How often does the internet say "I wish"?

The trail

1. 1 Corpus census — what is the denominator? 13 s

   SELECT relation, count() AS total FROM internet.text
   GROUP BY relation ORDER BY total DESC LIMIT 20

   50,745,831,089 documents across twelve relations.

2. 2 First corpus-wide attempt refused

   SELECT relation, count() AS docs_with_i_wish FROM internet.text
   WHERE hasAllTokens(search_text_lc, ['i','wish'])
     AND match(search_text_lc, '(^|[^a-z])i wish([^a-z]|$)')
   GROUP BY relation ORDER BY docs_with_i_wish DESC LIMIT 20

   Refused at admission: the runtime’s exposure ceiling rejected the unbudgeted full-corpus aggregate. The agent split it per relation, then returned for the one-shot with an explicit budget.

3. 3 Fast per-relation probes — full scans of the smaller surfaces 2.7 s 25.9M rows

   SELECT countIf(match(lower(payload), '(^|[^a-z])i wish([^a-z]|$)')) AS wish_docs,
          count() AS total
   FROM forums.posts LIMIT 10

   Hacker News: 1.1 s. Forums: 2.7 s. Bluesky (346M posts): 4.9 s. Mailing lists and Usenet (99.7M): 16.2 s. Academic full text: 329 s.

4. 4 Named-source fanout — who wishes most? 47 s

   SELECT source,
          countIf(hasAllTokens(search_text_lc, ['i','wish'])
              AND match(search_text_lc, '(^|[^a-z])i wish([^a-z]|$)')) AS wish,
          count() AS total
   FROM internet.documents
   GROUP BY source ORDER BY wish DESC LIMIT 100

   Same shape against forums.posts (1.2 s) and mailing_lists.messages grouped by newsgroup (2.5 s); newsletters isolated by URL filter.

5. 5 Occurrences, not documents — every utterance counted 15 m 46 s 50.5B rows

   SELECT relation,
          sum(countMatches(search_text_lc, '(^|[^a-z])i wish([^a-z]|$)')) AS occurrences
   FROM internet.text
   WHERE hasAllTokens(search_text_lc, ['i','wish'])
   GROUP BY relation ORDER BY occurrences DESC LIMIT 20

   109.1 million occurrences. 50.46 billion rows read in 15 m 46 s.

6. 6 Token denominators — how much text is that? 47 s

   SELECT source, sum(length(search_text_lc)) AS chars
   FROM internet.documents GROUP BY source ORDER BY chars DESC LIMIT 100

   Exact uncompressed column bytes per relation, character sums per source; tokens estimated at chars/4. Total ≈ 4.93 trillion tokens.

Documents containing "I wish", per relation

Token-normalized: occurrences per million tokens (rarest first)

Tokens estimated as characters / 4 (cl100k English prose average).

What it means

• Per document, long-form surfaces look wish-heavy simply because documents are long. Normalizing by tokens inverts the story: the wishing-est places per word actually said are fast conversational media — 4chan (72.8/M tokens) and Reddit comments (70.4).
• The wishing-est real community measured: misc.kids.pregnancy on Usenet, 50 occurrences per million tokens — one "I wish" every 20,000 tokens.
• Code and formal registers barely wish at all: GitHub 0.3/M, Metaculus question text 0.6/M, the linux-kernel list 1.3/M, academic papers 1.6/M.

Notes

• Counts are point-in-time; each relation has its own snapshot edge.
• Token counts are estimated as chars/4.
• Totals span the full corpus measured in the run, including relations not itemized in these tables.
• Occurrence totals (109.1M) exceed document totals (104.2M) because one document can wish repeatedly.

Measured 2026-08-06.