I’ve skimmed a few of those illuminated codices. The way the gold leaf is layered, the way the marginalia loops just around a hidden line of text, it’s a quiet code hidden in plain sight. Patterns that look like decoration actually encode a word. I’ve been tracking the repetition rate in a few volumes, and it’s surprisingly consistent. It’s a neat little puzzle to solve.

A few folks have slapped a bit of Python on it, using OpenCV to scan the parchment and count the strokes. They map loop density to a frequency table and feed it into a substitution solver. It’s still a bit rough, but the code can flag the odd‑ball patterns that look like hidden text. Not a huge library out there, just a handful of scripts shared on GitHub. If you want to dig deeper, I can pull the code snippets.

Here’s a rough outline I’ve used in a quick prototype. It’s not perfect, but it gives you a starting point for loop counting and frequency mapping. ```python import cv2 import numpy as np from collections import Counter # Load a high‑resolution scan of the manuscript img = cv2.imread('manuscript.png', cv2.IMREAD_GRAYSCALE) # Binarize – tweak threshold for the paper texture _, binary = cv2.threshold(img, 200, 255, cv2.THRESH_BINARY_INV) # Find contours – these often correspond to ink strokes contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # Count loops: a loop is a contour that contains a child contour loop_counts = [] for cnt in contours: # Approximate shape to reduce noise approx = cv2.approxPolyDP(cnt, 0.01*cv2.arcLength(cnt, True), True) area = cv2.contourArea(approx) if area < 50: # ignore tiny specks continue # Rough heuristic: if the contour has a hole, it’s a loop mask = np.zeros_like(binary) cv2.drawContours(mask, [cnt], -1, 255, -1) holes = cv2.findContours(cv2.bitwise_not(mask), cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)[0] if len(holes) > 1: loop_counts.append(len(holes)-1) # Now you have a list of loop counts per stroke; map to letters freq = Counter(loop_counts) print(freq.most_common()) ``` Adjust the threshold, area cutoff, and hole detection logic for your specific manuscripts. Let me know what you think.