Phase 0: Table extraction accuracy harness + fixtures + CI gate

.gitattributes

@@ -0,0 +1,14 @@
+# Enforce byte-exact checkout for binary fixtures.
+# Without this, git's autodetection heuristic can misclassify uncompressed
+# PDFs (pageCompression=0) as text on Windows (core.autocrlf=true), which
+# rewrites LF to CRLF inside the PDF and corrupts xref/trailer offsets.
+*.pdf      binary
+*.wasm     binary
+*.ttf      binary
+*.otf      binary
+*.woff     binary
+*.woff2    binary
+*.ico      binary
+*.png      binary
+*.jpg      binary
+*.jpeg     binary

.github/workflows/ci.yml

@@ -102,3 +102,51 @@ jobs:
         with:
           name: test-results-${{ matrix.os }}-py${{ matrix.python-version }}
           path: results.xml
+
+  accuracy:
+    name: Table Extraction Accuracy
+    needs: lint
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      - name: Set up Python 3.13
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.13"
+
+      - name: Install Rust toolchain
+        uses: dtolnay/rust-toolchain@stable
+
+      - uses: Swatinem/rust-cache@v2.9.1
+        with:
+          save-if: false
+
+      - uses: astral-sh/setup-uv@v7
+        with:
+          cache-dependency-glob: pyproject.toml
+
+      - name: Install dependencies
+        run: uv sync --extra dev
+
+      - name: Build native extension
+        uses: PyO3/maturin-action@v1
+        with:
+          command: develop
+          args: --release
+
+      - name: Verify fixtures are up to date
+        run: |
+          cp -r tests/fixtures/tables /tmp/tables_snapshot
+          uv run python scripts/generate_table_fixtures.py
+          diff -r /tmp/tables_snapshot tests/fixtures/tables
+
+      - name: Run accuracy harness
+        run: uv run pytest tests/python/ -m accuracy -v --tb=short
+
+      - name: Upload accuracy report
+        if: always()
+        uses: actions/upload-artifact@v7
+        with:
+          name: accuracy-report
+          path: tests/output/table_accuracy.json

.gitignore

@@ -11,6 +11,7 @@ dist/
 build/
 *.so
 *.dylib
+CLAUDE.md
 *.dll
 .venv/
 venv/
@@ -36,6 +37,9 @@ python/paperjam/libpdfium.so
 # Test fixtures (generated)
 tests/fixtures/large_*.pdf
 
+# Per-session test artifacts (accuracy reports, etc.)
+tests/output/
+
 # Sphinx
 _build
 

.pre-commit-config.yaml

@@ -47,3 +47,11 @@ repos:
         language: system
         files: ^docs-site/src/
         pass_filenames: false
+
+  - repo: local
+    hooks:
+      - id: strip-ai-attribution
+        name: strip AI attribution from commit messages
+        entry: .claude/hooks/strip-ai-attribution.sh
+        language: script
+        stages: [commit-msg]

crates/paperjam-core/Cargo.toml

@@ -45,3 +45,10 @@ render = ["dep:pdfium-render", "dep:image"]
 signatures = ["dep:x509-parser", "dep:cms", "dep:der", "dep:sha1", "dep:rsa", "dep:p256", "dep:pkcs8", "dep:spki"]
 ltv = ["signatures", "dep:ureq", "dep:rustls"]
 validation = ["dep:roxmltree"]
+
+[dev-dependencies]
+criterion = { version = "0.5", features = ["html_reports"] }
+
+[[bench]]
+name = "table_extraction"
+harness = false

crates/paperjam-core/benches/table_extraction.rs

@@ -0,0 +1,79 @@
+//! Criterion microbench for table extraction over the fixture corpus.
+//!
+//! Run with: `cargo bench -p paperjam-core --bench table_extraction`
+//!
+//! Each synthetic PDF fixture under `tests/fixtures/tables/` becomes a bench group.
+//! The document and page are parsed outside the measured section so the bench only
+//! measures `table::extract_tables` itself — that's the code subsequent phases will
+//! change.
+
+use std::path::PathBuf;
+
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion, Throughput};
+use paperjam_core::document::Document;
+use paperjam_core::table::{extract_tables, TableExtractionOptions};
+
+fn fixtures_dir() -> PathBuf {
+    let manifest = PathBuf::from(env!("CARGO_MANIFEST_DIR"));
+    manifest.join("../../tests/fixtures/tables")
+}
+
+fn collect_fixtures() -> Vec<PathBuf> {
+    let mut out = Vec::new();
+    let dir = fixtures_dir();
+    let Ok(rd) = std::fs::read_dir(&dir) else {
+        eprintln!(
+            "skipping bench: fixtures dir not found at {}",
+            dir.display()
+        );
+        return out;
+    };
+    for entry in rd.flatten() {
+        let path = entry.path();
+        if path.extension().and_then(|s| s.to_str()) == Some("pdf") {
+            out.push(path);
+        }
+    }
+    out.sort();
+    out
+}
+
+fn bench_extract(c: &mut Criterion) {
+    let fixtures = collect_fixtures();
+    let opts = TableExtractionOptions::default();
+
+    let mut group = c.benchmark_group("table_extraction");
+    for path in &fixtures {
+        let name = path
+            .file_stem()
+            .and_then(|s| s.to_str())
+            .unwrap_or("?")
+            .to_string();
+        // Parse the document once and keep its pages alive for the whole bench run.
+        let doc = match Document::open(path) {
+            Ok(d) => d,
+            Err(e) => {
+                eprintln!("skipping {name}: failed to open ({e:?})");
+                continue;
+            }
+        };
+        let mut pages = Vec::new();
+        for n in 1..=doc.page_count() as u32 {
+            if let Ok(p) = doc.page(n) {
+                pages.push(p);
+            }
+        }
+        group.throughput(Throughput::Elements(pages.len() as u64));
+        group.bench_with_input(BenchmarkId::from_parameter(&name), &pages, |b, pages| {
+            b.iter(|| {
+                for page in pages.iter() {
+                    let _ = extract_tables(page, &opts);
+                }
+            });
+        });
+    }
+    group.finish();
+}
+
+criterion_group!(benches, bench_extract);
+criterion_main!(benches);

crates/paperjam-core/src/forms/mod.rs

@@ -93,22 +93,14 @@ pub fn fill_form_fields(
 
     for (name, value) in values {
         match field_name_map.get(name) {
-            None => {
-                not_found.push(name.clone());
+            Some((false, field_type)) if set_field_value(&mut inner, name, value, field_type)? => {
+                filled += 1;
+                filled_fields.push((name.clone(), value.clone(), field_type.clone()));
             }
-            Some((true, _)) => {
-                // Read-only field, skip
+            // Not in the map, read-only, or set_field_value returned false: record as not-found.
+            _ => {
                 not_found.push(name.clone());
             }
-            Some((false, field_type)) => {
-                // Find and update the field object
-                if set_field_value(&mut inner, name, value, field_type)? {
-                    filled += 1;
-                    filled_fields.push((name.clone(), value.clone(), field_type.clone()));
-                } else {
-                    not_found.push(name.clone());
-                }
-            }
         }
     }
 

crates/paperjam-core/src/manipulation/insert.rs

@@ -40,7 +40,7 @@ pub fn insert_blank_pages(doc: &Document, positions: &[(u32, f64, f64)]) -> Resu
 
     // Sort positions in reverse order so insertions don't shift indices
     let mut sorted_positions = positions.to_vec();
-    sorted_positions.sort_by(|a, b| b.0.cmp(&a.0));
+    sorted_positions.sort_by_key(|p| std::cmp::Reverse(p.0));
 
     for (after_page, width, height) in sorted_positions {
         // Create an empty content stream

crates/paperjam-epub/src/parser.rs

@@ -218,22 +218,20 @@ fn parse_opf(xml: &str) -> Result<(OpfMetadata, HashMap<String, String>, Vec<Str
                     }
                 }
             }
-            Ok(Event::Text(ref e)) => {
-                if section == Section::Metadata {
-                    let text = e.unescape().unwrap_or_default().trim().to_string();
-                    if !text.is_empty() {
-                        match current_tag.as_str() {
-                            "title" => metadata.title = Some(text),
-                            "creator" => metadata.creator = Some(text),
-                            "subject" => metadata.subject = Some(text),
-                            "description" => metadata.description = Some(text),
-                            "publisher" => metadata.publisher = Some(text),
-                            "date" => metadata.date = Some(text),
-                            "language" => metadata.language = Some(text),
-                            "identifier" => metadata.identifier = Some(text),
-                            "rights" => metadata.rights = Some(text),
-                            _ => {}
-                        }
+            Ok(Event::Text(ref e)) if section == Section::Metadata => {
+                let text = e.unescape().unwrap_or_default().trim().to_string();
+                if !text.is_empty() {
+                    match current_tag.as_str() {
+                        "title" => metadata.title = Some(text),
+                        "creator" => metadata.creator = Some(text),
+                        "subject" => metadata.subject = Some(text),
+                        "description" => metadata.description = Some(text),
+                        "publisher" => metadata.publisher = Some(text),
+                        "date" => metadata.date = Some(text),
+                        "language" => metadata.language = Some(text),
+                        "identifier" => metadata.identifier = Some(text),
+                        "rights" => metadata.rights = Some(text),
+                        _ => {}
                     }
                 }
             }

crates/paperjam-epub/src/toc.rs

@@ -76,10 +76,8 @@ fn parse_nav_point(reader: &mut Reader<&[u8]>) -> Option<TocEntry> {
                     }
                 }
             }
-            Ok(Event::Text(ref e)) => {
-                if in_text {
-                    title = e.unescape().unwrap_or_default().trim().to_string();
-                }
+            Ok(Event::Text(ref e)) if in_text => {
+                title = e.unescape().unwrap_or_default().trim().to_string();
             }
             Ok(Event::End(ref e)) => {
                 let local = local_name(e.name().as_ref());

crates/paperjam-pptx/src/parser.rs

@@ -254,11 +254,9 @@ fn parse_slide(xml: &str, index: usize, notes_xml: Option<&str>) -> Result<Slide
                     _ => {}
                 }
             }
-            Ok(Event::Empty(ref e)) => {
-                if parser.in_shape {
-                    let local = local_name_owned(e.name().as_ref());
-                    handle_shape_empty(&mut parser, e, &local)?;
-                }
+            Ok(Event::Empty(ref e)) if parser.in_shape => {
+                let local = local_name_owned(e.name().as_ref());
+                handle_shape_empty(&mut parser, e, &local)?;
             }
             Ok(Event::Text(ref e)) => {
                 if parser.in_table_cell {
@@ -436,10 +434,8 @@ fn parse_notes_text(xml: &str) -> Result<String> {
                     text.push('\n');
                 }
             }
-            Ok(Event::Text(ref e)) => {
-                if in_text_body {
-                    text.push_str(&e.unescape().unwrap_or_default());
-                }
+            Ok(Event::Text(ref e)) if in_text_body => {
+                text.push_str(&e.unescape().unwrap_or_default());
             }
             Ok(Event::Eof) => break,
             Err(e) => return Err(PptxError::Xml(format!("notes XML error: {e}"))),

docs-site/src/pages/index.tsx

@@ -137,15 +137,6 @@ export default function Home(): React.JSX.Element {
             borderRadius: '16px',
           }}
         />
-        <h1
-          style={{
-            fontSize: '2.5rem',
-            fontWeight: 800,
-            marginBottom: '0.5rem',
-          }}
-        >
-          paperjam
-        </h1>
         <p
           style={{
             fontSize: '1.25rem',

pyproject.toml

@@ -42,6 +42,8 @@ dev = [
     "pdfplumber>=0.10",
     "mypy>=1.8",
     "ruff>=0.3",
+    "reportlab>=4.0,<4.2",
+    "pandas>=2.0",
 ]
 
 [tool.maturin]
@@ -56,6 +58,9 @@ include = [
 
 [tool.pytest.ini_options]
 testpaths = ["tests/python"]
+markers = [
+    "accuracy: table extraction accuracy harness (run with -m accuracy)",
+]
 
 [tool.ruff]
 target-version = "py312"