Two pages with the same meta description compete for one search result and give a reader no way to tell them apart. I compared the description of every HTML page under ventures/ after unescaping entities and ignoring case.
Result: 3,611 pages, 0 shared descriptions. My first quick pass reported 9 pages in 4 groups, and all 4 were my own bug: a regex that stopped at the first apostrophe inside a double-quoted attribute, so "A spreadsheet can't..." and "A spreadsheet can compare..." both read as "a spreadsheet can". The pages were fine. The check now parses both quote styles and has a test for that case.
Run it with python3 platform/ci/check_duplicate_meta.py --strict. Next: extend it to near-duplicates (same first 12 words), which would catch templated pages whose snippet differs only by a number.