How this dashboard was generated
Every number above came from a real query, run live against a real ClickHouse
server (a fresh, isolated clickhouse-server:24-alpine container, seeded with
analytics/'s actual structure.sql schema and synthetic
but realistic rows) — via ekos clickhouse ask (RFC 0056): a local Ollama
model (llama3:latest, no cloud API call) builds SQL grounded in EKOS's compiled
schema, ClickHouse itself validates and executes it, and every successful query is recorded as an
Event/Evidence pair in the ledger. Three of the real questions asked, unedited:
"how many total pageview events does site_id 101 have in events_v2 between 2026‑08‑01 00:00:00 and 2026‑08‑15 23:59:59?"
SELECT COUNT(*) FROM plausible_events_db.events_v2
WHERE site_id = 101 AND timestamp >= '2026-08-01 00:00:00'
AND timestamp <= '2026-08-15 23:59:59' LIMIT 1000
→ 56
"for site_id 101 in sessions_v2, group by referrer_source and count sessions, only where utm_medium is social or email, order by count descending"
SELECT referrer_source, COUNT(*) AS session_count
FROM plausible_events_db.sessions_v2
WHERE site_id = 101 AND utm_medium IN ('social', 'email')
GROUP BY referrer_source ORDER BY session_count DESC LIMIT 1000
→ Twitter 17, (email) 5, LinkedIn 4
"for site_id 101 in events_v2 where name is pageview, group by pathname and count distinct user_id, order by that count descending, limit 10"
SELECT pathname, COUNT(DISTINCT user_id) AS count
FROM plausible_events_db.events_v2
WHERE site_id = 101 AND name = 'pageview'
GROUP BY pathname ORDER BY count DESC LIMIT 10
→ / 14, /docs 10, /pricing 10, /contact 10, …
Two real, minor generation quirks, corrected before display rather than shipped:
the pageview-count question above got a same-day date boundary right (timestamp <=
'2026-08-15 23:59:59'), but an earlier, differently-worded visitor-count question compared a
datetime column against a bare date literal and silently dropped same-day results (25 instead of 26).
The sources query shown above also has no date filter at all — that's not a bug, the question
asked didn't request one, so it summed both the current and comparison periods together (17/5/4 instead
of 13/4/2). The stat-card totals, deltas, and both breakdown tables shown above use date-scoped,
boundary-corrected versions of these same queries, run directly against the same live instance, same
data, same schema. Every value on this page is real; none is copied from a reference screenshot.