] = benchmark_dir.name
-            result["benchmark_path"] = str(benchmark_file.relative_to(benchmark_dir / "train"))
-            results.append(result)
-    return results
-
-
-def display_path(p: Path) -> str:
-    # Used for serializing benchmark roots into data.json. Falls back to the
-    # absolute path when the benchmarks dir lives outside REPO_ROOT (e.g. when
-    # the combined-nightly orchestrator points POACH_BENCHMARKS_DIR at a
-    # shared clone above the worktree).
-    try:
-        return str(p.relative_to(REPO_ROOT))
-    except ValueError:
-        return str(p)
-
-
-def summarize_report(report: dict[str, Any]) -> dict[str, int]:
-    rule_running_millis = 0
-    extraction_millis = 0
-    other_millis = 0
-
-    for timing in report["timings"]:
-        if "running_rules" in timing["tags"]:
-            rule_running_millis += timing["total"]
-        elif "extraction" in timing["tags"]:
-            extraction_millis += timing["total"]
-        else:
-            other_millis += timing["total"]
-
-    return {
-        "rule_running_millis": rule_running_millis,
-        "extraction_millis": extraction_millis,
-        "other_millis": other_millis,
-        "timing_steps": len(report["timings"]),
-    }
-
-
-def aggregate_reports(
-    benchmark_dirs: list[Path], benchmark_results: list[dict[str, Any]]
-) -> dict[str, Any]:
-    if not benchmark_results:
-        raise SystemExit(f"No report files were generated under {REPORT_OUTPUT_DIR}")
-
-    benchmark_root = benchmark_dirs[0].parent
-    suites = []
-    for benchmark_dir in benchmark_dirs:
-        suite_results = [
-            result for result in benchmark_results if result["suite"] == benchmark_dir.name
-        ]
-        suites.append(
-            {
-                "name": benchmark_dir.name,
-                "benchmark_root": display_path(benchmark_dir),
-                "summary": {
-                    "total_time_seconds": sum(
-                        result["time_seconds"] for result in suite_results
-                    ),
-                },
-                "reports": [
-                    {
-                        "suite": benchmark_dir.name,
-                        "benchmark_path": result["benchmark_path"],
-                        "path": str(
-                            Path(result["report_path"]).relative_to(OUTPUT_DIR)
-                        ),
-                        "time_seconds": result["time_seconds"],
-                        "timing_summary": summarize_report(
-                            json.loads(
-                                Path(result["report_path"]).read_text(
-                                    encoding="utf-8"
-                                )
-                            )
-                        ),
-                    }
-                    for result in suite_results
-                ],
-            }
-        )
 
+# Determine directories
+SCRIPT_DIR = Path(__file__).resolve().parent
+POACH_ROOT = SCRIPT_DIR.parent
+NIGHTLY_DIR = POACH_ROOT / "nightly"
+POACH_BINARY = POACH_ROOT / "target" / "release" / "poach"
+
+def main(benchmark_dir):
+  print(benchmark_dir)
+
+  (benchmark_results, failing_benchmarks) = run_benchmarks(benchmark_dir)
+
+  data = {
+    "generated_at": datetime.now(timezone.utc).isoformat(),
+    "failing_benchmarks": [str(b) for b in failing_benchmarks],
+    "data": benchmark_results
+  }
+  data_out_path = NIGHTLY_DIR / "output" / "data" / "data.json"
+  data_out_path.parent.mkdir(parents=True, exist_ok=True)
+  data_out_path.write_text(json.dumps(data, indent=2), encoding="utf-8")
+
+def run_command(cmd):
+  started = time.perf_counter()
+  cmd_result = subprocess.run(
+    cmd,
+    cwd=POACH_ROOT,
+    capture_output=True,
+    text=True # decode stderr/stdout as string instead of raw bytes
+  )
+  if cmd_result.returncode != 0:
+    time_seconds = time.perf_counter() - started
+    print(f"Command failed after {time_seconds:.2f}s: {' '.join(cmd)}", file=sys.stderr)
     return {
-        "generated_at": datetime.now(timezone.utc).isoformat(),
-        "benchmark_root": display_path(benchmark_root),
-        "summary": {
-            "benchmark_count": len(benchmark_results),
-            "total_time_seconds": sum(
-                result["time_seconds"] for result in benchmark_results
-            ),
-        },
-        "suites": suites,
-        "reports": [report for suite in suites for report in suite["reports"]],
+      "cmd": cmd,
+      "status": "error"
     }
 
+  report = json.loads(cmd_result.stderr)
+
+  return {
+    "cmd": " ".join(cmd),
+    "status": "success",
+    "report": summarize_report(report),
+    "wall_time_s": time.perf_counter() - started
+  }
+  
+def summarize_report(report):
+  # aggregate timing steps by type
+  rule_ms = 0
+  extraction_ms = 0
+  other_ms = 0
+  total_ms = 0
+  for time_step in report["timings"]:
+    total_ms += time_step["total"]
+    if "running_rules" in time_step["tags"]:
+      rule_ms += time_step["total"]
+    elif "extraction" in time_step["tags"]:
+      extraction_ms += time_step["total"]
+    else:
+      other_ms += time_step["total"]
+
+  # No sizes in vanilla egglog reports
+
+  return {
+    "rule_ms": rule_ms,
+    "extraction_ms": extraction_ms,
+    "other_ms": other_ms,
+    "timing_steps": len(report["timings"])
+  }
+
+def run_benchmarks(benchmark_dir):
+  report_dir = NIGHTLY_DIR / "reports"
+  report_dir.mkdir(parents=True, exist_ok=True)
+
+  # Find benchmarks
+  # benchmark_dir is the root of the benchmark directory 
+  benchmarks = list(Path(benchmark_dir).rglob("train/*.egg"))
+  # For this treatment, we don't do anything at train time,
+  # we just use the train benchmarks at serve time
+
+  results = []
+  failing_benchmarks = []
+  for benchmark in benchmarks:
+    if len(results) > 10:
+      break
+    relative_path = benchmark.relative_to(benchmark_dir)
+    suite_name = str(relative_path.parent)
+    benchmark_name = relative_path.name
+    command = [
+      str(POACH_BINARY),
+      "serve",
+      "--debug",
+      "EMPTY.MODEL",
+      "single",
+      str(benchmark)
+    ]
+    
+    result = run_command(command)
+    result["benchmark_name"] = benchmark_name
+    result["suite_name"] = suite_name
+    if result["status"] == "success":
+      print(f"Success: {benchmark_name}")
+      results.append(result)
+    else:
+      failing_benchmarks.append(relative_path)
+
+  return (results, failing_benchmarks)
 
 if __name__ == "__main__":
-    try:
-        main()
-    except subprocess.CalledProcessError as err:
-        print(f"Command failed with exit code {err.returncode}: {err.cmd}", file=sys.stderr)
-        raise
+  if len(sys.argv) != 2:
+    raise SystemExit(f"Usage: nightly.py <benchmark-dir>")
+  
+  main(sys.argv[1])
\ No newline at end of file
diff --git a/infra/nightly.sh b/infra/nightly.sh
index 8a04ee749..f4ce9ffe0 100755
--- a/infra/nightly.sh
+++ b/infra/nightly.sh
@@ -13,7 +13,7 @@ export PATH=~/.cargo/bin:$PATH
 
 # Ensure we start from a clean slate
 rm -rf nightly
-mkdir -p nightly/output nightly/tmp
+mkdir -p nightly
 
 # Standalone runs do their own setup (toolchain + benchmarks clone). When
 # driven by the combined orchestrator, POACH_NIGHTLY_COMBINED=1 and the

From 4a8edc95475d5ff1889caef8dd39af5804bfbd81 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 14:23:18 -0700
Subject: [PATCH 18/31] remove early exit

---
 infra/nightly.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index 9af258dba..c55a1139a 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -89,8 +89,6 @@ def run_benchmarks(benchmark_dir):
   results = []
   failing_benchmarks = []
   for benchmark in benchmarks:
-    if len(results) > 10:
-      break
     relative_path = benchmark.relative_to(benchmark_dir)
     suite_name = str(relative_path.parent)
     benchmark_name = relative_path.name

From cce3f12dd73d2c28eb027ceea179d8ae8149c335 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 15:26:55 -0700
Subject: [PATCH 19/31] frontend

---
 infra/nightly-resources/web/index.html |  14 +-
 infra/nightly-resources/web/index.js   | 218 +++++++++++--------------
 infra/nightly-resources/web/table.js   |  45 +++++
 infra/nightly.py                       |   2 +-
 4 files changed, 145 insertions(+), 134 deletions(-)
 create mode 100644 infra/nightly-resources/web/table.js

diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index ed9d726eb..45a0b73f9 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -19,19 +19,7 @@ <h2>Summary</h2>
           <h2>Reports</h2>
           <div id="suite-tabs"></div>
           <div id="active-suite-summary"></div>
-          <table>
-            <thead>
-              <tr id="benchmarks-header">
-                <th data-sort-key="benchmark_path">Path</th>
-                <th data-sort-key="time_seconds">Wall Time</th>
-                <th data-sort-key="timing_summary.rule_running_millis">Rule Running</th>
-                <th data-sort-key="timing_summary.extraction_millis">Extraction</th>
-                <th data-sort-key="timing_summary.other_millis">Other</th>
-                <th data-sort-key="timing_summary.timing_steps"># Timing Steps</th>
-              </tr>
-            </thead>
-            <tbody id="benchmarks-body"></tbody>
-          </table>
+          <div id="active-suite-table"></div>
         </div>
       </div>
     </main>
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 51c4d3238..96826b66c 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -1,148 +1,126 @@
-import { formatMillis } from "./util.js";
+import { convertToTable } from "./table.js";
+
+const STATE = {
+  activeSuite: null,
+};
+
+const GLOBAL_DATA = {
+  data: null,
+  suites: null,
+};
 
-let suites = [];
-let activeSuiteName = null;
-let sortKey = "benchmark_path";
-let sortDir = "asc";
 load();
-installHeaderSortHandlers();
 
 async function load() {
   const statusNode = document.querySelector("#status");
 
-  try {
-    const response = await fetch("./data/data.json");
-    if (!response.ok) {
-      throw new Error(`Failed to load data.json (${response.status})`);
-    }
-
-    const data = await response.json();
-    suites = [...data.suites].sort((a, b) => a.name.localeCompare(b.name));
-    activeSuiteName = suites[0]?.name ?? null;
-    statusNode.textContent = "Loaded data/data.json";
-    renderSummary(data);
-    renderSuites();
-  } catch (error) {
+  const response = await fetch("./data/data.json");
+  if (!response.ok) {
     statusNode.textContent = `Failed to load data/data.json: ${error}`;
+    return;
   }
-}
 
-function renderSummary(data) {
-  let ruleRunningMillis = 0;
-  let extractionMillis = 0;
-  let otherMillis = 0;
+  GLOBAL_DATA.data = await response.json();
+  GLOBAL_DATA.suites = [
+    ...new Set(GLOBAL_DATA.data.passing_benchmarks.map((x) => x.suite_name)),
+  ].sort();
+  STATE.activeSuite = GLOBAL_DATA.suites[0] ?? null;
 
-  for (const { timing_summary } of data.reports) {
-    ruleRunningMillis += timing_summary.rule_running_millis;
-    extractionMillis += timing_summary.extraction_millis;
-    otherMillis += timing_summary.other_millis;
-  }
+  statusNode.textContent = "Loaded data/data.json";
+  renderSummary();
 
-  document.querySelector("#summary-text").textContent =
-    `${data.summary.benchmark_count} benchmarks across ${data.suites.length} suites | ` +
-    `Nightly time: ${data.summary.total_time_seconds.toFixed(1)} s | ` +
-    `Rule running: ${ruleRunningMillis} ms | ` +
-    `Extraction: ${extractionMillis} ms | ` +
-    `Other: ${otherMillis} ms`;
+  renderSuiteSelectors();
+  renderTable();
 }
 
-function renderSuites() {
-  document.querySelector("#suite-tabs").innerHTML = suites
-    .map((suite) => {
-      return `
-        <button
-          type="button"
-          class="suite-tab${suite.name === activeSuiteName ? " is-active" : ""}"
-          data-suite-name="${suite.name}"
-        >
-          ${suite.name}
-        </button>
-      `;
-    })
-    .join("");
+function renderSummary() {
+  let ruleMs = 0;
+  let extractMs = 0;
+  let otherMs = 0;
 
-  for (const button of document.querySelectorAll(".suite-tab")) {
-    button.addEventListener("click", () => {
-      activeSuiteName = button.dataset.suiteName;
-      renderSuites();
-    });
+  for (const reportSummary of GLOBAL_DATA.data.passing_benchmarks) {
+    ruleMs += reportSummary.report.rule_ms;
+    extractMs += reportSummary.report.extraction_ms;
+    otherMs += reportSummary.report.other_ms;
   }
 
-  const activeSuite = suites.find((suite) => suite.name === activeSuiteName);
-  if (!activeSuite) {
-    document.querySelector("#active-suite-summary").textContent = "";
-    document.querySelector("#benchmarks-body").innerHTML = "";
-    return;
-  }
+  const numPassing = GLOBAL_DATA.data.passing_benchmarks.length;
+  const numFailing = GLOBAL_DATA.data.failing_benchmarks.length;
+  const totalTime = GLOBAL_DATA.data.passing_benchmarks
+    .map((x) => x.wall_time_s)
+    .reduce((a, b) => a + b, 0);
 
-  document.querySelector("#active-suite-summary").innerHTML = `
-    <div class="suite-header">
-      <h3>${activeSuite.name}</h3>
-      <p>${activeSuite.reports.length} benchmarks | ${activeSuite.summary.total_time_seconds.toFixed(1)} s</p>
-    </div>
-  `;
-  document.querySelector("#benchmarks-body").innerHTML = renderRows(
-    sortReports(activeSuite.reports),
-  );
-  updateHeaderIndicators();
+  document.querySelector("#summary-text").textContent =
+    `Passing Benchmarks: ${numPassing} | ` +
+    `Failing Benchmarks: ${numFailing} | ` +
+    `Nightly time: ${totalTime.toFixed(1)} s | ` +
+    `Rule running: ${(ruleMs / 1000).toFixed(1)} s | ` +
+    `Extraction: ${(extractMs / 1000).toFixed(1)} s | ` +
+    `Other: ${(otherMs / 1000).toFixed(1)} s`;
 }
 
-function getSortValue(report, key) {
-  return key.split(".").reduce((acc, part) => acc?.[part], report) ?? 0;
-}
+function renderSuiteSelectors() {
+  document.querySelector("#suite-tabs").innerHTML = GLOBAL_DATA.suites
+    .map(
+      (suite) =>
+        `<button
+    type="button"
+    class="suite-tab ${suite === STATE.activeSuite ? " is-active" : ""}"
+    data-suite-name="${suite}"
+      >
+      ${suite}
+      </button>
+      `,
+    )
+    .join("");
 
-function sortReports(reports) {
-  const sorted = [...reports];
-  const dir = sortDir === "asc" ? 1 : -1;
-  sorted.sort((a, b) => {
-    const av = getSortValue(a, sortKey);
-    const bv = getSortValue(b, sortKey);
-    if (typeof av === "string" || typeof bv === "string") {
-      return String(av).localeCompare(String(bv)) * dir;
-    }
-    return (av - bv) * dir;
-  });
-  return sorted;
-}
+  for (const button of document.querySelectorAll(".suite-tab")) {
+    button.addEventListener("click", () => {
+      STATE.activeSuite = button.dataset.suiteName;
 
-function installHeaderSortHandlers() {
-  for (const th of document.querySelectorAll("#benchmarks-header th")) {
-    th.style.cursor = "pointer";
-    th.addEventListener("click", () => {
-      const key = th.dataset.sortKey;
-      if (sortKey === key) {
-        sortDir = sortDir === "asc" ? "desc" : "asc";
-      } else {
-        sortKey = key;
-        sortDir = "asc";
+      for (const btn of document.querySelectorAll(".suite-tab")) {
+        btn.classList.toggle(
+          "is-active",
+          btn.dataset.suiteName === STATE.activeSuite,
+        );
       }
-      renderSuites();
+
+      renderTable();
     });
   }
 }
 
-function updateHeaderIndicators() {
-  for (const th of document.querySelectorAll("#benchmarks-header th")) {
-    const label = th.textContent.replace(/[ ▲▼]+$/, "");
-    const arrow =
-      th.dataset.sortKey === sortKey ? (sortDir === "asc" ? " ▲" : " ▼") : "";
-    th.textContent = label + arrow;
-  }
-}
+function renderTable() {
+  const benchmarks = GLOBAL_DATA.data.passing_benchmarks.filter(
+    (x) => x.suite_name === STATE.activeSuite,
+  );
+  const totalTime = benchmarks
+    .map((x) => x.wall_time_s)
+    .reduce((a, b) => a + b, 0);
 
-function renderRows(reports) {
-  return reports
-    .map(({ benchmark_path, time_seconds, timing_summary }) => {
-      return `
-        <tr>
-          <td>${benchmark_path}</td>
-          <td>${time_seconds.toFixed(3)} s</td>
-          <td>${formatMillis(timing_summary.rule_running_millis)}</td>
-          <td>${formatMillis(timing_summary.extraction_millis)}</td>
-          <td>${formatMillis(timing_summary.other_millis)}</td>
-          <td>${timing_summary.timing_steps}</td>
-        </tr>
-      `;
-    })
-    .join("");
+  document.querySelector("#active-suite-summary").innerHTML = `
+  <div>
+    <h3>${STATE.activeSuite}</h3>
+    <p>${benchmarks.length} benchmarks | ${totalTime} s</p>
+  </div>`;
+
+  const columns = [
+    "Benchmark",
+    "Wall Time (s)",
+    "Rules (ms)",
+    "Extraction (ms)",
+    "Other (ms)",
+  ];
+
+  const rows = benchmarks.map((b) => ({
+    Benchmark: b.benchmark_name,
+    "Wall Time (s)": b.wall_time_s,
+    "Rules (ms)": b.report.rule_ms,
+    "Extraction (ms)": b.report.extraction_ms,
+    "Other (ms)": b.report.other_ms,
+  }));
+
+  const tableDiv = document.querySelector("#active-suite-table");
+  tableDiv.innerHTML = "";
+  tableDiv.appendChild(convertToTable(columns, rows));
 }
diff --git a/infra/nightly-resources/web/table.js b/infra/nightly-resources/web/table.js
new file mode 100644
index 000000000..f2c1e583a
--- /dev/null
+++ b/infra/nightly-resources/web/table.js
@@ -0,0 +1,45 @@
+/**
+ * Given a list of column names and a list of row objects, produce an
+ * HTML table that displays the data
+ *
+ * @param {String[]} columns
+ * Defines the set of columns to be displayed and their order.
+ * If empty, an empty table element will be returned
+ *
+ * @param {Object[]} rows
+ * Each object corresponds to one row in the table.
+ * Each object maps column name -> cell value.
+ * Values should be numbers or strings only.
+ * Rows may contain keys not listed in `columns` and may be missing `columns`
+ * Extra values are ignored. Missing values are displayed as `-` in the table
+ * If empty, a table with only the header row will be returned
+ *
+ * @return An HTML <table> DOM element
+ */
+export function convertToTable(columns, rows) {
+  const table = document.createElement("table");
+
+  const thead = document.createElement("thead");
+  const headerRow = document.createElement("tr");
+  for (const column of columns) {
+    const th = document.createElement("th");
+    th.textContent = column;
+    headerRow.appendChild(th);
+  }
+  thead.appendChild(headerRow);
+  table.appendChild(thead);
+
+  const tbody = document.createElement("tbody");
+  for (const row of rows) {
+    const tr = document.createElement("tr");
+    for (const column of columns) {
+      const td = document.createElement("td");
+      td.textContent = row[column] ?? "-"; // We want to leave 0 and "" intact, so don't use ||
+      tr.appendChild(td);
+    }
+    tbody.appendChild(tr);
+  }
+  table.appendChild(tbody);
+
+  return table;
+}
diff --git a/infra/nightly.py b/infra/nightly.py
index c55a1139a..d100c8898 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -21,7 +21,7 @@ def main(benchmark_dir):
   data = {
     "generated_at": datetime.now(timezone.utc).isoformat(),
     "failing_benchmarks": [str(b) for b in failing_benchmarks],
-    "data": benchmark_results
+    "passing_benchmarks": benchmark_results
   }
   data_out_path = NIGHTLY_DIR / "output" / "data" / "data.json"
   data_out_path.parent.mkdir(parents=True, exist_ok=True)

From c2bae370ba852f7303b14e40170827b39f24c616 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 15:46:21 -0700
Subject: [PATCH 20/31] sortable table

---
 infra/nightly-resources/web/table.js | 83 ++++++++++++++++++++++------
 1 file changed, 67 insertions(+), 16 deletions(-)

diff --git a/infra/nightly-resources/web/table.js b/infra/nightly-resources/web/table.js
index f2c1e583a..e63e36b4b 100644
--- a/infra/nightly-resources/web/table.js
+++ b/infra/nightly-resources/web/table.js
@@ -15,31 +15,82 @@
  * If empty, a table with only the header row will be returned
  *
  * @return An HTML <table> DOM element
+ * Sortable by column, defaults to sorted by first column
  */
 export function convertToTable(columns, rows) {
   const table = document.createElement("table");
 
-  const thead = document.createElement("thead");
-  const headerRow = document.createElement("tr");
-  for (const column of columns) {
-    const th = document.createElement("th");
-    th.textContent = column;
-    headerRow.appendChild(th);
+  const STATE = {
+    sortCol: columns[0],
+    sortDir: "asc",
+  };
+
+  function sortedRows() {
+    const dir = STATE.sortDir === "asc" ? 1 : -1;
+    return [...rows].sort((a, b) => {
+      // Look up value to sort by in each row
+      const av = a[STATE.sortCol];
+      const bv = b[STATE.sortCol];
+
+      // First check if either/both are missing. Sort missing values to the bottom
+      const aMissing = av === undefined || av === null;
+      const bMissing = bv === undefined || bv === null;
+      if (aMissing && bMissing) return 0;
+      if (aMissing) return 1;
+      if (bMissing) return -1;
+
+      // Compare
+      return (av > bv ? 1 : av < bv ? -1 : 0) * dir;
+    });
   }
-  thead.appendChild(headerRow);
-  table.appendChild(thead);
 
-  const tbody = document.createElement("tbody");
-  for (const row of rows) {
-    const tr = document.createElement("tr");
+  function toggleSortDir() {
+    if (STATE.sortDir === "asc") {
+      STATE.sortDir = "desc";
+    } else {
+      STATE.sortDir = "asc";
+    }
+  }
+
+  function render() {
+    // Reset table element before rebuilding
+    table.innerHTML = "";
+
+    const thead = document.createElement("thead");
+    const headerRow = document.createElement("tr");
     for (const column of columns) {
-      const td = document.createElement("td");
-      td.textContent = row[column] ?? "-"; // We want to leave 0 and "" intact, so don't use ||
-      tr.appendChild(td);
+      const th = document.createElement("th");
+      const arrow =
+        column === STATE.sortCol ? (STATE.sortDir === "asc" ? " ▲" : " ▼") : "";
+      th.textContent = column + arrow;
+      th.style.cursor = "pointer";
+      th.addEventListener("click", () => {
+        if (STATE.sortCol === column) {
+          toggleSortDir();
+        } else {
+          STATE.sortCol = column;
+          STATE.sortDir = "asc";
+        }
+        render();
+      });
+      headerRow.appendChild(th);
+    }
+    thead.appendChild(headerRow);
+    table.appendChild(thead);
+
+    const tbody = document.createElement("tbody");
+    for (const row of sortedRows()) {
+      const tr = document.createElement("tr");
+      for (const column of columns) {
+        const td = document.createElement("td");
+        td.textContent = row[column] ?? "-"; // We want to leave 0 and "" intact, so don't use ||
+        tr.appendChild(td);
+      }
+      tbody.appendChild(tr);
     }
-    tbody.appendChild(tr);
+    table.appendChild(tbody);
   }
-  table.appendChild(tbody);
 
+  render();
   return table;
 }

From 93d080bd1579493f133426d4f648a4c64d1ca916 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 16:35:41 -0700
Subject: [PATCH 21/31] round wall time

---
 infra/nightly-resources/web/index.js | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 96826b66c..8c47b30fe 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -114,7 +114,7 @@ function renderTable() {
 
   const rows = benchmarks.map((b) => ({
     Benchmark: b.benchmark_name,
-    "Wall Time (s)": b.wall_time_s,
+    "Wall Time (s)": b.wall_time_s.toFixed(2),
     "Rules (ms)": b.report.rule_ms,
     "Extraction (ms)": b.report.extraction_ms,
     "Other (ms)": b.report.other_ms,

From 5a742be185d10d84c05494f5558f1c10fd3057be Mon Sep 17 00:00:00 2001
From: Noah Huck <noahhuck@cs.washington.edu>
Date: Tue, 12 May 2026 17:57:34 -0700
Subject: [PATCH 22/31] register custom scheduler with egraph

---
 src/lib.rs | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/lib.rs b/src/lib.rs
index 9db873cf0..5db3a2e5d 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -363,8 +363,10 @@ impl Default for EGraph {
         eg.rulesets
             .insert("".into(), Ruleset::Rules(Default::default()));
 
-        // support get-size! macro for Herbie
+        // support get-size! macro, custom scheduler for Herbie
         eg.add_primitive(get_size_prim::GetSizePrimitive);
+        eg.add_command("run-schedule".into(), Arc::new(RunExtendedSchedule))
+            .unwrap();
 
         eg
     }

From bd34aa41ec411aea38581003d6f0ca4fac56bea3 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 21:00:03 -0700
Subject: [PATCH 23/31] serialize durations as micros

---
 src/report.rs | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/src/report.rs b/src/report.rs
index 0d85dc6fd..36b4412dc 100644
--- a/src/report.rs
+++ b/src/report.rs
@@ -28,7 +28,7 @@ pub struct RunReport {
 struct TimingStep {
     name: String,
     tags: Vec<String>,
-    #[serde(with = "serde_millis")]
+    #[serde(serialize_with = "serialize_duration_total")]
     total: Duration,
     #[serde(serialize_with = "serialize_duration_breakdown")]
     breakdown: Vec<Duration>,
@@ -122,11 +122,18 @@ impl Display for TimingStep {
     }
 }
 
+fn serialize_duration_total<S>(total: &Duration, serializer: S) -> Result<S::Ok, S::Error>
+where
+    S: Serializer,
+{
+    total.as_micros().serialize(serializer)
+}
+
 fn serialize_duration_breakdown<S>(breakdown: &[Duration], serializer: S) -> Result<S::Ok, S::Error>
 where
     S: Serializer,
 {
-    let breakdown_millis: Vec<_> = breakdown.iter().map(Duration::as_millis).collect();
+    let breakdown_millis: Vec<_> = breakdown.iter().map(Duration::as_micros).collect();
     breakdown_millis.serialize(serializer)
 }
 

From f94ca002d009e161d111d70a05f4460b741a424e Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 21:26:58 -0700
Subject: [PATCH 24/31] micros everywhere

---
 infra/nightly-resources/web/index.js | 40 ++++++++++++++--------------
 infra/nightly.py                     | 34 ++++++++++++-----------
 src/report.rs                        | 12 ++++++---
 3 files changed, 46 insertions(+), 40 deletions(-)

diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 8c47b30fe..a4752efb1 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -34,29 +34,29 @@ async function load() {
 }
 
 function renderSummary() {
-  let ruleMs = 0;
-  let extractMs = 0;
-  let otherMs = 0;
+  let ruleMicros = 0;
+  let extractMicros = 0;
+  let otherMicros = 0;
 
   for (const reportSummary of GLOBAL_DATA.data.passing_benchmarks) {
-    ruleMs += reportSummary.report.rule_ms;
-    extractMs += reportSummary.report.extraction_ms;
-    otherMs += reportSummary.report.other_ms;
+    ruleMicros += reportSummary.report.rule_micros;
+    extractMicros += reportSummary.report.extraction_micros;
+    otherMicros += reportSummary.report.other_micros;
   }
 
   const numPassing = GLOBAL_DATA.data.passing_benchmarks.length;
   const numFailing = GLOBAL_DATA.data.failing_benchmarks.length;
   const totalTime = GLOBAL_DATA.data.passing_benchmarks
-    .map((x) => x.wall_time_s)
+    .map((x) => x.wall_time_micros)
     .reduce((a, b) => a + b, 0);
 
   document.querySelector("#summary-text").textContent =
     `Passing Benchmarks: ${numPassing} | ` +
     `Failing Benchmarks: ${numFailing} | ` +
-    `Nightly time: ${totalTime.toFixed(1)} s | ` +
-    `Rule running: ${(ruleMs / 1000).toFixed(1)} s | ` +
-    `Extraction: ${(extractMs / 1000).toFixed(1)} s | ` +
-    `Other: ${(otherMs / 1000).toFixed(1)} s`;
+    `Nightly time: ${totalTime} μs | ` +
+    `Rule running: ${ruleMicros} μs | ` +
+    `Extraction: ${extractMicros} μs | ` +
+    `Other: ${otherMicros} μs`;
 }
 
 function renderSuiteSelectors() {
@@ -95,7 +95,7 @@ function renderTable() {
     (x) => x.suite_name === STATE.activeSuite,
   );
   const totalTime = benchmarks
-    .map((x) => x.wall_time_s)
+    .map((x) => x.wall_time_micros)
     .reduce((a, b) => a + b, 0);
 
   document.querySelector("#active-suite-summary").innerHTML = `
@@ -106,18 +106,18 @@ function renderTable() {
 
   const columns = [
     "Benchmark",
-    "Wall Time (s)",
-    "Rules (ms)",
-    "Extraction (ms)",
-    "Other (ms)",
+    "Wall Time (μs)",
+    "Rules (μs)",
+    "Extraction (μs)",
+    "Other (μs)",
   ];
 
   const rows = benchmarks.map((b) => ({
     Benchmark: b.benchmark_name,
-    "Wall Time (s)": b.wall_time_s.toFixed(2),
-    "Rules (ms)": b.report.rule_ms,
-    "Extraction (ms)": b.report.extraction_ms,
-    "Other (ms)": b.report.other_ms,
+    "Wall Time (μs)": b.wall_time_micros,
+    "Rules (μs)": b.report.rule_micros,
+    "Extraction (μs)": b.report.extraction_micros,
+    "Other (μs)": b.report.other_micros,
   }));
 
   const tableDiv = document.querySelector("#active-suite-table");
diff --git a/infra/nightly.py b/infra/nightly.py
index d100c8898..7802fddcc 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -28,19 +28,21 @@ def main(benchmark_dir):
   data_out_path.write_text(json.dumps(data, indent=2), encoding="utf-8")
 
 def run_command(cmd):
-  started = time.perf_counter()
+  started = time.perf_counter_ns()
   cmd_result = subprocess.run(
     cmd,
     cwd=POACH_ROOT,
     capture_output=True,
     text=True # decode stderr/stdout as string instead of raw bytes
   )
+  # Clock granularity is ~50-100 ns.
+  # Report as micros to avoid reporting false precision.
+  time_micros = (time.perf_counter_ns() - started) // 1000
   if cmd_result.returncode != 0:
-    time_seconds = time.perf_counter() - started
-    print(f"Command failed after {time_seconds:.2f}s: {' '.join(cmd)}", file=sys.stderr)
     return {
       "cmd": cmd,
-      "status": "error"
+      "status": "error",
+      "wall_time_micros": time_micros
     }
 
   report = json.loads(cmd_result.stderr)
@@ -49,30 +51,30 @@ def run_command(cmd):
     "cmd": " ".join(cmd),
     "status": "success",
     "report": summarize_report(report),
-    "wall_time_s": time.perf_counter() - started
+    "wall_time_micros": time_micros
   }
   
 def summarize_report(report):
   # aggregate timing steps by type
-  rule_ms = 0
-  extraction_ms = 0
-  other_ms = 0
-  total_ms = 0
+  rule_micros = 0
+  extraction_micros = 0
+  other_micros = 0
+  total_micros = 0
   for time_step in report["timings"]:
-    total_ms += time_step["total"]
+    total_micros += time_step["total"]
     if "running_rules" in time_step["tags"]:
-      rule_ms += time_step["total"]
+      rule_micros += time_step["total"]
     elif "extraction" in time_step["tags"]:
-      extraction_ms += time_step["total"]
+      extraction_micros += time_step["total"]
     else:
-      other_ms += time_step["total"]
+      other_micros += time_step["total"]
 
   # No sizes in vanilla egglog reports
 
   return {
-    "rule_ms": rule_ms,
-    "extraction_ms": extraction_ms,
-    "other_ms": other_ms,
+    "rule_micros": rule_micros,
+    "extraction_micros": extraction_micros,
+    "other_micros": other_micros,
     "timing_steps": len(report["timings"])
   }
 
diff --git a/src/report.rs b/src/report.rs
index 36b4412dc..657664418 100644
--- a/src/report.rs
+++ b/src/report.rs
@@ -109,13 +109,13 @@ impl Display for TimingStep {
         writeln!(f, "timing:")?;
         writeln!(f, "  name: {}", self.name)?;
         writeln!(f, "  tags: {:?}", self.tags)?;
-        writeln!(f, "  total: {}", self.total.as_secs_f64())?;
+        writeln!(f, "  total: {}", self.total.as_micros())?;
         writeln!(
             f,
             "  breakdown: {:?}",
             self.breakdown
                 .iter()
-                .map(Duration::as_millis)
+                .map(Duration::as_micros)
                 .collect::<Vec<_>>()
         )?;
         Ok(())
@@ -126,6 +126,8 @@ fn serialize_duration_total<S>(total: &Duration, serializer: S) -> Result<S::Ok,
 where
     S: Serializer,
 {
+    // Clock granularity is ~50-100 ns.
+    // Serialize as micros to avoid reporting false precision.
     total.as_micros().serialize(serializer)
 }
 
@@ -133,8 +135,10 @@ fn serialize_duration_breakdown<S>(breakdown: &[Duration], serializer: S) -> Res
 where
     S: Serializer,
 {
-    let breakdown_millis: Vec<_> = breakdown.iter().map(Duration::as_micros).collect();
-    breakdown_millis.serialize(serializer)
+    // Clock granularity is ~50-100 ns.
+    // Serialize as micros to avoid reporting false precision.
+    let breakdown_micros: Vec<_> = breakdown.iter().map(Duration::as_micros).collect();
+    breakdown_micros.serialize(serializer)
 }
 
 fn write_metric_section<T>(f: &mut Formatter<'_>, title: &str, metrics: &[T]) -> fmt::Result

From cacb7ed3adcaed793afcdc354af6a074924c2315 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 21:45:28 -0700
Subject: [PATCH 25/31] toggle time display

---
 infra/nightly-resources/web/index.html |  5 ++
 infra/nightly-resources/web/index.js   | 73 +++++++++++++++++++-------
 infra/nightly-resources/web/style.css  | 15 ++++++
 3 files changed, 73 insertions(+), 20 deletions(-)

diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index 45a0b73f9..d4a2d12c2 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -9,6 +9,11 @@
   <body>
     <main class="page">
       <h1>Poach Nightly</h1>
+      <fieldset id="time-display-selector">
+        <legend>Time Display</legend>
+        <label><input type="radio" name="time-display" value="raw" checked /> Raw</label>
+        <label><input type="radio" name="time-display" value="readable" /> Readable</label>
+      </fieldset>
       <p id="status">Loading <code>data/data.json</code>...</p>
       <div id="contents">
         <div id="summary">
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index a4752efb1..9b5a80545 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -2,6 +2,7 @@ import { convertToTable } from "./table.js";
 
 const STATE = {
   activeSuite: null,
+  timeDisplay: "raw",
 };
 
 const GLOBAL_DATA = {
@@ -21,16 +22,49 @@ async function load() {
   }
 
   GLOBAL_DATA.data = await response.json();
+  statusNode.textContent = "Loaded data/data.json";
+
   GLOBAL_DATA.suites = [
     ...new Set(GLOBAL_DATA.data.passing_benchmarks.map((x) => x.suite_name)),
   ].sort();
   STATE.activeSuite = GLOBAL_DATA.suites[0] ?? null;
 
-  statusNode.textContent = "Loaded data/data.json";
-  renderSummary();
+  // Set up interactive elements
+  setupSuiteSelectors();
+  setupTimeDisplaySelector();
 
-  renderSuiteSelectors();
-  renderTable();
+  render();
+}
+
+function setupTimeDisplaySelector() {
+  for (const radio of document.querySelectorAll('input[name="time-display"]')) {
+    radio.addEventListener("change", () => {
+      if (radio.checked) {
+        STATE.timeDisplay = radio.value;
+        render();
+      }
+    });
+  }
+}
+
+function displayTime(rawValue) {
+  const ONE_MIN = 60000000;
+  const ONE_SEC = 1000000;
+  const ONE_MILLI = 1000;
+  if (STATE.timeDisplay === "raw") {
+    return `${rawValue} μs`;
+  } else {
+    console.assert(STATE.timeDisplay === "readable");
+    if (rawValue >= ONE_MIN) {
+      return `${(rawValue / ONE_MIN).toFixed(2)} min`;
+    } else if (rawValue >= ONE_SEC) {
+      return `${(rawValue / ONE_SEC).toFixed(2)} s`;
+    } else if (rawValue >= ONE_MILLI) {
+      return `${(rawValue / ONE_MILLI).toFixed(2)} ms`;
+    } else {
+      return `${rawValue} μs`;
+    }
+  }
 }
 
 function renderSummary() {
@@ -53,13 +87,13 @@ function renderSummary() {
   document.querySelector("#summary-text").textContent =
     `Passing Benchmarks: ${numPassing} | ` +
     `Failing Benchmarks: ${numFailing} | ` +
-    `Nightly time: ${totalTime} μs | ` +
-    `Rule running: ${ruleMicros} μs | ` +
-    `Extraction: ${extractMicros} μs | ` +
-    `Other: ${otherMicros} μs`;
+    `Nightly time: ${displayTime(totalTime)} | ` +
+    `Rule running: ${displayTime(ruleMicros)} | ` +
+    `Extraction: ${displayTime(extractMicros)} | ` +
+    `Other: ${displayTime(otherMicros)}`;
 }
 
-function renderSuiteSelectors() {
+function setupSuiteSelectors() {
   document.querySelector("#suite-tabs").innerHTML = GLOBAL_DATA.suites
     .map(
       (suite) =>
@@ -104,23 +138,22 @@ function renderTable() {
     <p>${benchmarks.length} benchmarks | ${totalTime} s</p>
   </div>`;
 
-  const columns = [
-    "Benchmark",
-    "Wall Time (μs)",
-    "Rules (μs)",
-    "Extraction (μs)",
-    "Other (μs)",
-  ];
+  const columns = ["Benchmark", "Wall Time", "Rules", "Extraction", "Other"];
 
   const rows = benchmarks.map((b) => ({
     Benchmark: b.benchmark_name,
-    "Wall Time (μs)": b.wall_time_micros,
-    "Rules (μs)": b.report.rule_micros,
-    "Extraction (μs)": b.report.extraction_micros,
-    "Other (μs)": b.report.other_micros,
+    "Wall Time": displayTime(b.wall_time_micros),
+    Rules: displayTime(b.report.rule_micros),
+    Extraction: displayTime(b.report.extraction_micros),
+    Other: displayTime(b.report.other_micros),
   }));
 
   const tableDiv = document.querySelector("#active-suite-table");
   tableDiv.innerHTML = "";
   tableDiv.appendChild(convertToTable(columns, rows));
 }
+
+function render() {
+  renderSummary();
+  renderTable();
+}
diff --git a/infra/nightly-resources/web/style.css b/infra/nightly-resources/web/style.css
index 8eb1262d4..dd70666f4 100644
--- a/infra/nightly-resources/web/style.css
+++ b/infra/nightly-resources/web/style.css
@@ -12,6 +12,21 @@ body {
 .page {
   width: 100%;
   padding: 24px;
+  position: relative;
+}
+
+#time-display-selector {
+  position: absolute;
+  top: 24px;
+  right: 24px;
+  display: flex;
+  gap: 12px;
+  border: 1px solid #ccc;
+  padding: 8px 12px;
+}
+
+#time-display-selector legend {
+  padding: 0 4px;
 }
 
 h1 {

From ac3765cf2fe5e153cce1b53f6040ad50bb0f02e6 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 21:51:35 -0700
Subject: [PATCH 26/31] fix sorting

---
 infra/nightly-resources/web/index.js | 17 ++++++++++++-----
 infra/nightly-resources/web/table.js | 16 ++++++++++++++--
 2 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 9b5a80545..09f01a708 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -142,15 +142,22 @@ function renderTable() {
 
   const rows = benchmarks.map((b) => ({
     Benchmark: b.benchmark_name,
-    "Wall Time": displayTime(b.wall_time_micros),
-    Rules: displayTime(b.report.rule_micros),
-    Extraction: displayTime(b.report.extraction_micros),
-    Other: displayTime(b.report.other_micros),
+    "Wall Time": b.wall_time_micros,
+    Rules: b.report.rule_micros,
+    Extraction: b.report.extraction_micros,
+    Other: b.report.other_micros,
   }));
 
+  const displayFns = {
+    "Wall Time": displayTime,
+    Rules: displayTime,
+    Extraction: displayTime,
+    Other: displayTime,
+  };
+
   const tableDiv = document.querySelector("#active-suite-table");
   tableDiv.innerHTML = "";
-  tableDiv.appendChild(convertToTable(columns, rows));
+  tableDiv.appendChild(convertToTable(columns, rows, displayFns));
 }
 
 function render() {
diff --git a/infra/nightly-resources/web/table.js b/infra/nightly-resources/web/table.js
index e63e36b4b..9666695e0 100644
--- a/infra/nightly-resources/web/table.js
+++ b/infra/nightly-resources/web/table.js
@@ -14,10 +14,15 @@
  * Extra values are ignored. Missing values are displayed as `-` in the table
  * If empty, a table with only the header row will be returned
  *
+ * @param {Object} [displayFns]
+ * Optional map from column name -> formatter function (rawValue) => string.
+ * Sorting still uses the raw value from `rows`; rendering uses the formatted
+ * string. Columns not listed are rendered as-is.
+ *
  * @return An HTML <table> DOM element
  * Sortable by column, defaults to sorted by first column
  */
-export function convertToTable(columns, rows) {
+export function convertToTable(columns, rows, displayFns = {}) {
   const table = document.createElement("table");
 
   const STATE = {
@@ -83,7 +88,14 @@ export function convertToTable(columns, rows) {
       const tr = document.createElement("tr");
       for (const column of columns) {
         const td = document.createElement("td");
-        td.textContent = row[column] ?? "-"; // We want to leave 0 and "" intact, so don't use ||
+        const rawValue = row[column];
+        if (rawValue === undefined || rawValue === null) {
+          td.textContent = "-";
+        } else if (displayFns[column]) {
+          td.textContent = displayFns[column](rawValue);
+        } else {
+          td.textContent = rawValue;
+        }
         tr.appendChild(td);
       }
       tbody.appendChild(tr);

From c2584bef7e25e41a63f313b3acc90290760e73bf Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Tue, 12 May 2026 21:52:43 -0700
Subject: [PATCH 27/31] switch default

---
 infra/nightly-resources/web/index.html | 4 ++--
 infra/nightly-resources/web/index.js   | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/infra/nightly-resources/web/index.html b/infra/nightly-resources/web/index.html
index d4a2d12c2..d8dda7f72 100644
--- a/infra/nightly-resources/web/index.html
+++ b/infra/nightly-resources/web/index.html
@@ -11,8 +11,8 @@
       <h1>Poach Nightly</h1>
       <fieldset id="time-display-selector">
         <legend>Time Display</legend>
-        <label><input type="radio" name="time-display" value="raw" checked /> Raw</label>
-        <label><input type="radio" name="time-display" value="readable" /> Readable</label>
+        <label><input type="radio" name="time-display" value="readable" checked /> Readable</label>
+        <label><input type="radio" name="time-display" value="raw" /> Raw</label>
       </fieldset>
       <p id="status">Loading <code>data/data.json</code>...</p>
       <div id="contents">
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 09f01a708..3615bbbf4 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -2,7 +2,7 @@ import { convertToTable } from "./table.js";
 
 const STATE = {
   activeSuite: null,
-  timeDisplay: "raw",
+  timeDisplay: "readable",
 };
 
 const GLOBAL_DATA = {

From c3865b2d98482038945861042c154486c9728834 Mon Sep 17 00:00:00 2001
From: Haobin Ni <haobin.ni@gmail.com>
Date: Mon, 18 May 2026 10:20:43 -0700
Subject: [PATCH 28/31] Keep main up-to-date with PTP_Vanilla's nightly
 improvements

---
 infra/nightly.py |  8 ++++-
 src/poach.rs     | 89 +++++-------------------------------------------
 2 files changed, 15 insertions(+), 82 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index 7802fddcc..95eef4469 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -69,7 +69,7 @@ def summarize_report(report):
     else:
       other_micros += time_step["total"]
 
-  # No sizes in vanilla egglog reports
+  # Add size and other information to the report here
 
   return {
     "rule_micros": rule_micros,
@@ -88,6 +88,12 @@ def run_benchmarks(benchmark_dir):
   # For this treatment, we don't do anything at train time,
   # we just use the train benchmarks at serve time
 
+  # TODO: invoke the poach commands appropriate for this branch (e.g.
+  # `poach train ...` and/or `poach serve ...`) for each benchmark file
+  # and append one result dict per command to `results`. Each result
+  # must include at least: "suite_name", "benchmark_name", "status",
+  # "wall_time_micros" (plus any branch-specific fields like "phase").
+
   results = []
   failing_benchmarks = []
   for benchmark in benchmarks:
diff --git a/src/poach.rs b/src/poach.rs
index 5de7cc439..1ade70033 100644
--- a/src/poach.rs
+++ b/src/poach.rs
@@ -1,6 +1,4 @@
-use poach::{EGraph, report::Reporter};
-
-use std::{fs::File, io::Write, path::PathBuf, process::exit};
+use std::path::PathBuf;
 
 use clap::{Args, Parser, Subcommand};
 
@@ -113,94 +111,23 @@ pub fn poach() {
             fine_tune(arg);
         }
         Commands::Test(arg) => {
-            eprintln!("test({:?})", arg);
-            //TODO: run vanilla egglog tests
+            println!("test({:?})", arg);
         }
     }
     // TODO handle report IO
 }
 
-/// VanillaEgglog's model is just unit
-/// Still, it would create an empty file
 fn train(arg: TrainArgs) {
-    let _ = File::create(arg.output_model_file.as_path());
+    println!("train({:?})", arg);
+    //TODO
 }
 
-/// VanillaEgglog
 fn serve(arg: ServeArgs) {
-    match arg.mode {
-        ServeMode::Streaming => {
-            let mut egraph = EGraph::default();
-
-            rayon::ThreadPoolBuilder::new()
-                .num_threads(1)
-                .build_global()
-                .unwrap();
-
-            match egraph.repl(poach::RunMode::Normal) {
-                Ok(_) => {}
-                _ => {
-                    exit(-1);
-                }
-            }
-        }
-        ServeMode::Single { input_file: input } => {
-            let mut egraph = EGraph::default();
-
-            rayon::ThreadPoolBuilder::new()
-                .num_threads(1)
-                .build_global()
-                .unwrap();
-
-            let program = std::fs::read_to_string(input.as_path()).unwrap_or_else(|_| {
-                let arg = input.to_string_lossy();
-                panic!("Failed to read file {arg}")
-            });
-
-            let result: Result<_, poach::Error> = if arg.debug {
-                let filename = Some(input.to_str().unwrap().into());
-                match egraph.parser.get_program_from_string(filename, &program) {
-                    Ok(parsed) => {
-                        let mut reporter = Reporter::new();
-                        let result = egraph.run_program_with_reporter(parsed, &mut reporter);
-                        if result.is_ok() {
-                            serde_json::to_writer(&mut std::io::stderr(), &reporter.build_report())
-                                .expect("failed to serialize debug report");
-                            writeln!(std::io::stderr()).expect("failed to terminate debug report");
-                        }
-                        result
-                    }
-                    Err(err) => Err(err.into()),
-                }
-            } else {
-                egraph
-                    .parse_and_run_program(Some(input.to_str().unwrap().into()), &program)
-                    .map_err(Into::into)
-            };
-
-            match result {
-                Ok(msgs) => {
-                    for msg in msgs {
-                        print!("{msg}");
-                    }
-                }
-                _ => {
-                    exit(-1);
-                }
-            }
-        }
-        ServeMode::Batch {
-            input_dir: _,
-            output_dir: _,
-        } => {
-            //TODO
-            panic!("Batch not implemented");
-        }
-    }
+    println!("serve({:?})", arg);
+    //TODO
 }
 
-/// VanillaEgglog's model is just unit
-/// Still, it would create an empty file
 fn fine_tune(arg: FineTuneArgs) {
-    let _ = File::create(arg.output_model_file.as_path());
+    println!("fine_tune({:?})", arg);
+    //TODO
 }

From 2aa3dded9a95e6b7026ba356fd3e452bc1e34cd9 Mon Sep 17 00:00:00 2001
From: Haobin Ni <haobin.ni@gmail.com>
Date: Mon, 18 May 2026 10:51:36 -0700
Subject: [PATCH 29/31] Fixing obvious erros

---
 Cargo.lock                           | 10 ----------
 Cargo.toml                           |  1 -
 infra/nightly-resources/web/index.js |  2 +-
 infra/nightly.py                     |  4 ++--
 4 files changed, 3 insertions(+), 14 deletions(-)

diff --git a/Cargo.lock b/Cargo.lock
index 79e7c8882..b2ef14f63 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1219,7 +1219,6 @@ dependencies = [
  "rustc-hash",
  "serde",
  "serde_json",
- "serde_millis",
  "testing_logger",
  "thiserror",
  "tracing",
@@ -1496,15 +1495,6 @@ dependencies = [
  "serde_core",
 ]
 
-[[package]]
-name = "serde_millis"
-version = "0.1.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e6e2dc780ca5ee2c369d1d01d100270203c4ff923d2a4264812d723766434d00"
-dependencies = [
- "serde",
-]
-
 [[package]]
 name = "sha2"
 version = "0.10.9"
diff --git a/Cargo.toml b/Cargo.toml
index 465755148..cdcbd20ce 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -150,7 +150,6 @@ lazy_static = "1.5.0"
 serde = { workspace = true }
 tracing-subscriber = "0.3.23"
 tracing = "0.1.44"
-serde_millis = "0.1.1"
 
 [build-dependencies]
 chrono = { workspace = true, features = ["now"], optional = true }
diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index 3615bbbf4..dcf54fbe8 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -135,7 +135,7 @@ function renderTable() {
   document.querySelector("#active-suite-summary").innerHTML = `
   <div>
     <h3>${STATE.activeSuite}</h3>
-    <p>${benchmarks.length} benchmarks | ${totalTime} s</p>
+    <p>${benchmarks.length} benchmarks | ${displayTime(totalTime)} </p>
   </div>`;
 
   const columns = ["Benchmark", "Wall Time", "Rules", "Extraction", "Other"];
diff --git a/infra/nightly.py b/infra/nightly.py
index 95eef4469..420cad8e6 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -40,7 +40,7 @@ def run_command(cmd):
   time_micros = (time.perf_counter_ns() - started) // 1000
   if cmd_result.returncode != 0:
     return {
-      "cmd": cmd,
+      "cmd": " ".join(cmd),
       "status": "error",
       "wall_time_micros": time_micros
     }
@@ -124,4 +124,4 @@ def run_benchmarks(benchmark_dir):
   if len(sys.argv) != 2:
     raise SystemExit(f"Usage: nightly.py <benchmark-dir>")
   
-  main(sys.argv[1])
\ No newline at end of file
+  main(sys.argv[1])

From f1b4dbd7de170d349ff4217a0317c687d29a846e Mon Sep 17 00:00:00 2001
From: Haobin Ni <haobin.ni@gmail.com>
Date: Mon, 18 May 2026 14:24:48 -0700
Subject: [PATCH 30/31] Fix

---
 infra/nightly-resources/web/index.js | 2 +-
 infra/nightly.py                     | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/infra/nightly-resources/web/index.js b/infra/nightly-resources/web/index.js
index dcf54fbe8..0c71de62a 100644
--- a/infra/nightly-resources/web/index.js
+++ b/infra/nightly-resources/web/index.js
@@ -17,7 +17,7 @@ async function load() {
 
   const response = await fetch("./data/data.json");
   if (!response.ok) {
-    statusNode.textContent = `Failed to load data/data.json: ${error}`;
+    statusNode.textContent = `Failed to load data/data.json: ${response.status} ${response.statusText}`;
     return;
   }
 
diff --git a/infra/nightly.py b/infra/nightly.py
index 420cad8e6..a70647ad4 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -1,4 +1,4 @@
-#!/user/bin/env python3
+#!/usr/bin/env python3
 
 import json
 import subprocess

From 08b2b102b91c420248b899a666fffb0a440b9ca8 Mon Sep 17 00:00:00 2001
From: Anjali Pal <anjalip@cs.washington.edu>
Date: Mon, 18 May 2026 17:21:54 -0700
Subject: [PATCH 31/31] remove extra prints

---
 infra/nightly.py | 2 --
 1 file changed, 2 deletions(-)

diff --git a/infra/nightly.py b/infra/nightly.py
index 6efb3fcb0..30166c93c 100644
--- a/infra/nightly.py
+++ b/infra/nightly.py
@@ -154,7 +154,6 @@ def run_benchmarks(benchmark_dir):
     ]
     train_result = run_command(train_command, summarize_train_report)
     if train_result["status"] == "success":
-        print(f"Train Success: {benchmark_name}")
         report["train"] = train_result
     else:
         print(f"Failure: {benchmark_name}")
@@ -172,7 +171,6 @@ def run_benchmarks(benchmark_dir):
     
     serve_result = run_command(serve_command, summarize_serve_report)
     if serve_result["status"] == "success":
-      print(f"Serve Success: {benchmark_name}")
       report["serve"] = serve_result
       reports.append(report)
     else: