Count a term on Hacker News by month, sparkline included, with SQL

Answer shape

One GROUP BY over hackernews.items returns the monthly count of the token 'mcp' since January 2025, and bar() renders each month as a bar, so the shape of the series is visible in the rows themselves. Measured 2026-09-09 through the public door: 0.76 s, 6,027,149 rows read, 20 rows returned, $0.0030 of machine time, $0 charged.

Sources

• Hacker News

Methods

• source-native SQL
• token-indexed lexical search
• time series
• in-query rendering

SQL

SELECT toStartOfMonth(original_timestamp) AS m, count() AS n,
       bar(n, 0, 2200, 20) AS spark
FROM hackernews.items
WHERE original_timestamp >= '2025-01-01'
  AND original_timestamp < toStartOfMonth(today())
  AND hasToken(search_text_lc, 'mcp')
GROUP BY m ORDER BY m LIMIT 100;

Output, no account

scry.io/s/dikjdnv0 holds this query's rows and its envelope — elapsed, rows read, machine cost — frozen from a run through the public door.

Verification notes

• NotehasToken(search_text_lc, 'mcp') is a whole-token match over the full-text index: it counts the token, not the sense.
• NoteThe current month is excluded (original_timestamp < toStartOfMonth(today())), so every bar covers a complete month.
• NoteRows land every few minutes, so a rerun differs from the frozen output by the day; the difference is freshness, not an error.
• NoteFrozen output and envelope, no account: https://scry.io/s/dikjdnv0 (2026-09-09: 0.76 s, 6,027,149 rows read, $0.0030 machine cost, $0 charged).

Related sources