diff --git a/.claude/codebase/types.md b/.claude/codebase/types.md
index 1ac09653..430f8609 100644
--- a/.claude/codebase/types.md
+++ b/.claude/codebase/types.md
@@ -120,12 +120,13 @@ Col* (e.g., ColID = "id", ColName = "name", ColDeletedAt = "deleted_at")
 - LLM (provider, model, baseURL, apiKey, timeout, thinking, extraContext)
   - Chat/Extraction overrides (LLMChatOverride, LLMExtractionOverride)
 - Documents (MaxFileSize ByteSize, CacheTTL Duration)
-- Extraction (MaxPages int, Enabled *bool, TextTimeout, LLMTimeout)
+- Extraction (MaxPages int, Enable *bool, LLMTimeout)
+  - OCR (Enable *bool, ConfidenceThreshold int)
 - Locale (Currency string)
 
 ### Defaults
 - Provider: "ollama", Model: "qwen3", BaseURL: "http://localhost:11434"
-- MaxPages: 20, CacheTTL: 30 days, TextTimeout: 30s, LLMTimeout: 5m
+- MaxPages: 20, CacheTTL: 30 days, LLMTimeout: 5m
 
 ## LLM Types (internal/llm/)
 
diff --git a/cmd/micasa/main.go b/cmd/micasa/main.go
index be6906ff..348a54b1 100644
--- a/cmd/micasa/main.go
+++ b/cmd/micasa/main.go
@@ -161,7 +161,9 @@ func (cmd *runCmd) Run() error {
 	exCfg := cfg.LLM.ExtractionConfig()
 	extractors := extract.DefaultExtractors(
 		cfg.Extraction.MaxPages,
-		cfg.Extraction.TextTimeoutDuration(),
+		0, // pdftotext uses its own internal default timeout (30s)
+		cfg.Extraction.IsOCREnabled(),
+		cfg.Extraction.OCR.ConfidenceThreshold,
 	)
 	opts.SetExtraction(
 		exCfg.Provider,
diff --git a/docs/content/docs/reference/configuration.md b/docs/content/docs/reference/configuration.md
index 56533326..29e21546 100644
--- a/docs/content/docs/reference/configuration.md
+++ b/docs/content/docs/reference/configuration.md
@@ -111,18 +111,19 @@ You can always infer the env var name from the config key.
 | `MICASA_LLM_MODEL` | `qwen3` | `llm.model` | LLM model name |
 | `MICASA_LLM_API_KEY` | (empty) | `llm.api_key` | LLM API key for cloud providers |
 | `MICASA_LLM_EXTRA_CONTEXT` | (empty) | `llm.extra_context` | Custom context appended to LLM system prompts |
-| `MICASA_LLM_TIMEOUT` | `5s` | `llm.timeout` | LLM operation timeout |
+| `MICASA_LLM_TIMEOUT` | `5m` | `llm.timeout` | Max time for a single LLM response |
 | `MICASA_LLM_THINKING` | (unset) | `llm.thinking` | Enable model thinking for chat |
 | `MICASA_DOCUMENTS_MAX_FILE_SIZE` | `50 MiB` | `documents.max_file_size` | Max document import size |
 | `MICASA_DOCUMENTS_CACHE_TTL` | `30d` | `documents.cache_ttl` | Document cache lifetime |
 | `MICASA_DOCUMENTS_CACHE_TTL_DAYS` | -- | `documents.cache_ttl_days` | Deprecated; use `MICASA_DOCUMENTS_CACHE_TTL` |
 | `MICASA_DOCUMENTS_FILE_PICKER_DIR` | (Downloads) | `documents.file_picker_dir` | Starting directory for the file picker |
 | `MICASA_EXTRACTION_MODEL` | (chat model) | `extraction.model` | LLM model for document extraction |
-| `MICASA_EXTRACTION_ENABLED` | `true` | `extraction.enabled` | Enable/disable LLM extraction |
+| `MICASA_EXTRACTION_ENABLE` | `true` | `extraction.enable` | Enable/disable LLM extraction |
 | `MICASA_EXTRACTION_THINKING` | `false` | `extraction.thinking` | Enable model thinking for extraction |
-| `MICASA_EXTRACTION_TEXT_TIMEOUT` | `30s` | `extraction.text_timeout` | pdftotext timeout |
 | `MICASA_EXTRACTION_MAX_PAGES` | `0` | `extraction.max_pages` | Max pages to OCR per document (0 = no limit) |
 | `MICASA_EXTRACTION_LLM_TIMEOUT` | `5m` | `extraction.llm_timeout` | LLM extraction timeout |
+| `MICASA_EXTRACTION_OCR_ENABLE` | `true` | `extraction.ocr.enable` | Enable/disable OCR on documents |
+| `MICASA_EXTRACTION_OCR_CONFIDENCE_THRESHOLD` | `0` | `extraction.ocr.confidence_threshold` | Min tesseract confidence (0-100) |
 | `MICASA_LOCALE_CURRENCY` | (auto-detect) | `locale.currency` | ISO 4217 currency code (e.g. `USD`, `EUR`, `GBP`) |
 
 {{% details title="Deprecated env var names" closed="true" %}}
@@ -139,8 +140,8 @@ warning. They will be removed in a future release.
 | `MICASA_CURRENCY` | `MICASA_LOCALE_CURRENCY` |
 | `MICASA_EXTRACTION_MAX_EXTRACT_PAGES` | `MICASA_EXTRACTION_MAX_PAGES` |
 | `MICASA_MAX_EXTRACT_PAGES` | `MICASA_EXTRACTION_MAX_PAGES` |
-| `MICASA_TEXT_TIMEOUT` | `MICASA_EXTRACTION_TEXT_TIMEOUT` |
 | `MICASA_MAX_OCR_PAGES` | `MICASA_EXTRACTION_MAX_PAGES` |
+| `MICASA_EXTRACTION_ENABLED` | `MICASA_EXTRACTION_ENABLE` |
 | `MICASA_EXTRACTION_MODEL` | `MICASA_LLM_EXTRACTION_MODEL` |
 | `MICASA_EXTRACTION_THINKING` | `MICASA_LLM_EXTRACTION_THINKING` |
 
@@ -177,12 +178,12 @@ micasa   # uses llama3.3 instead of the default qwen3
 
 ### `MICASA_LLM_TIMEOUT`
 
-Sets the LLM timeout for quick operations (ping, model listing), overriding
-the config file value. Uses Go duration syntax:
+Sets the maximum time for a single LLM response (including streaming),
+overriding the config file value. Uses Go duration syntax:
 
 ```sh
-export MICASA_LLM_TIMEOUT=15s
-micasa   # waits up to 15s for LLM server responses
+export MICASA_LLM_TIMEOUT=10m
+micasa   # waits up to 10m for LLM responses
 ```
 
 ### `MICASA_DOCUMENTS_MAX_FILE_SIZE`
@@ -278,10 +279,10 @@ model = "qwen3"
 # Use this to inject domain-specific details about your house, region, etc.
 # extra_context = "My house is a 1920s craftsman in Portland, OR."
 
-# Timeout for quick LLM server operations (ping, model listing).
-# Go duration syntax: "5s", "10s", "500ms", etc. Default: "5s".
-# Increase if your LLM server is slow to respond.
-# timeout = "5s"
+# Max time for a single LLM response (including streaming).
+# Go duration syntax: "5m", "10m", etc. Default: "5m".
+# Increase for slow models or complex queries.
+# timeout = "5m"
 
 # Enable model thinking mode for chat (e.g. qwen3 <think> blocks).
 # Unset = don't send (server default), true = enable, false = disable.
@@ -302,10 +303,6 @@ model = "qwen3"
 # with small, fast models optimized for structured JSON output.
 # model = "qwen2.5:7b"
 
-# Timeout for pdftotext. Go duration syntax: "30s", "1m", etc. Default: "30s".
-# Increase if you routinely process very large PDFs.
-# text_timeout = "30s"
-
 # Maximum pages to OCR for scanned documents. 0 = no limit. Default: 0.
 # max_pages = 0
 
@@ -338,7 +335,7 @@ set in `[llm.chat]` and `[llm.extraction]`.
 | `model` | string | `qwen3` | Model identifier sent in chat requests. Must be available on the server. |
 | `api_key` | string | (empty) | Authentication credential. Required for cloud providers (Anthropic, OpenAI, etc.). Leave empty for local servers. |
 | `extra_context` | string | (empty) | Free-form text appended to all LLM system prompts. Useful for telling the model about your house or regional conventions. Currency is handled automatically via `[locale]`. |
-| `timeout` | string | `"5s"` | Max wait time for quick LLM operations (ping, model listing). Go duration syntax, e.g. `"10s"`, `"500ms"`. Increase for slow servers. |
+| `timeout` | string | `"5m"` | Max time for a single LLM response (including streaming). Go duration syntax, e.g. `"10m"`. Increase for slow models. |
 | `thinking` | bool | (unset) | Enable model thinking mode (e.g. qwen3 `<think>` blocks). Unset = don't send the option (server default). |
 
 ### `[llm.chat]` section
@@ -391,11 +388,20 @@ dates, vendor matching) from uploaded documents.
 | Key | Type | Default | Description |
 |-----|------|---------|-------------|
 | `model` | string | (chat model) | **Deprecated.** Use `[llm.extraction] model` instead. Falls back to `llm.model` if empty. |
-| `text_timeout` | string | `"30s"` | Max time for `pdftotext` to run. Go duration syntax, e.g. `"1m"`. Increase for very large PDFs. |
 | `max_pages` | int | `0` | Maximum pages to OCR per scanned document. 0 means no limit. |
-| `enabled` | bool | `true` | Set to `false` to disable LLM-powered extraction. When disabled, no structured data is extracted from documents. |
+| `enable` | bool | `true` | Set to `false` to disable LLM-powered structured extraction. OCR and pdftotext still run (see `[extraction.ocr]`). |
+| `enabled` | bool | -- | **Deprecated.** Use `enable` instead. |
 | `thinking` | bool | `false` | **Deprecated.** Use `[llm.extraction] thinking` instead. |
 
+### `[extraction.ocr]` section
+
+OCR sub-pipeline settings. Requires `tesseract` and `pdftocairo`.
+
+| Key | Type | Default | Description |
+|-----|------|---------|-------------|
+| `enable` | bool | `true` | Set to `false` to disable OCR on documents. When disabled, scanned pages and images produce no text. |
+| `confidence_threshold` | int | `0` | Minimum tesseract word confidence (0-100) to keep. Words below this threshold are dropped. 0 means no filtering. |
+
 ### `[locale]` section
 
 Locale and currency settings. Controls currency formatting across all money
diff --git a/internal/config/config.go b/internal/config/config.go
index af9108b4..880ed530 100644
--- a/internal/config/config.go
+++ b/internal/config/config.go
@@ -264,14 +264,14 @@ type Extraction struct {
 	// documents. 0 means no limit (all pages). Default: 0.
 	MaxPages int `toml:"max_pages"`
 
-	// Enabled controls whether LLM-powered extraction runs when a document
-	// is uploaded. When disabled, no structured data is extracted -- OCR and
-	// pdftotext are internal pipeline steps, not standalone features. Default: true.
-	Enabled *bool `toml:"enabled,omitempty"`
+	// Enable controls whether LLM-powered structured extraction runs when
+	// a document is uploaded. When disabled, no structured data is extracted
+	// from documents. OCR and pdftotext still run independently (controlled
+	// by [extraction.ocr]) to populate the document's stored text. Default: true.
+	Enable *bool `toml:"enable,omitempty"`
 
-	// TextTimeout is the maximum time to wait for pdftotext. Go duration
-	// string, e.g. "30s", "1m". Default: "30s".
-	TextTimeout string `toml:"text_timeout"`
+	// Enabled is the deprecated spelling; migrated to Enable on load.
+	Enabled *bool `toml:"enabled,omitempty"`
 
 	// LLMTimeout is the maximum time to wait for the LLM extraction
 	// inference step. Go duration string, e.g. "5m", "90s". Default: "5m".
@@ -281,28 +281,39 @@ type Extraction struct {
 	// Supported values: none, low, medium, high, auto.
 	// Empty string = don't send (server default). Default: empty.
 	Thinking string `toml:"thinking,omitempty"`
+
+	// OCR holds settings for the OCR sub-pipeline.
+	OCR OCR `toml:"ocr" doc:"OCR sub-pipeline. Requires tesseract and pdftocairo."`
+}
+
+// OCR holds settings for the OCR sub-pipeline within extraction.
+type OCR struct {
+	// Enable controls whether OCR runs on uploaded documents.
+	// When disabled, scanned pages and images produce no text. Default: true.
+	Enable *bool `toml:"enable,omitempty"`
+
+	// ConfidenceThreshold is the minimum tesseract word confidence (0-100)
+	// to keep in OCR output. Words below this threshold are dropped.
+	// 0 means no filtering (all words kept). Default: 0.
+	ConfidenceThreshold int `toml:"confidence_threshold"`
 }
 
 // IsEnabled returns whether LLM extraction is enabled. Defaults to true
 // when the field is unset.
 func (e Extraction) IsEnabled() bool {
-	if e.Enabled != nil {
-		return *e.Enabled
+	if e.Enable != nil {
+		return *e.Enable
 	}
 	return true
 }
 
-// TextTimeoutDuration returns the parsed text extraction timeout, falling
-// back to DefaultTextTimeout if the value is empty or unparseable.
-func (e Extraction) TextTimeoutDuration() time.Duration {
-	if e.TextTimeout == "" {
-		return DefaultTextTimeout
-	}
-	d, err := time.ParseDuration(e.TextTimeout)
-	if err != nil {
-		return DefaultTextTimeout
+// IsOCREnabled returns whether OCR is enabled. Defaults to true when
+// the field is unset.
+func (e Extraction) IsOCREnabled() bool {
+	if e.OCR.Enable != nil {
+		return *e.OCR.Enable
 	}
-	return d
+	return true
 }
 
 // LLMTimeoutDuration returns the parsed LLM extraction timeout, falling
@@ -341,7 +352,6 @@ const (
 	DefaultLLMExtractionTimeout = DefaultLLMTimeout
 	DefaultCacheTTL             = 30 * 24 * time.Hour // 30 days
 	DefaultMaxPages             = 0
-	DefaultTextTimeout          = 30 * time.Second
 	configRelPath               = "micasa/config.toml"
 )
 
@@ -378,6 +388,15 @@ func LoadFromPath(path string) (Config, error) {
 		return cfg, err
 	}
 
+	// Clear deprecated Enabled again: applyEnvOverrides may have
+	// repopulated it from MICASA_EXTRACTION_ENABLED.
+	if cfg.Extraction.Enabled != nil {
+		if cfg.Extraction.Enable == nil {
+			cfg.Extraction.Enable = cfg.Extraction.Enabled
+		}
+		cfg.Extraction.Enabled = nil
+	}
+
 	// Normalize base URLs: strip trailing slash and /v1 suffix --
 	// providers handle their own path construction.
 	cfg.LLM.BaseURL = normalizeBaseURL(cfg.LLM.BaseURL)
@@ -488,22 +507,6 @@ func LoadFromPath(path string) (Config, error) {
 		)
 	}
 
-	if cfg.Extraction.TextTimeout != "" {
-		d, err := time.ParseDuration(cfg.Extraction.TextTimeout)
-		if err != nil {
-			return cfg, fmt.Errorf(
-				"extraction.text_timeout: invalid duration %q -- use Go syntax like \"30s\" or \"1m\"",
-				cfg.Extraction.TextTimeout,
-			)
-		}
-		if d <= 0 {
-			return cfg, fmt.Errorf(
-				"extraction.text_timeout must be positive, got %s",
-				cfg.Extraction.TextTimeout,
-			)
-		}
-	}
-
 	if cfg.Extraction.LLMTimeout != "" {
 		d, err := time.ParseDuration(cfg.Extraction.LLMTimeout)
 		if err != nil {
@@ -527,6 +530,13 @@ func LoadFromPath(path string) (Config, error) {
 		)
 	}
 
+	if cfg.Extraction.OCR.ConfidenceThreshold < 0 || cfg.Extraction.OCR.ConfidenceThreshold > 100 {
+		return cfg, fmt.Errorf(
+			"extraction.ocr.confidence_threshold must be 0-100, got %d",
+			cfg.Extraction.OCR.ConfidenceThreshold,
+		)
+	}
+
 	checkFilePermissions(&cfg, path)
 
 	return cfg, nil
@@ -905,6 +915,17 @@ func migrateRenamedKeys(cfg *Config, md toml.MetaData, path string) {
 		)
 	}
 
+	// extraction.enabled -> extraction.enable (v1.78)
+	if md.IsDefined("extraction", "enabled") {
+		if !md.IsDefined("extraction", "enable") {
+			cfg.Extraction.Enable = cfg.Extraction.Enabled
+		}
+		cfg.Warnings = append(cfg.Warnings,
+			"extraction.enabled is deprecated -- use extraction.enable instead",
+		)
+	}
+	cfg.Extraction.Enabled = nil // never propagate the deprecated field
+
 	// extraction.model -> llm.extraction.model (v1.59)
 	if md.IsDefined("extraction", "model") && !md.IsDefined("llm", "extraction", "model") {
 		cfg.LLM.Extraction.Model = cfg.Extraction.Model
@@ -926,6 +947,9 @@ func migrateRenamedKeys(cfg *Config, md toml.MetaData, path string) {
 // replacements. Processed newest-first so that the most recent intermediate
 // name wins when multiple generations of the same variable are set.
 var envRenames = []struct{ old, canonical string }{
+	// v1.78: extraction.enabled -> extraction.enable.
+	{"MICASA_EXTRACTION_ENABLED", "MICASA_EXTRACTION_ENABLE"},
+
 	// v1.77: env var names now derived from dotted config paths.
 	{"MICASA_CURRENCY", "MICASA_LOCALE_CURRENCY"},
 	{"MICASA_MAX_DOCUMENT_SIZE", "MICASA_DOCUMENTS_MAX_FILE_SIZE"},
@@ -934,7 +958,6 @@ var envRenames = []struct{ old, canonical string }{
 	{"MICASA_FILE_PICKER_DIR", "MICASA_DOCUMENTS_FILE_PICKER_DIR"},
 	{"MICASA_EXTRACTION_MAX_EXTRACT_PAGES", "MICASA_EXTRACTION_MAX_PAGES"},
 	{"MICASA_MAX_EXTRACT_PAGES", "MICASA_EXTRACTION_MAX_PAGES"},
-	{"MICASA_TEXT_TIMEOUT", "MICASA_EXTRACTION_TEXT_TIMEOUT"},
 
 	// v1.59
 	{"MICASA_EXTRACTION_MODEL", "MICASA_LLM_EXTRACTION_MODEL"},
@@ -1102,9 +1125,9 @@ model = "` + DefaultModel + `"
 # file_picker_dir = "/home/user/Documents"
 
 [extraction]
-# Timeout for pdftotext. Go duration syntax: "30s", "1m", etc. Default: "30s".
-# Increase if you routinely process very large PDFs.
-# text_timeout = "30s"
+# Set to false to disable LLM-powered structured extraction. OCR and pdftotext
+# still run (see [extraction.ocr]) to populate document text for search/display.
+# enable = true
 
 # Timeout for LLM extraction inference. Go duration syntax: "5m", "90s", etc.
 # Default: "5m". Increase for slow local models or complex documents.
@@ -1113,9 +1136,14 @@ model = "` + DefaultModel + `"
 # Maximum pages for async extraction of scanned documents. 0 = no limit. Default: 0.
 # max_pages = 0
 
-# Set to false to disable LLM-powered extraction even when LLM is configured.
-# When disabled, no structured data is extracted from documents.
-# enabled = true
+# [extraction.ocr]
+# Set to false to disable OCR on uploaded documents. When disabled, scanned
+# pages and images produce no text. Default: true.
+# enable = true
+
+# Minimum tesseract word confidence (0-100) to keep. Words below this
+# threshold are dropped. 0 = no filtering. Default: 0.
+# confidence_threshold = 0
 
 [locale]
 # ISO 4217 currency code. Stored in the database on first run; after that the
diff --git a/internal/config/config_test.go b/internal/config/config_test.go
index 486f57a0..e8b616d4 100644
--- a/internal/config/config_test.go
+++ b/internal/config/config_test.go
@@ -462,7 +462,7 @@ func TestExtractionFromFile(t *testing.T) {
 	path := writeConfig(t, `[extraction]
 model = "qwen2.5:7b"
 max_pages = 10
-enabled = false
+enable = false
 `)
 	cfg, err := LoadFromPath(path)
 	require.NoError(t, err)
@@ -485,7 +485,7 @@ func TestExtractionResolvedModel(t *testing.T) {
 func TestExtractionEnvOverrides(t *testing.T) {
 	t.Setenv("MICASA_EXTRACTION_MODEL", "phi3")
 	t.Setenv("MICASA_EXTRACTION_MAX_PAGES", "5")
-	t.Setenv("MICASA_EXTRACTION_ENABLED", "false")
+	t.Setenv("MICASA_EXTRACTION_ENABLE", "false")
 
 	cfg, err := LoadFromPath(noConfig(t))
 	require.NoError(t, err)
@@ -510,7 +510,7 @@ func TestInvalidEnvVarReturnsError(t *testing.T) {
 		wantMsg string
 	}{
 		{"MICASA_EXTRACTION_MAX_PAGES", "not-a-number", "expected integer"},
-		{"MICASA_EXTRACTION_ENABLED", "maybe", "expected true or false"},
+		{"MICASA_EXTRACTION_ENABLE", "maybe", "expected true or false"},
 		{"MICASA_DOCUMENTS_MAX_FILE_SIZE", "lots", "expected byte size"},
 		{"MICASA_DOCUMENTS_CACHE_TTL", "forever", "expected duration"},
 		{"MICASA_DOCUMENTS_CACHE_TTL_DAYS", "many", "expected integer"},
@@ -629,12 +629,15 @@ func TestEnvVars(t *testing.T) {
 		"MICASA_DOCUMENTS_CACHE_TTL_DAYS":  "documents.cache_ttl_days",
 		"MICASA_DOCUMENTS_FILE_PICKER_DIR": "documents.file_picker_dir",
 
-		"MICASA_EXTRACTION_MODEL":        "extraction.model",
-		"MICASA_EXTRACTION_MAX_PAGES":    "extraction.max_pages",
-		"MICASA_EXTRACTION_ENABLED":      "extraction.enabled",
-		"MICASA_EXTRACTION_TEXT_TIMEOUT": "extraction.text_timeout",
-		"MICASA_EXTRACTION_LLM_TIMEOUT":  "extraction.llm_timeout",
-		"MICASA_EXTRACTION_THINKING":     "extraction.thinking",
+		"MICASA_EXTRACTION_MODEL":       "extraction.model",
+		"MICASA_EXTRACTION_MAX_PAGES":   "extraction.max_pages",
+		"MICASA_EXTRACTION_ENABLE":      "extraction.enable",
+		"MICASA_EXTRACTION_ENABLED":     "extraction.enabled",
+		"MICASA_EXTRACTION_LLM_TIMEOUT": "extraction.llm_timeout",
+		"MICASA_EXTRACTION_THINKING":    "extraction.thinking",
+
+		"MICASA_EXTRACTION_OCR_ENABLE":               "extraction.ocr.enable",
+		"MICASA_EXTRACTION_OCR_CONFIDENCE_THRESHOLD": "extraction.ocr.confidence_threshold",
 
 		"MICASA_LOCALE_CURRENCY": "locale.currency",
 
@@ -664,7 +667,7 @@ func TestEnvVarName(t *testing.T) {
 	}{
 		{"llm.model", "MICASA_LLM_MODEL"},
 		{"documents.max_file_size", "MICASA_DOCUMENTS_MAX_FILE_SIZE"},
-		{"extraction.text_timeout", "MICASA_EXTRACTION_TEXT_TIMEOUT"},
+		{"extraction.llm_timeout", "MICASA_EXTRACTION_LLM_TIMEOUT"},
 		{"locale.currency", "MICASA_LOCALE_CURRENCY"},
 		{"llm.chat.provider", "MICASA_LLM_CHAT_PROVIDER"},
 	}
@@ -739,6 +742,86 @@ func TestMaxPagesEnvMigrationChain(t *testing.T) {
 		"MICASA_MAX_EXTRACT_PAGES (newer) should take precedence")
 }
 
+func TestEnabledTOMLMigration(t *testing.T) {
+	path := writeConfig(t, "[extraction]\nenabled = false\n")
+	cfg, err := LoadFromPath(path)
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsEnabled())
+	require.Len(t, cfg.Warnings, 1)
+	assert.Contains(t, cfg.Warnings[0], "extraction.enabled")
+	assert.Contains(t, cfg.Warnings[0], "extraction.enable")
+}
+
+func TestEnabledTOMLMigrationIgnoredWhenNewKeySet(t *testing.T) {
+	path := writeConfig(t, "[extraction]\nenabled = true\nenable = false\n")
+	cfg, err := LoadFromPath(path)
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsEnabled(), "new key takes precedence")
+	require.Len(t, cfg.Warnings, 1)
+	assert.Contains(t, cfg.Warnings[0], "extraction.enabled")
+}
+
+func TestEnabledEnvMigration(t *testing.T) {
+	t.Setenv("MICASA_EXTRACTION_ENABLED", "false")
+	cfg, err := LoadFromPath(noConfig(t))
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsEnabled())
+	assert.Nil(t, cfg.Extraction.Enabled, "deprecated Enabled must be cleared after env migration")
+	require.Len(t, cfg.Warnings, 1)
+	assert.Contains(t, cfg.Warnings[0], "MICASA_EXTRACTION_ENABLED")
+	assert.Contains(t, cfg.Warnings[0], "MICASA_EXTRACTION_ENABLE")
+}
+
+func TestEnabledEnvIgnoredWhenNewEnvSet(t *testing.T) {
+	t.Setenv("MICASA_EXTRACTION_ENABLED", "true")
+	t.Setenv("MICASA_EXTRACTION_ENABLE", "false")
+	cfg, err := LoadFromPath(noConfig(t))
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsEnabled())
+	assert.Empty(t, cfg.Warnings)
+}
+
+// --- OCR config ---
+
+func TestOCRDefaults(t *testing.T) {
+	cfg, err := LoadFromPath(noConfig(t))
+	require.NoError(t, err)
+	assert.True(t, cfg.Extraction.IsOCREnabled())
+	assert.Equal(t, 0, cfg.Extraction.OCR.ConfidenceThreshold)
+}
+
+func TestOCRFromFile(t *testing.T) {
+	path := writeConfig(t, "[extraction.ocr]\nenable = false\nconfidence_threshold = 70\n")
+	cfg, err := LoadFromPath(path)
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsOCREnabled())
+	assert.Equal(t, 70, cfg.Extraction.OCR.ConfidenceThreshold)
+}
+
+func TestOCREnvOverrides(t *testing.T) {
+	t.Setenv("MICASA_EXTRACTION_OCR_ENABLE", "false")
+	t.Setenv("MICASA_EXTRACTION_OCR_CONFIDENCE_THRESHOLD", "80")
+	cfg, err := LoadFromPath(noConfig(t))
+	require.NoError(t, err)
+	assert.False(t, cfg.Extraction.IsOCREnabled())
+	assert.Equal(t, 80, cfg.Extraction.OCR.ConfidenceThreshold)
+}
+
+func TestOCRConfidenceThresholdValidation(t *testing.T) {
+	t.Run("rejects negative", func(t *testing.T) {
+		path := writeConfig(t, "[extraction.ocr]\nconfidence_threshold = -1\n")
+		_, err := LoadFromPath(path)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "confidence_threshold must be 0-100")
+	})
+	t.Run("rejects over 100", func(t *testing.T) {
+		path := writeConfig(t, "[extraction.ocr]\nconfidence_threshold = 101\n")
+		_, err := LoadFromPath(path)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "confidence_threshold must be 0-100")
+	})
+}
+
 // --- Per-pipeline LLM config ---
 
 func TestChatConfigInheritsBase(t *testing.T) {
@@ -1075,7 +1158,6 @@ func TestDeprecatedEnvVarRenames(t *testing.T) {
 		{"MICASA_MAX_DOCUMENT_SIZE", "MICASA_DOCUMENTS_MAX_FILE_SIZE", "100 MiB"},
 		{"MICASA_CACHE_TTL", "MICASA_DOCUMENTS_CACHE_TTL", "7d"},
 		{"MICASA_FILE_PICKER_DIR", "MICASA_DOCUMENTS_FILE_PICKER_DIR", "/tmp"},
-		{"MICASA_TEXT_TIMEOUT", "MICASA_EXTRACTION_TEXT_TIMEOUT", "1m"},
 	}
 	for _, tt := range tests {
 		t.Run(tt.old, func(t *testing.T) {
diff --git a/internal/config/show.go b/internal/config/show.go
index 57933dc6..2042a5d8 100644
--- a/internal/config/show.go
+++ b/internal/config/show.go
@@ -28,6 +28,7 @@ var hiddenPaths = map[string]bool{
 // "DEPRECATED: use <replacement>" warning.
 var deprecatedPaths = map[string]string{
 	"documents.cache_ttl_days": "documents.cache_ttl",
+	"extraction.enabled":       "extraction.enable",
 	"extraction.model":         "llm.extraction.model",
 	"extraction.thinking":      "llm.extraction.thinking",
 }
@@ -74,12 +75,13 @@ func (c Config) forDisplay() Config {
 		d.Documents.CacheTTL = &Duration{dur}
 	}
 	// CacheTTLDays preserved when user-set so the dump warns about it.
-	if d.Extraction.Enabled == nil {
+	if d.Extraction.Enable == nil {
 		t := true
-		d.Extraction.Enabled = &t
+		d.Extraction.Enable = &t
 	}
-	if d.Extraction.TextTimeout == "" {
-		d.Extraction.TextTimeout = DefaultTextTimeout.String()
+	if d.Extraction.OCR.Enable == nil {
+		t := true
+		d.Extraction.OCR.Enable = &t
 	}
 	if d.Locale.Currency == "" {
 		d.Locale.Currency = detectCurrencyCode()
@@ -335,16 +337,21 @@ func writeAligned(w io.Writer, blk sectionBlock) error {
 	return nil
 }
 
-// FormatDuration formats a duration in a human-friendly way, using day
-// notation for whole-day multiples.
+// FormatDuration formats a duration in a human-friendly way, using
+// clean notation for whole-unit multiples (days, hours, minutes).
 func FormatDuration(d time.Duration) string {
-	if d == 0 {
+	switch {
+	case d == 0:
 		return "0s"
-	}
-	if d%(24*time.Hour) == 0 {
+	case d%(24*time.Hour) == 0:
 		return fmt.Sprintf("%dd", d/(24*time.Hour))
+	case d%time.Hour == 0:
+		return fmt.Sprintf("%dh", d/time.Hour)
+	case d%time.Minute == 0:
+		return fmt.Sprintf("%dm", d/time.Minute)
+	default:
+		return d.String()
 	}
-	return d.String()
 }
 
 // formatTOMLValue formats a reflected value as a TOML value string.
diff --git a/internal/config/show_test.go b/internal/config/show_test.go
index 2a84940b..57bf27ac 100644
--- a/internal/config/show_test.go
+++ b/internal/config/show_test.go
@@ -38,8 +38,7 @@ func TestShowConfigDefaults(t *testing.T) {
 	assert.Contains(t, out, `max_file_size = "50 MiB"`)
 	assert.Contains(t, out, `cache_ttl = "30d"`)
 	assert.Contains(t, out, "max_pages = 0")
-	assert.Contains(t, out, "enabled = true")
-	assert.Contains(t, out, `text_timeout = "30s"`)
+	assert.Contains(t, out, "enable = true")
 
 	assert.NotContains(t, out, "cache_ttl_days")
 	assert.NotContains(t, out, "[llm.chat]")
@@ -63,7 +62,7 @@ cache_ttl = "7d"
 
 [extraction]
 max_pages = 10
-enabled = false
+enable = false
 `)
 	cfg, err := LoadFromPath(path)
 	require.NoError(t, err)
@@ -95,8 +94,7 @@ cache_ttl = "7d"
 
 [extraction]
 max_pages = 10
-enabled = false
-text_timeout = "1m"
+enable = false
 `)
 	orig, err := LoadFromPath(path)
 	require.NoError(t, err)
@@ -123,7 +121,6 @@ text_timeout = "1m"
 		parsed.Documents.CacheTTLDuration())
 	assert.Equal(t, orig.Extraction.MaxPages, parsed.Extraction.MaxPages)
 	assert.Equal(t, orig.Extraction.IsEnabled(), parsed.Extraction.IsEnabled())
-	assert.Equal(t, orig.Extraction.TextTimeout, parsed.Extraction.TextTimeout)
 
 	// API keys are hidden -- the parsed config must NOT have them.
 	assert.Empty(t, parsed.LLM.APIKey)
@@ -352,9 +349,12 @@ func TestFormatDuration(t *testing.T) {
 	}{
 		{0, "0s"},
 		{5 * time.Second, "5s"},
+		{5 * time.Minute, "5m"},
+		{90 * time.Minute, "90m"},
+		{2 * time.Hour, "2h"},
 		{30 * 24 * time.Hour, "30d"},
 		{7 * 24 * time.Hour, "7d"},
-		{90 * time.Minute, "1h30m0s"},
+		{90*time.Minute + 30*time.Second, "1h30m30s"},
 		{500 * time.Millisecond, "500ms"},
 	}
 	for _, tt := range tests {
diff --git a/internal/extract/extractor.go b/internal/extract/extractor.go
index bf491f0b..ac51ee0e 100644
--- a/internal/extract/extractor.go
+++ b/internal/extract/extractor.go
@@ -31,13 +31,25 @@ type Extractor interface {
 // DefaultExtractors returns the standard extractors in priority order:
 // pdftotext, plaintext, PDF OCR, image OCR. maxPages of 0 means no limit
 // (all pages). Zero timeout causes the concrete extractor to use its default.
-func DefaultExtractors(maxPages int, timeout time.Duration) []Extractor {
-	return []Extractor{
+// ocrEnabled controls whether OCR extractors are included (default true).
+// confidenceThreshold filters OCR words below this confidence (0 = no filter).
+func DefaultExtractors(
+	maxPages int,
+	timeout time.Duration,
+	ocrEnabled bool,
+	confidenceThreshold int,
+) []Extractor {
+	ext := []Extractor{
 		&PDFTextExtractor{Timeout: timeout},
 		&PlainTextExtractor{},
-		&PDFOCRExtractor{MaxPages: maxPages},
-		&ImageOCRExtractor{},
 	}
+	if ocrEnabled {
+		ext = append(ext,
+			&PDFOCRExtractor{MaxPages: maxPages, ConfidenceThreshold: confidenceThreshold},
+			&ImageOCRExtractor{ConfidenceThreshold: confidenceThreshold},
+		)
+	}
+	return ext
 }
 
 // HasMatchingExtractor reports whether any extractor in the list with
@@ -141,7 +153,8 @@ func (e *PlainTextExtractor) Extract(_ context.Context, data []byte) (TextSource
 
 // PDFOCRExtractor wraps ocrPDF for scanned PDF pages.
 type PDFOCRExtractor struct {
-	MaxPages int
+	MaxPages            int
+	ConfidenceThreshold int
 }
 
 func (e *PDFOCRExtractor) Tool() string             { return "tesseract" }
@@ -156,6 +169,10 @@ func (e *PDFOCRExtractor) Extract(ctx context.Context, data []byte) (TextSource,
 	if err != nil {
 		return TextSource{}, err
 	}
+	if e.ConfidenceThreshold > 0 {
+		tsv = filterTSVByConfidence(tsv, e.ConfidenceThreshold)
+		text = textFromTSV(tsv)
+	}
 	return TextSource{
 		Tool: "tesseract",
 		Desc: "Text recognized from rasterized page images. Covers scanned pages that pdftotext misses, but may contain OCR errors.",
@@ -165,7 +182,9 @@ func (e *PDFOCRExtractor) Extract(ctx context.Context, data []byte) (TextSource,
 }
 
 // ImageOCRExtractor wraps ocrImage for direct image OCR.
-type ImageOCRExtractor struct{}
+type ImageOCRExtractor struct {
+	ConfidenceThreshold int
+}
 
 func (e *ImageOCRExtractor) Tool() string             { return "tesseract" }
 func (e *ImageOCRExtractor) Matches(mime string) bool { return IsImageMIME(mime) }
@@ -179,6 +198,10 @@ func (e *ImageOCRExtractor) Extract(ctx context.Context, data []byte) (TextSourc
 	if err != nil {
 		return TextSource{}, err
 	}
+	if e.ConfidenceThreshold > 0 {
+		tsv = filterTSVByConfidence(tsv, e.ConfidenceThreshold)
+		text = textFromTSV(tsv)
+	}
 	return TextSource{
 		Tool: "tesseract",
 		Desc: "Text recognized from the image. May contain OCR errors.",
diff --git a/internal/extract/extractor_test.go b/internal/extract/extractor_test.go
index c9aa0232..247a8299 100644
--- a/internal/extract/extractor_test.go
+++ b/internal/extract/extractor_test.go
@@ -150,7 +150,7 @@ func TestImageOCRExtractor_Available(t *testing.T) {
 
 func TestDefaultExtractors_Order(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	require.Len(t, extractors, 4)
 	assert.Equal(t, "pdftotext", extractors[0].Tool())
 	assert.Equal(t, "plaintext", extractors[1].Tool())
@@ -166,7 +166,7 @@ func TestDefaultExtractors_Order(t *testing.T) {
 
 func TestDefaultExtractors_Passthrough(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(42, 99)
+	extractors := DefaultExtractors(42, 99, true, 0)
 	pdfExt, ok := extractors[0].(*PDFTextExtractor)
 	require.True(t, ok)
 	assert.Equal(t, 99, int(pdfExt.Timeout))
@@ -176,32 +176,52 @@ func TestDefaultExtractors_Passthrough(t *testing.T) {
 	assert.Equal(t, 42, ocrExt.MaxPages)
 }
 
+func TestDefaultExtractors_OCRDisabled(t *testing.T) {
+	t.Parallel()
+	extractors := DefaultExtractors(0, 0, false, 0)
+	require.Len(t, extractors, 2)
+	assert.Equal(t, "pdftotext", extractors[0].Tool())
+	assert.Equal(t, "plaintext", extractors[1].Tool())
+}
+
+func TestDefaultExtractors_ConfidencePassthrough(t *testing.T) {
+	t.Parallel()
+	extractors := DefaultExtractors(0, 0, true, 70)
+	pdfOCR, ok := extractors[2].(*PDFOCRExtractor)
+	require.True(t, ok)
+	assert.Equal(t, 70, pdfOCR.ConfidenceThreshold)
+
+	imgOCR, ok := extractors[3].(*ImageOCRExtractor)
+	require.True(t, ok)
+	assert.Equal(t, 70, imgOCR.ConfidenceThreshold)
+}
+
 // --- HasMatchingExtractor ---
 
 func TestHasMatchingExtractor_Tesseract_PDF(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	got := HasMatchingExtractor(extractors, "tesseract", "application/pdf")
 	assert.Equal(t, OCRAvailable(), got)
 }
 
 func TestHasMatchingExtractor_Tesseract_Image(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	got := HasMatchingExtractor(extractors, "tesseract", "image/png")
 	assert.Equal(t, ImageOCRAvailable(), got)
 }
 
 func TestHasMatchingExtractor_Pdftotext(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	got := HasMatchingExtractor(extractors, "pdftotext", "application/pdf")
 	assert.Equal(t, HasPDFToText(), got)
 }
 
 func TestHasMatchingExtractor_NoMatch(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	assert.False(t, HasMatchingExtractor(extractors, "tesseract", "text/plain"))
 	assert.False(t, HasMatchingExtractor(extractors, "pdftotext", "image/png"))
 	assert.False(t, HasMatchingExtractor(extractors, "nonexistent", "application/pdf"))
@@ -211,21 +231,21 @@ func TestHasMatchingExtractor_NoMatch(t *testing.T) {
 
 func TestNeedsOCR_PDF(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	got := NeedsOCR(extractors, "application/pdf")
 	assert.Equal(t, OCRAvailable(), got)
 }
 
 func TestNeedsOCR_Image(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	got := NeedsOCR(extractors, "image/png")
 	assert.Equal(t, ImageOCRAvailable(), got)
 }
 
 func TestNeedsOCR_PlainText(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 0)
+	extractors := DefaultExtractors(0, 0, true, 0)
 	assert.False(t, NeedsOCR(extractors, "text/plain"))
 }
 
@@ -240,7 +260,7 @@ func TestNeedsOCR_NoOCRExtractors(t *testing.T) {
 
 func TestExtractorTimeout(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(0, 42)
+	extractors := DefaultExtractors(0, 42, true, 0)
 	assert.Equal(t, time.Duration(42), ExtractorTimeout(extractors))
 }
 
@@ -252,7 +272,7 @@ func TestExtractorTimeout_NoPDFText(t *testing.T) {
 
 func TestExtractorMaxPages(t *testing.T) {
 	t.Parallel()
-	extractors := DefaultExtractors(15, 0)
+	extractors := DefaultExtractors(15, 0, true, 0)
 	assert.Equal(t, 15, ExtractorMaxPages(extractors))
 }
 
diff --git a/internal/extract/ocr.go b/internal/extract/ocr.go
index 26b2b7c7..e01d3055 100644
--- a/internal/extract/ocr.go
+++ b/internal/extract/ocr.go
@@ -367,6 +367,33 @@ func atoi(b []byte) int {
 	return n
 }
 
+// filterTSVByConfidence removes rows from tesseract TSV output whose
+// confidence (column 10) is below the threshold.
+func filterTSVByConfidence(tsv []byte, threshold int) []byte {
+	lines := bytes.Split(tsv, []byte("\n"))
+	if len(lines) < 2 {
+		return tsv
+	}
+	var out bytes.Buffer
+	out.Write(lines[0]) // header
+	out.WriteByte('\n')
+	for _, line := range lines[1:] {
+		if len(line) == 0 {
+			continue
+		}
+		fields := bytes.Split(line, []byte("\t"))
+		if len(fields) < 12 {
+			continue
+		}
+		conf := atoi(fields[10])
+		if conf >= threshold {
+			out.Write(line)
+			out.WriteByte('\n')
+		}
+	}
+	return out.Bytes()
+}
+
 // IsImageMIME reports whether the MIME type is an image format that
 // tesseract can process.
 func IsImageMIME(mime string) bool {
diff --git a/internal/extract/ocr_coverage_test.go b/internal/extract/ocr_coverage_test.go
index 9041f5f4..a57170e3 100644
--- a/internal/extract/ocr_coverage_test.go
+++ b/internal/extract/ocr_coverage_test.go
@@ -807,7 +807,7 @@ func TestExtractWithProgress_PDF_InvalidData(t *testing.T) {
 		context.Background(),
 		[]byte("not a pdf"),
 		"application/pdf",
-		DefaultExtractors(5, 0),
+		DefaultExtractors(5, 0, true, 0),
 	)
 
 	var gotErr bool
@@ -1018,7 +1018,7 @@ func TestExtractWithProgress_Image_InvalidData(t *testing.T) {
 		context.Background(),
 		[]byte("not an image"),
 		"image/png",
-		DefaultExtractors(5, 0),
+		DefaultExtractors(5, 0, true, 0),
 	)
 
 	var gotErr bool
diff --git a/internal/extract/ocr_progress_test.go b/internal/extract/ocr_progress_test.go
index b46cc7e4..b7674bcd 100644
--- a/internal/extract/ocr_progress_test.go
+++ b/internal/extract/ocr_progress_test.go
@@ -22,7 +22,7 @@ func TestExtractWithProgress_EmptyData(t *testing.T) {
 		context.Background(),
 		nil,
 		"application/pdf",
-		DefaultExtractors(20, 0),
+		DefaultExtractors(20, 0, true, 0),
 	)
 	msg := <-ch
 	assert.True(t, msg.Done)
@@ -37,7 +37,12 @@ func TestExtractWithProgress_EmptyData(t *testing.T) {
 // TestExtractWithProgress_EmptyImage verifies the image path with empty data.
 func TestExtractWithProgress_EmptyImage(t *testing.T) {
 	t.Parallel()
-	ch := ExtractWithProgress(context.Background(), nil, "image/png", DefaultExtractors(20, 0))
+	ch := ExtractWithProgress(
+		context.Background(),
+		nil,
+		"image/png",
+		DefaultExtractors(20, 0, true, 0),
+	)
 	msg := <-ch
 	assert.True(t, msg.Done)
 	assert.Empty(t, msg.Text)
@@ -52,7 +57,12 @@ func TestExtractWithProgress_ContextCancelled(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	cancel() // cancel immediately
 
-	ch := ExtractWithProgress(ctx, []byte("fake image data"), "image/png", DefaultExtractors(20, 0))
+	ch := ExtractWithProgress(
+		ctx,
+		[]byte("fake image data"),
+		"image/png",
+		DefaultExtractors(20, 0, true, 0),
+	)
 
 	var gotErr bool
 	for msg := range ch {
@@ -78,7 +88,12 @@ func TestExtractWithProgress_Image_Integration(t *testing.T) {
 		skipOrFatalCI(t, "test fixture not found: "+imgPath)
 	}
 
-	ch := ExtractWithProgress(context.Background(), data, "image/png", DefaultExtractors(20, 0))
+	ch := ExtractWithProgress(
+		context.Background(),
+		data,
+		"image/png",
+		DefaultExtractors(20, 0, true, 0),
+	)
 
 	var progressCount int
 	var finalText string
@@ -117,7 +132,7 @@ func TestExtractWithProgress_PDF_Integration(t *testing.T) {
 		context.Background(),
 		data,
 		"application/pdf",
-		DefaultExtractors(5, 0),
+		DefaultExtractors(5, 0, true, 0),
 	)
 
 	var phases []string
diff --git a/internal/extract/ocr_test.go b/internal/extract/ocr_test.go
index 24e47604..ae1f2ca9 100644
--- a/internal/extract/ocr_test.go
+++ b/internal/extract/ocr_test.go
@@ -78,6 +78,28 @@ func TestAtoi(t *testing.T) {
 	}
 }
 
+func TestFilterTSVByConfidence(t *testing.T) {
+	t.Parallel()
+	tsv := []byte(
+		"level\tpage_num\tblock_num\tpar_num\tline_num\tword_num\tleft\ttop\twidth\theight\tconf\ttext\n" +
+			"5\t1\t1\t1\t1\t1\t100\t200\t50\t12\t96\tHigh\n" +
+			"5\t1\t1\t1\t1\t2\t160\t200\t50\t12\t30\tLow\n" +
+			"5\t1\t1\t1\t2\t1\t100\t220\t50\t12\t70\tBorder\n",
+	)
+
+	filtered := filterTSVByConfidence(tsv, 70)
+	text := textFromTSV(filtered)
+	assert.Contains(t, text, "High")
+	assert.NotContains(t, text, "Low")
+	assert.Contains(t, text, "Border")
+}
+
+func TestFilterTSVByConfidence_Empty(t *testing.T) {
+	t.Parallel()
+	assert.Equal(t, []byte(nil), filterTSVByConfidence(nil, 70))
+	assert.Equal(t, []byte(""), filterTSVByConfidence([]byte(""), 70))
+}
+
 func TestIsImageMIME(t *testing.T) {
 	t.Parallel()
 	assert.True(t, IsImageMIME("image/png"))
diff --git a/internal/extract/pipeline.go b/internal/extract/pipeline.go
index eb727d53..8420be03 100644
--- a/internal/extract/pipeline.go
+++ b/internal/extract/pipeline.go
@@ -16,7 +16,7 @@ import (
 // and gracefully degrades when its dependencies are unavailable.
 type Pipeline struct {
 	LLMClient  *llm.Client   // nil = skip LLM extraction
-	Extractors []Extractor   // nil = DefaultExtractors(0, 0)
+	Extractors []Extractor   // nil = DefaultExtractors(0, 0, true, 0)
 	Schema     SchemaContext // DDL + entity rows for prompt
 	DocID      uint          // document ID for UPDATE operations
 }
@@ -72,7 +72,7 @@ func (p *Pipeline) Run(
 
 	extractors := p.Extractors
 	if extractors == nil {
-		extractors = DefaultExtractors(0, 0)
+		extractors = DefaultExtractors(0, 0, true, 0)
 	}
 
 	// Run all matching, available extractors.
diff --git a/internal/extract/pipeline_test.go b/internal/extract/pipeline_test.go
index b1dfbfc7..4f17eb17 100644
--- a/internal/extract/pipeline_test.go
+++ b/internal/extract/pipeline_test.go
@@ -112,7 +112,7 @@ func TestPipeline_OCRIntegration(t *testing.T) {
 	}
 
 	// Both pdftotext and OCR should run for PDFs.
-	p := &Pipeline{Extractors: DefaultExtractors(5, 0)}
+	p := &Pipeline{Extractors: DefaultExtractors(5, 0, true, 0)}
 	r := p.Run(context.Background(), data, "sample.pdf", "application/pdf")
 	require.NoError(t, r.Err)
 	assert.True(t, r.HasSource("tesseract"), "OCR always runs for PDFs")
@@ -141,7 +141,7 @@ func TestPipeline_MixedPDF(t *testing.T) {
 		t.Skipf("test fixture not found (pdfunite unavailable?): %s", pdfPath)
 	}
 
-	p := &Pipeline{Extractors: DefaultExtractors(5, 0)}
+	p := &Pipeline{Extractors: DefaultExtractors(5, 0, true, 0)}
 	r := p.Run(context.Background(), data, "mixed-inspection.pdf", "application/pdf")
 	require.NoError(t, r.Err)