Checking a quotation (verify_quote)

Many sayings shared online as “the Buddha said” are not in the early texts. verify_quote checks a quotation against the texts themselves and reports what it finds. It is a library tool: it says what the searched texts contain, never who did or did not say something.

What it searches

The language is detected automatically, or set language to pali or english. Text with Pāli diacritics is Pāli; text with common English words (the, is, you, of…) is English. Only when neither is clear does detection look words up in the Pāli index, which builds that index (see below).

What it returns

• verdict: found, close_paraphrase or not_found_in_scope.
• matches: the best passages, each with its citation (text_id, citation, segment, VRI paragraph, PTS page for Pāli; translator, URL, attribution and licence for English) and a similarity from 0 to 1.
• scope: exactly which texts and translations were searched, and any page that could not be retrieved.
• equally_similar_texts, citation_note: present when the same wording matches several texts equally well, such as a stock opening (“Evaṃ me sutaṃ…”) or a verse repeated in another book. The note gives the number of texts (with equally_similar_at_least when there are more than were counted) and says that the matches listed are the first in canon order, not a unique source.
• next_step: for an English quotation that is not found, how to check the Pāli instead.

not_found_in_scope means only that the searched texts do not contain the words. For Pāli, the saying may come from a commentary, a later book or a modern author.

For English, it is not a finding about the saying at all. Many popular English wordings of real verses come from older or freer translations that are not searched. For example, a widely shared older wording of Dhp 5 (“Hatred does not cease by hatred…”) shares few words with the translations searched and is reported as not found. The same verse in Pāli is found. So for an English quotation that is not found, the result says this and points to the Pāli check: find the Pāli of the passage (with search_texts or get_text) and run verify_quote on it with language: "pali". The tests check this case.

How matching works

Text is lower-cased and diacritics and punctuation are removed (so manopubbaṅgamā and manopubbangama are the same). Candidate passages are found through a word index built the first time a Pāli quotation is checked (a few seconds; see memory below); spelling variants are matched through shared three-letter sequences (trigrams). Each candidate is then scored:

• similarity = Dice coefficient of character trigrams between the quotation and the passage, weighted by how many of the quotation’s distinctive words the passage contains. In Pāli, spaces are ignored (compounds are split differently by different writers) and the word weighting is gentler.
• found at 0.82 or more, close_paraphrase from 0.5, below that not_found_in_scope.
• Texts whose best passage scores the same as the best match (to two decimals) are counted as equally similar. Up to 200 equally strong candidates are compared, so a larger count is reported as “at least”.

Memory

The Pāli index is built once per server process, on the first Pāli check, and then kept. Measured on 1 November 2026 (Windows 11, Node.js 26): the build takes about 3.5 to 4 seconds. After garbage collection it holds about 30 MB of typed arrays plus about 16 MB of heap, and resident memory (RSS) stays roughly 50 to 80 MB higher than before. During the build, heap use grows by about 70 MB before it is collected, and resident memory can briefly be 100 MB or more above the starting point. These figures are approximate and vary between runs. English checks do not build the index unless the language has to be detected from a quotation with no common English words.

Tests

The tests (tests/verify.test.mjs) check 20 genuine quotations and 20 well-known misattributions. The genuine Pāli quotations are built from the bundled text (some without diacritics, one misspelt) and the English ones are sliced from the fixture pages, so no scripture is typed into the tests. The misattributions include “Holding on to anger is like grasping a hot coal…” and “Peace comes from within. Do not seek it without.”