Prediction ledger
Extract forecasts with their dates, confidence, and checks.
Extract the predictions actually present in the supplied content. First list each distinct prediction as a bullet of at most 16 words. Then create a Markdown table with these columns: Prediction, Confidence, Date, and How to verify. Keep the predicted event, scope, timeframe, and conditions accurate. Report confidence only when the source states it. Preserve dates and quantities exactly. Describe an objective verification test when the source supplies one; label a test you propose as "Suggested." Use "Not stated" for missing dates or confidence. Do not fabricate forecasts from missing content or convert a possibility into a promise. Remove duplicates. Output the list and table without unrelated commentary.
Example
Forecast extraction with a conditional target and missing confidence
Before 85 words
The report forecasts that weekly deliveries in the pilot region will exceed 1,000 by 2027-06-30, provided the second warehouse opens on schedule. The authors give this forecast 70% confidence and say to verify it against the weekly operations report. They also predict that pilot-region cold-storage costs will fall after the second warehouse opens, but give no amount, deadline, confidence, or verification method for that prediction. The report makes no forecasts for other regions. These are forecasts rather than commitments; neither outcome has been observed yet.
After 81 words
- Pilot-region deliveries: >1,000/week by 2027-06-30, if the second warehouse opens on schedule. - Pilot-region cold-storage costs fall after the second warehouse opens.
| Prediction | Confidence | Date | How to verify |
|---|---|---|---|
| Pilot-region deliveries >1,000/week if warehouse opens on schedule | 70% | 2027-06-30 | Weekly operations report |
| Pilot-region cold-storage costs fall after warehouse opens | Not stated | Not stated | Suggested: Compare pre/post-opening costs. |
Examples illustrate the method. They do not measure model output.
Files and sources
- prompt.txt Complete prompt as plain text.
- SKILL.md The same prompt with a skill header.
- llms.txt (JSON) Prompt, metadata, example, and sources.
- LICENSE MIT license text.
Adapted from Fabric · extract_predictions. Source notes.