fix: ocr_done state triggers review load after OCR completes; add Gruffalo word corrections

smoke_signal/scripts/punct_corrector.py

@@ -38,10 +38,16 @@ KNOWN_SUBSTITUTIONS: dict[str, str] = {
 # I misread as T at word start; G misread as s.
 # Extend this as more errors are found in review.
 WORD_CORRECTIONS: dict[str, str] = {
+    # I misread as T at word start
     "Tndeed": "Indeed",
     "Tll":    "I'll",
     "Tt":     "It",
     "Ti":     "It",
+    "Tn":     "In",
+    "Ts":     "Is",
+    # G misread as s at word start
+    "sruffalo": "Gruffalo",
+    "sruffalo's": "Gruffalo's",
     "sood":   "good",
     "sreat":  "great",
     "sot":    "got",
@@ -49,6 +55,7 @@ WORD_CORRECTIONS: dict[str, str] = {
     "srew":   "grew",
     "save":   "gave",
     "soing":  "going",
+    "so":     "Go",   # only catches capitalised — won't corrupt "so" mid sentence
 }
 
 MIN_PUNCT_PER_10_WORDS = 0.8

smoke_signal_tab.py

@@ -1946,7 +1946,7 @@ def run_ocr(page_selection: str = "", page_exclusion: str = "", replace_book_que
             )
         )
 
-    yield _ocr_status_html(), "\n".join(log)  # stream: models loaded
+    yield _ocr_status_html(), "\n".join(log), 0  # stream: models loaded
     punct_mod = _load_punct_module()
     if punct_mod is None:
         log.append(log_line(f"⚠ Punctuation corrector unavailable ({_PUNCT_MODULE_ERROR or 'unknown'})"))
@@ -1973,7 +1973,7 @@ def run_ocr(page_selection: str = "", page_exclusion: str = "", replace_book_que
     for _, row in eligible.iterrows():
         book_id = row["book_id"]
         log.append(log_line(f"▶ {book_id}: starting OCR"))
-        yield _ocr_status_html(), "\n".join(log)  # stream: book start
+        yield _ocr_status_html(), "\n".join(log), 0  # stream: book start
         profile_path = PROFILES_DIR / f"{book_id}_page_profile.json"
         if not profile_path.exists():
             log.append(log_line(f"✗ {book_id}: no profile"))
@@ -2239,10 +2239,10 @@ def run_ocr(page_selection: str = "", page_exclusion: str = "", replace_book_que
                 # Stream progress after each batch
                 done_pages = min(start + batch_size, len(ocr_targets))
                 log.append(log_line(f"  ✓ {book_id}: batch {start//batch_size + 1} complete ({done_pages}/{len(ocr_targets)} pages)"))
-                yield _ocr_status_html(), "\n".join(log)
+                yield _ocr_status_html(), "\n".join(log), 0
 
         log.append(log_line(f"  → {book_id}: batch OCR complete — building review queue"))
-        yield _ocr_status_html(), "\n".join(log)  # stream: batch OCR done
+        yield _ocr_status_html(), "\n".join(log), 0  # stream: batch OCR done
         # Second pass: build page-level OCR output and review queue.
         for page_data in profile.get("pages", []):
             page_num = page_data["page_number"]
@@ -2371,7 +2371,7 @@ def run_ocr(page_selection: str = "", page_exclusion: str = "", replace_book_que
         df.loc[df["book_id"] == book_id, "status"] = "ocred"
         total_review = len(review_pages) + len(quarantine_pages)
         log.append(log_line(f"✓ {book_id}: {len(ocr_pages)} selected pp — review queue: {total_review} · queue write complete"))
-        yield _ocr_status_html(), "\n".join(log)  # stream: book done
+        yield _ocr_status_html(), "\n".join(log), 0  # stream: book done
 
     qdf_new = pd.DataFrame(queue_rows)
     if QUEUE_CSV.exists():
@@ -2395,7 +2395,7 @@ def run_ocr(page_selection: str = "", page_exclusion: str = "", replace_book_que
         QUEUE_CSV.unlink()
 
     save_manifest_df(df)
-    yield _ocr_status_html(), "\n".join(log)  # stream: final
+    yield _ocr_status_html(), "\n".join(log), 1  # stream: final — triggers review load
 
 
 def _ocr_status_html() -> str:
@@ -2874,6 +2874,7 @@ def smoke_signal_tab():
                 </div>""")
 
                 ocr_status = gr.HTML(_ocr_status_html())
+                ocr_done = gr.State(0)  # increments when OCR completes — triggers review load
 
                 gr.HTML("""<div class="ss-card">
                     <div class="ss-card-title">Self-Improvement Loop</div>
@@ -2911,7 +2912,7 @@ def smoke_signal_tab():
                 ocr_run_event = ocr_btn.click(
                     run_ocr,
                     inputs=[ocr_page_selection, ocr_page_exclusion, ocr_replace_queue],
-                    outputs=[ocr_status, ocr_log],
+                    outputs=[ocr_status, ocr_log, ocr_done],
                     show_progress="minimal",
                     trigger_mode="once",
                 )
@@ -3035,11 +3036,8 @@ def smoke_signal_tab():
                     outputs=review_load_outputs
                 )
                 review_tab.select(load_review, outputs=review_load_outputs)
-                ocr_run_event.then(
-                    load_review,
-                    outputs=review_load_outputs,
-                    show_progress="hidden",
-                )
+                ocr_done.change(load_review, outputs=review_load_outputs)
+
                 accept_btn.click(do_accept, inputs=[current_idx, final_text_box, reviewer_name, reason_code], outputs=action_outputs)
                 edit_btn.click(do_accept, inputs=[current_idx, final_text_box, reviewer_name, reason_code], outputs=action_outputs)
                 reject_btn.click(do_reject, inputs=[current_idx, final_text_box, reviewer_name, reason_code], outputs=action_outputs)