feat(extract): allow accepting extraction results without LLM (#721)

## Summary

- Remove the `HasError` gate on the extraction overlay's accept action
so `a` works regardless of whether the LLM step ran, failed, or was
skipped
- Extracted text (pdftotext/OCR) can now be persisted without requiring
LLM analysis to succeed
- Fix bug where `acceptDeferredExtraction` did not call
`reloadAfterMutation` when there were no non-document operations,
causing accepted deferred documents to not appear in the table

closes #710

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: cpcloud

internal/app/extraction.go

@@ -813,20 +813,19 @@ func applyStringField(data map[string]any, key string, dst *string) {
 }
 
 // acceptExtraction persists all pending results and closes the overlay.
+// Works regardless of whether LLM ran, failed, or was skipped.
 func (m *Model) acceptExtraction() {
 	ex := m.ex.extraction
 	if ex == nil || !ex.Done || ex.accepted {
 		return
 	}
 
 	if ex.pendingDoc != nil {
-		// Deferred creation (magic-add): create document now.
 		if err := m.acceptDeferredExtraction(); err != nil {
 			m.setStatusError(err.Error())
 			return
 		}
 	} else {
-		// Existing document: persist extraction results and dispatch ops.
 		if err := m.acceptExistingExtraction(); err != nil {
 			m.setStatusError(err.Error())
 			return
@@ -879,6 +878,7 @@ func (m *Model) acceptDeferredExtraction() error {
 	if err := m.commitShadowOperations(ex, nonDocOps); err != nil {
 		return fmt.Errorf("dispatch operations: %w", err)
 	}
+	m.reloadAfterMutation()
 	return nil
 }
 
@@ -1075,7 +1075,7 @@ func (m *Model) handleExtractionPipelineKey(msg tea.KeyMsg) tea.Cmd {
 			return m.activateExtractionModelPicker()
 		}
 	case keyA:
-		if ex.Done && !ex.HasError {
+		if ex.Done {
 			m.acceptExtraction()
 		}
 	case keyX:
@@ -1264,7 +1264,7 @@ func (m *Model) handleExtractionExploreKey(msg tea.KeyMsg) tea.Cmd {
 			ex.previewCol = len(g.specs) - 1
 		}
 	case keyA:
-		if ex.Done && !ex.HasError {
+		if ex.Done {
 			m.acceptExtraction()
 		}
 	case keyX:
@@ -1495,10 +1495,7 @@ func (m *Model) buildExtractionPipelineOverlay(
 		if len(ex.previewGroups) > 1 {
 			hints = append(hints, m.helpItem(keyB+"/"+keyF, "tabs"))
 		}
-		if !ex.HasError {
-			hints = append(hints, m.helpItem(keyA, "accept"))
-		}
-		hints = append(hints, m.helpItem(keyX, "back"), m.helpItem(keyEsc, "discard"))
+		hints = append(hints, m.helpItem(keyA, "accept"), m.helpItem(keyX, "back"), m.helpItem(keyEsc, "discard"))
 	} else {
 		hints = append(hints, m.helpItem(keyJ+"/"+keyK, "navigate"))
 		cursorStatus := ex.Steps[ex.cursorStep()].Status
@@ -1509,10 +1506,7 @@ func (m *Model) buildExtractionPipelineOverlay(
 			hints = append(hints, m.helpItem(keyX, "explore"))
 		}
 		if ex.Done {
-			if !ex.HasError {
-				hints = append(hints, m.helpItem(keyA, "accept"))
-			}
-			hints = append(hints, m.helpItem(keyEsc, "discard"))
+			hints = append(hints, m.helpItem(keyA, "accept"), m.helpItem(keyEsc, "discard"))
 		} else {
 			hints = append(hints,
 				m.helpItem(symCtrlC, "int"),

internal/app/extraction_test.go

@@ -2719,3 +2719,115 @@ func TestExtractionExtractFails_PingSkipPreventsLLM(t *testing.T) {
 	assert.Equal(t, stepSkipped, ex.Steps[stepLLM].Status)
 	assert.True(t, ex.Done)
 }
+
+// --- Accept works with or without LLM ---
+
+func TestAccept_WorksWhenLLMFailed(t *testing.T) {
+	t.Parallel()
+	m := newExtractionModel(t, map[extractionStep]stepStatus{
+		stepText: stepDone,
+		stepLLM:  stepFailed,
+	})
+	ex := m.ex.extraction
+	ex.Done = true
+	ex.HasError = true
+	ex.pendingText = "ocr text"
+
+	doc := &data.Document{
+		FileName: "invoice.pdf",
+		MIMEType: "application/pdf",
+		Data:     []byte("pdf-bytes"),
+	}
+	require.NoError(t, m.store.CreateDocument(doc))
+	ex.DocID = doc.ID
+
+	sendExtractionKey(m, "a")
+
+	assert.Nil(t, m.ex.extraction, "accept should work even when LLM failed")
+
+	saved, err := m.store.GetDocument(doc.ID)
+	require.NoError(t, err)
+	assert.Equal(t, "ocr text", saved.ExtractedText)
+}
+
+func TestAccept_WorksWithoutLLMStep(t *testing.T) {
+	t.Parallel()
+	m := newExtractionModel(t, map[extractionStep]stepStatus{
+		stepText:    stepDone,
+		stepExtract: stepDone,
+	})
+	ex := m.ex.extraction
+	ex.Done = true
+	ex.pendingText = "ocr text"
+
+	doc := &data.Document{
+		FileName: "scan.pdf",
+		MIMEType: "application/pdf",
+		Data:     []byte("pdf-bytes"),
+	}
+	require.NoError(t, m.store.CreateDocument(doc))
+	ex.DocID = doc.ID
+
+	sendExtractionKey(m, "a")
+
+	assert.Nil(t, m.ex.extraction, "accept should work without LLM step")
+
+	saved, err := m.store.GetDocument(doc.ID)
+	require.NoError(t, err)
+	assert.Equal(t, "ocr text", saved.ExtractedText)
+}
+
+func TestAccept_DeferredDoc_WorksWhenLLMFailed(t *testing.T) {
+	t.Parallel()
+	m := newExtractionModel(t, map[extractionStep]stepStatus{
+		stepText: stepDone,
+		stepLLM:  stepFailed,
+	})
+	ex := m.ex.extraction
+	ex.Done = true
+	ex.HasError = true
+	ex.extractedText = "invoice text"
+	ex.pendingDoc = &data.Document{
+		FileName:      "invoice.pdf",
+		MIMEType:      "application/pdf",
+		Data:          []byte("pdf-bytes"),
+		ExtractedText: "invoice text",
+	}
+
+	sendExtractionKey(m, "a")
+
+	assert.Nil(t, m.ex.extraction, "accept should work for deferred doc when LLM failed")
+
+	docs, err := m.store.ListDocuments(false)
+	require.NoError(t, err)
+	require.Len(t, docs, 1)
+	assert.Equal(t, "invoice.pdf", docs[0].FileName)
+}
+
+func TestAccept_DeferredDoc_WorksWithoutLLMStep(t *testing.T) {
+	t.Parallel()
+	m := newExtractionModel(t, map[extractionStep]stepStatus{
+		stepText:    stepDone,
+		stepExtract: stepDone,
+	})
+	ex := m.ex.extraction
+	ex.Done = true
+	ex.pendingText = "better ocr text"
+	ex.pendingDoc = &data.Document{
+		FileName: "scan.pdf",
+		MIMEType: "application/pdf",
+		Data:     []byte("pdf-bytes"),
+	}
+
+	sendExtractionKey(m, "a")
+
+	assert.Nil(t, m.ex.extraction, "accept should work for deferred doc without LLM")
+
+	docs, err := m.store.ListDocuments(false)
+	require.NoError(t, err)
+	require.Len(t, docs, 1)
+
+	full, err := m.store.GetDocument(docs[0].ID)
+	require.NoError(t, err)
+	assert.Equal(t, "better ocr text", full.ExtractedText)
+}