Split Multi-Promise Text into Separate Unique Promise Records
Problem
The current AI extraction logic combines multiple distinct promises into a single output record when they appear within the same paragraph or compound sentence.
This prevents accurate tracking of individual promises (e.g., a university project, TVET construction, and an internship programme being grouped as one record).
Goal
- Ensure the AI outputs one unique record per actionable promise, even when multiple promises appear in the same sentence, paragraph, or category.
- Each promise should be independently trackable in the system.
- Expected Behaviour
- Each promise must be extracted as a separate, standalone record.
- If a sentence or paragraph contains multiple commitments, they must be split into distinct outputs.
- Compound statements should be broken down so that each primary action (e.g., Build, Establish, Set up, Launch, Construct, Recruit, Allocate, Implement) results in a new promise record.
- No merging of multiple deliverables into a single field.
- No aggregation of distinct commitments.
Acceptance Criteria
- Multiple promises within a single paragraph are split into separate records.
- Each record contains only one actionable promise.
- No compound deliverables exist within a single output field.
- Existing single-promise behaviour remains unchanged.
Suggested additional rules to the existing AI prompt logic
- Extract one record per actionable promise.
- If a sentence or paragraph contains multiple commitments, split them into separate records.
- Each primary action verb (e.g., Build, Establish, Set up, Launch, Construct, Recruit, Allocate, Implement) must produce a new record.
- Do not combine multiple projects or deliverables into one output.
- Do not summarise or aggregate promises.
Link to the example issue on Check Media
https://checkmedia.org/pesacheck-promise-tracker/media/3439740?listQuery=%7B%22sort%22%3A%22recent_activity%22%7D&listIndex=3&tab=articles

Split Multi-Promise Text into Separate Unique Promise Records
Problem
The current AI extraction logic combines multiple distinct promises into a single output record when they appear within the same paragraph or compound sentence.
This prevents accurate tracking of individual promises (e.g., a university project, TVET construction, and an internship programme being grouped as one record).
Goal
Acceptance Criteria
Suggested additional rules to the existing AI prompt logic
Link to the example issue on Check Media
https://checkmedia.org/pesacheck-promise-tracker/media/3439740?listQuery=%7B%22sort%22%3A%22recent_activity%22%7D&listIndex=3&tab=articles