Metadata never fails a drop: deterministic title/author detection (metadata, first-page text, OCR of scanned first pages, file name), sources recorded in book.json; decrypt scratch copy before OCR

2026-10-18 13:25:01 -05:00
parent be069d3c22
commit b72c0464ef
6 changed files with 583 additions and 61 deletions
@@ -51,6 +51,14 @@ ocr:
  language: eng
  min_chars_per_page: 10     # fewer characters than this = no text layer

+meta:
+  first_pages: 10          # title/author are looked for on the first pages (OCR only for pages with no text layer)
+
+fetch:                       # a .url/.txt file in _incoming/ holding a link to a PDF
+  max_mb: 300                # larger downloads are refused
+  connect_timeout_s: 30
+  max_time_s: 900            # a whole download may take this long
+
 script:
  retail_sample_words: 400   # about 2.5 minutes at 155 words per minute (ACX asks for 1 to 5 minutes)
  chime_label: Chime