Use when you want a number of your own that assistants can attribute to you.
original-statistic-asset-plan.md
You are planning original data that others could cite. You are not producing the numbers and you must not estimate them.
What our data covers, with row counts and date range: {{DATA_WE_HOLD}}
Questions our audience asks that a number would settle: {{AUDIENCE_QUESTIONS}}
Topic: {{TOPIC}}
Privacy, contractual and commercial limits on what we can publish: {{PUBLISHING_LIMITS}}
Produce:
1. Up to ten candidate figures, ranked by how likely each is to be quoted. For each: the question from {{AUDIENCE_QUESTIONS}} it settles, the field in {{DATA_WE_HOLD}} it comes from, the sample and period needed, and why someone writing about {{TOPIC}} would cite us rather than a public source.
2. A table: | Candidate | Sample adequacy (adequate / thin / cannot tell) | Confound a critic would raise first | Blocked by {{PUBLISHING_LIMITS}} | Method note required |
3. For the top three, a sentence template stating the figure, the sample, the period and the source, leaving the number itself blank.
4. What must be published alongside the number for it to survive scrutiny.
Constraints:
- Produce no figures, ranges or illustrative examples. Every number position stays blank.
- Reject candidates whose sample cannot carry a general claim. Say why rather than softening the wording.
- Skip anything that restates a public statistic; that citation goes to the public source.
Replace each placeholder with your own detail. The more specific you are, the less the model invents.
When is planning a statistic the right move?
When you want something cited that a competitor cannot copy from the same public source. It plans the study rather than reporting it, so run it before anyone queries the database. If the number already exists in a public dataset, the citation goes to that dataset and this is the wrong prompt.
What do I need before running it?
An honest description of the data you hold, with row counts and a date range, the questions your audience asks, and the privacy, contractual and commercial limits on publishing. Row counts drive the sample adequacy verdict, so a vague claim that you have lots of data returns cannot tell throughout.
What comes back, and which column is worth the most?
Up to ten candidate figures ranked by likelihood of being quoted, a table covering sample adequacy, the first confound a critic would raise and the method note required, then sentence templates with the number left blank. The confound column is the useful bit, since that is the objection arriving after publication.
What is the mistake that costs me?
Filling in a template blank with an estimate to see how the sentence reads. Every number position is left empty on purpose, and an illustrative figure has a way of surviving into the published version. Publish the sample and period alongside the number, or it gets quoted once and challenged thereafter.