diff --git a/docs/errors-array-format.md b/docs/errors-array-format.md
new file mode 100644
index 0000000..d47452c
--- /dev/null
+++ b/docs/errors-array-format.md
@@ -0,0 +1,482 @@
+# Errors Array Format and Test Integration Guide
+
+This document explains the complete structure of the errors/diagnostics array in pdftract extraction results and how to integrate assertions in tests.
+
+## Overview
+
+The pdftract extraction system uses a unified diagnostic system to report errors, warnings, and informational messages. All diagnostics are emitted during PDF parsing and extraction without panicking (per INV-8), allowing the parser to always attempt recovery and continue processing.
+
+## Extraction Result Structure
+
+### Main Result Structure
+
+```rust
+pub struct ExtractionResult {
+    pub fingerprint: String,
+    pub pages: Vec<PageResult>,
+    pub metadata: ExtractionMetadata,
+    pub signatures: Vec<SignatureJson>,
+    pub form_fields: Vec<FormFieldJson>,
+    pub links: Vec<LinkJson>,
+    pub attachments: Vec<AttachmentJson>,
+    pub threads: Vec<ThreadJson>,
+    pub javascript_actions: Vec<JavascriptActionJson>,
+}
+```
+
+### Metadata Structure (Contains Errors/Diagnostics)
+
+```rust
+pub struct ExtractionMetadata {
+    pub page_count: usize,
+    pub receipts_mode: ReceiptsMode,
+    pub span_count: usize,
+    pub block_count: usize,
+    pub cache_status: Option<String>,
+    pub cache_age_seconds: Option<u64>,
+    pub error_count: usize,              // Number of pages that failed to extract
+    pub reading_order_algorithm: Option<String>,
+    pub diagnostics: Vec<String>,        // THE ERRORS ARRAY
+    pub profile_name: Option<String>,
+    pub profile_version: Option<String>,
+    pub profile_fields: Option<serde_json::Value>,
+}
+```
+
+## Errors Array Format
+
+### Location
+- **Field**: `ExtractionResult.metadata.diagnostics`
+- **Type**: `Vec<String>`
+- **Description**: Array of diagnostic message strings emitted during extraction
+
+### String Format
+
+Each diagnostic in the array follows this format:
+
+```
+{DIAGNOSTIC_CODE}: {Human-readable message} (byte offset {offset})?
+```
+
+**Examples:**
+- `STREAM_DECODE_ERROR: zlib stream truncated mid-inflation (byte offset 12345)`
+- `STRUCT_INVALID_NAME: Name object exceeds 127-byte limit (byte offset 67890)`
+- `FONT_GLYPH_UNMAPPED: Glyph could not be mapped to Unicode`
+- `XREF_REPAIRED: Xref was reconstructed via forward scan`
+
+### Components
+
+1. **Diagnostic Code** (e.g., `STREAM_DECODE_ERROR`):
+   - Identifies the type of error/warning/info
+   - Follows naming convention: `CATEGORY_SPECIFIC_ISSUE`
+   - Categories: `STRUCT_*`, `STREAM_*`, `XREF_*`, `ENCRYPTION_*`, `PAGE_*`, `FONT_*`, `OCR_*`, `REMOTE_*`, `GSTATE_*`, `LAYOUT_*`, `MCP_*`, `CACHE_*`, etc.
+
+2. **Message**: Human-readable description of what occurred
+
+3. **Byte Offset** (optional): Location in the PDF file where the error occurred
+
+## Accessing Error Information
+
+### From Extraction Result
+
+```rust
+use pdftract_core::extract::extract_pdf;
+
+let result = extract_pdf("test.pdf", &Default::default())?;
+
+// Access the errors array
+let diagnostics = &result.metadata.diagnostics;
+
+// Get error count
+let error_count = result.metadata.error_count;
+
+// Iterate through all diagnostics
+for diagnostic in diagnostics {
+    println!("{}", diagnostic);
+}
+```
+
+### Checking for Specific Errors
+
+```rust
+// Check if any stream decode errors occurred
+let has_stream_errors = diagnostics.iter()
+    .any(|d| d.contains("STREAM_DECODE_ERROR"));
+
+// Check for encryption errors
+let has_encryption_errors = diagnostics.iter()
+    .any(|d| d.contains("ENCRYPTION"));
+
+// Count specific error types
+let glyph_unmapped_count = diagnostics.iter()
+    .filter(|d| d.contains("FONT_GLYPH_UNMAPPED"))
+    .count();
+```
+
+## Diagnostic Categories and Codes
+
+### Structure Errors (STRUCT_*)
+- `STRUCT_INVALID_NAME` - Invalid name character or malformed name object
+- `STRUCT_INVALID_HEX` - Invalid hex character in hex string
+- `STRUCT_MISSING_KEY` - Missing required dictionary key
+- `STRUCT_CIRCULAR_REF` - Circular reference detected
+- `STRUCT_UNEXPECTED_BYTE` - Unexpected byte during parsing
+- `STRUCT_UNEXPECTED_EOF` - Unexpected end of file
+
+### Stream Errors (STREAM_*)
+- `STREAM_DECODE_ERROR` - Stream decompression failed (corrupt data)
+- `STREAM_BOMB` - Decompression bomb limit exceeded
+- `STREAM_UNKNOWN_FILTER` - Unknown filter name
+- `STREAM_INVALID_PARAMS` - Invalid filter parameters
+- `STREAM_TRUNCATED` - Stream data truncated
+
+### XRef Errors (XREF_*)
+- `XREF_INVALID_HEADER` - Invalid xref keyword or header
+- `XREF_REPAIRED` - Xref was reconstructed via forward scan
+- `XREF_TRUNCATED` - Truncated xref table
+
+### Encryption Errors (ENCRYPTION_*)
+- `ENCRYPTION_UNSUPPORTED` - Unsupported encryption or no password supplied
+- `ENCRYPTION_WRONG_PASSWORD` - Password incorrect
+- `ENCRYPTION_INVALID_DICT` - Invalid encryption dictionary
+
+### Font Errors (FONT_*)
+- `FONT_GLYPH_UNMAPPED` - Glyph could not be mapped to Unicode
+- `FONT_NOT_FOUND` - Font not found or couldn't be parsed
+- `FONT_INVALID_CMAP` - Invalid CMap format
+- `FONT_PARSE_FAILED` - Font program parsing failed
+
+### Page Errors (PAGE_*)
+- `PAGE_OUT_OF_RANGE` - Page number out of range
+- `PAGE_INVALID_COUNT` - Invalid /Count key in /Pages tree
+- `PAGE_INVALID_ROTATE` - Invalid /Rotate value (not multiple of 90)
+
+### Graphics State Errors (GSTATE_*)
+- `GSTATE_STACK_OVERFLOW` - Graphics state stack overflow
+- `GSTATE_STACK_UNDERFLOW` - Graphics state stack underflow
+- `CM_ARG_COUNT` - Invalid argument count for cm operator
+
+### OCR Errors (OCR_*)
+- `OCR_JBIG2_UNSUPPORTED` - JBIG2 decoder not available
+- `OCR_TESSERACT_FAILED` - Tesseract OCR failed
+- `OCR_LANGUAGE_UNAVAILABLE` - Requested language pack not available
+
+## Assertion Patterns for Tests
+
+### Pattern 1: Check for No Errors (Clean Extraction)
+
+```rust
+#[test]
+fn test_clean_pdf_extracts_without_errors() {
+    let result = extract_pdf("tests/fixtures/clean.pdf", &Default::default())
+        .expect("Extraction should succeed");
+    
+    // Verify no diagnostics were emitted
+    assert!(
+        result.metadata.diagnostics.is_empty(),
+        "Expected no diagnostics, but got: {:?}",
+        result.metadata.diagnostics
+    );
+    
+    // Verify error count is zero
+    assert_eq!(result.metadata.error_count, 0);
+}
+```
+
+### Pattern 2: Check for Specific Error Code
+
+```rust
+#[test]
+fn test_truncated_stream_emits_decode_error() {
+    let result = extract_pdf("tests/fixtures/truncated.pdf", &Default::default())
+        .expect("Extraction should succeed (with recovery)");
+    
+    // Check for STREAM_DECODE_ERROR diagnostic
+    let has_decode_error = result.metadata.diagnostics.iter()
+        .any(|d| d.contains("STREAM_DECODE_ERROR"));
+    
+    assert!(
+        has_decode_error,
+        "Expected STREAM_DECODE_ERROR diagnostic, got: {:?}",
+        result.metadata.diagnostics
+    );
+}
+```
+
+### Pattern 3: Check for Multiple Error Types
+
+```rust
+#[test]
+fn test_malformed_pdf_emits_multiple_warnings() {
+    let result = extract_pdf("tests/fixtures/malformed.pdf", &Default::default())
+        .expect("Extraction should succeed (with recovery)");
+    
+    // Check for multiple expected diagnostics
+    let diagnostics = &result.metadata.diagnostics;
+    
+    assert!(
+        diagnostics.iter().any(|d| d.contains("STRUCT_INVALID_NAME")),
+        "Expected STRUCT_INVALID_NAME"
+    );
+    
+    assert!(
+        diagnostics.iter().any(|d| d.contains("STREAM_DECODE_ERROR")),
+        "Expected STREAM_DECODE_ERROR"
+    );
+    
+    assert!(
+        diagnostics.iter().any(|d| d.contains("XREF_REPAIRED")),
+        "Expected XREF_REPAIRED"
+    );
+}
+```
+
+### Pattern 4: Count Specific Errors
+
+```rust
+#[test]
+fn test_pdf_with_unmapped_glyphs() {
+    let result = extract_pdf("tests/fixtures/custom-font.pdf", &Default::default())
+        .expect("Extraction should succeed");
+    
+    // Count unmapped glyph diagnostics
+    let unmapped_count = result.metadata.diagnostics.iter()
+        .filter(|d| d.contains("FONT_GLYPH_UNMAPPED"))
+        .count();
+    
+    assert!(
+        unmapped_count > 0,
+        "Expected at least one FONT_GLYPH_UNMAPPED diagnostic"
+    );
+    
+    println!("Found {} unmapped glyphs", unmapped_count);
+}
+```
+
+### Pattern 5: Verify Error Message Content
+
+```rust
+#[test]
+fn test_encryption_error_message() {
+    let result = extract_pdf_encrypted("tests/fixtures/encrypted.pdf", None)
+        .expect("Should produce error result");
+    
+    // Find the encryption error diagnostic
+    let encryption_diag = result.metadata.diagnostics.iter()
+        .find(|d| d.contains("ENCRYPTION_UNSUPPORTED"))
+        .expect("Expected ENCRYPTION_UNSUPPORTED diagnostic");
+    
+    // Verify message contains expected text
+    assert!(
+        encryption_diag.contains("no password supplied") || 
+        encryption_diag.contains("Unsupported encryption"),
+        "Encryption error message should mention password or unsupported algorithm"
+    );
+}
+```
+
+### Pattern 6: Check Error Count Matches
+
+```rust
+#[test]
+fn test_partial_extraction_error_count() {
+    let result = extract_pdf("tests/fixtures/partial-corrupt.pdf", &Default::default())
+        .expect("Extraction should succeed");
+    
+    // Verify error_count field matches actual diagnostics
+    let actual_error_count = result.metadata.diagnostics.iter()
+        .filter(|d| d.contains("ERROR") || d.contains("FATAL"))
+        .count();
+    
+    assert_eq!(
+        result.metadata.error_count as usize,
+        actual_error_count,
+        "error_count field should match number of ERROR/FATAL diagnostics"
+    );
+}
+```
+
+### Pattern 7: Assert No Fatal Errors
+
+```rust
+#[test]
+fn test_extraction_has_no_fatal_errors() {
+    let result = extract_pdf("tests/fixtures/complex.pdf", &Default::default())
+        .expect("Extraction should succeed");
+    
+    // Ensure no fatal-level diagnostics were emitted
+    let has_fatal = result.metadata.diagnostics.iter()
+        .any(|d| {
+            // Fatal codes include ENCRYPTION_UNSUPPORTED, REMOTE_TLS_FAILED, etc.
+            d.contains("ENCRYPTION_UNSUPPORTED") ||
+            d.contains("REMOTE_TLS_FAILED") ||
+            d.contains("REMOTE_DNS_FAILED")
+        });
+    
+    assert!(
+        !has_fatal,
+        "Extraction should not have fatal errors: {:?}",
+        result.metadata.diagnostics
+    );
+}
+```
+
+### Pattern 8: Verify Expected Number of Diagnostics
+
+```rust
+#[test]
+fn test_specific_diagnostic_count() {
+    let result = extract_pdf("tests/fixtures/known-issues.pdf", &Default::default())
+        .expect("Extraction should succeed");
+    
+    // If we know this PDF produces exactly 3 warnings
+    assert_eq!(
+        result.metadata.diagnostics.len(),
+        3,
+        "Expected exactly 3 diagnostics, got {}: {:?}",
+        result.metadata.diagnostics.len(),
+        result.metadata.diagnostics
+    );
+}
+```
+
+## Test Integration Examples
+
+### Example 1: Truncated Flate Stream Test
+
+```rust
+#[test]
+fn test_truncated_flate_stream_recovery() {
+    let fixture_path = "tests/fixtures/malformed/truncated-flate.pdf";
+    let result = extract_pdf(fixture_path, &Default::default())
+        .expect("Extraction should succeed with recovery");
+    
+    // Should have emitted STREAM_DECODE_ERROR
+    let has_stream_error = result.metadata.diagnostics.iter()
+        .any(|d| d.contains("STREAM_DECODE_ERROR") || d.contains("STREAM_TRUNCATED"));
+    
+    assert!(has_stream_error, 
+        "Expected stream decode error for truncated-flate.pdf");
+    
+    // Should still produce some output (recovery)
+    assert!(!result.pages.is_empty(), 
+        "Should produce at least one page despite stream error");
+    
+    // Verify the error message mentions truncation
+    let error_msg = result.metadata.diagnostics.iter()
+        .find(|d| d.contains("STREAM"))
+        .expect("Should have stream-related diagnostic");
+    
+    assert!(error_msg.to_lowercase().contains("truncat") ||
+            error_msg.to_lowercase().contains("incomplete"),
+        "Stream error message should mention truncation or incomplete data");
+}
+```
+
+### Example 2: Encryption Error Test
+
+```rust
+#[test]
+fn test_encrypted_pdf_without_password() {
+    let fixture_path = "tests/fixtures/encrypted/livecycle.pdf";
+    let result = extract_pdf_encrypted(fixture_path, None);
+    
+    // Should fail with encryption error
+    assert!(result.is_err(), "Should fail when encrypted PDF lacks password");
+    
+    // Check the error contains ENCRYPTION_UNSUPPORTED
+    if let Err(e) = result {
+        let error_str = e.to_string();
+        assert!(error_str.contains("ENCRYPTION_UNSUPPORTED") ||
+                error_str.contains("encryption"),
+                "Error should mention unsupported encryption: {}", error_str);
+    }
+}
+```
+
+### Example 3: Font Glyph Unmapped Test
+
+```rust
+#[test]
+fn test_custom_font_with_unmapped_glyphs() {
+    let fixture_path = "tests/fixtures/fonts/custom-subset.pdf";
+    let result = extract_pdf(fixture_path, &Default::default())
+        .expect("Extraction should succeed");
+    
+    // Should have FONT_GLYPH_UNMAPPED diagnostics
+    let unmapped_diagnostics: Vec<_> = result.metadata.diagnostics.iter()
+        .filter(|d| d.contains("FONT_GLYPH_UNMAPPED"))
+        .collect();
+    
+    assert!(!unmapped_diagnostics.is_empty(),
+        "Custom font PDF should have unmapped glyph diagnostics");
+    
+    // Each unmapped glyph produces U+FFFD in output
+    let text = result.pages.iter()
+        .flat_map(|p| p.blocks.iter().flat_map(|b| b.spans.iter().map(|s| &s.text)))
+        .collect::<String>();
+    
+    let fffd_count = text.matches('�').count();
+    assert!(fffd_count > 0,
+        "Output should contain replacement character for unmapped glyphs");
+}
+```
+
+## Best Practices for Test Assertions
+
+1. **Be Specific**: Check for exact diagnostic codes when testing specific error conditions
+2. **Allow Recovery**: Most tests should expect extraction to succeed despite errors (per INV-8 recovery principle)
+3. **Verify Output**: When errors occur, verify that meaningful output was still produced
+4. **Count Errors**: Use counts to verify expected number of diagnostics
+5. **Check Messages**: Verify diagnostic messages contain expected information
+6. **Test Severity**: Ensure fatal errors only occur for truly unrecoverable conditions
+7. **Use Helper Functions**: Create reusable assertion helpers for common patterns
+
+## Helper Functions for Common Assertions
+
+```rust
+/// Assert that extraction produces a specific diagnostic code
+pub fn assert_has_diagnostic(result: &ExtractionResult, code: &str) {
+    assert!(
+        result.metadata.diagnostics.iter().any(|d| d.contains(code)),
+        "Expected diagnostic containing '{}', got: {:?}",
+        code,
+        result.metadata.diagnostics
+    );
+}
+
+/// Assert that extraction has no diagnostics
+pub fn assert_no_diagnostics(result: &ExtractionResult) {
+    assert!(
+        result.metadata.diagnostics.is_empty(),
+        "Expected no diagnostics, got: {:?}",
+        result.metadata.diagnostics
+    );
+}
+
+/// Count diagnostics containing a specific code
+pub fn count_diagnostics(result: &ExtractionResult, code: &str) -> usize {
+    result.metadata.diagnostics.iter()
+        .filter(|d| d.contains(code))
+        .count()
+}
+
+/// Get all diagnostics containing a specific code
+pub fn get_diagnostics(result: &ExtractionResult, code: &str) -> Vec<&String> {
+    result.metadata.diagnostics.iter()
+        .filter(|d| d.contains(code))
+        .collect()
+}
+```
+
+## Summary
+
+The errors array (`ExtractionResult.metadata.diagnostics`) provides comprehensive visibility into the PDF extraction process:
+
+- **Format**: `Vec<String>` with each string formatted as `CODE: message (offset)?`
+- **Access**: `result.metadata.diagnostics`
+- **Error Count**: `result.metadata.error_count`
+- **Assertion Patterns**: Check for specific codes, count errors, verify messages, ensure recovery
+- **Categories**: STRUCT, STREAM, XREF, ENCRYPTION, PAGE, FONT, OCR, REMOTE, GSTATE, LAYOUT, MCP, CACHE, etc.
+
+This unified diagnostic system allows tests to verify that errors are properly detected, reported, and recovered from during PDF extraction.