Trial to paid conversion, weeks of 18 and 25 August 2026
Produced by: Run EXAMPLE-DQ-01 · one read-only query · answer card on ACME-318 · demo values
- Answer
- 38 of the 412 trials that started in the week of 25 August converted to paid, which is 9.2%. The week before was 31 of 389, or 8% — up 1.2 points. Neither percentage was written by a model: the reply carries the four whole numbers and nothing else, and there is no field in it for a rate.
- What was counted
- 3 counting rules, in the words a person can disagree with rather than in SQL. A conversion is the first paid subscription for that account beginning within 14 days of the trial start. An account that started two trials in the window counts twice, once per trial. This is the section that turns "I do not trust that number" into "I would have counted it differently", which is a conversation a team can finish.
- The query that ran
- The exact query, in a code block, with its 4 bound values listed in order and the 2 rows the warehouse returned. Every value travels as a parameter — even the period unit and the conversion window — because the platform refuses a value written into the query text. It refuses anything that is not one SELECT, and the connection is read-only.
- Trend
- One point per period, each with its two counts and the rate computed from them, drawn as a chart in the document itself. The thread reply gets the number; the card gets the chart, because a chat message cannot render one. A figure resting on fewer than 5 rows is never drawn or reported at all.
- Read this before you quote it
- 3 caveats, and the first is the one that matters: analytics.trial starts on 2026-03-01, so a question about anything earlier has no denominator here. A result that hit the row cap is marked truncated and is never presented as a total — the run narrows the query and asks again instead.
- Where the definition came from
- Where each counting rule came from, with its reference. A rule the run had to settle itself is proposed as a workspace memory claim, cited and left unreviewed until somebody approves it — never written in as settled fact, because a definition nobody agreed to is how the next argument starts.