Story Selection

Evaluate and rank candidate stories for newsletter inclusion using weighted editorial criteria.

Overview

Not every story from a show makes the newsletter. Story selection applies a structured evaluation framework to pick the strongest 3-5 stories from a larger candidate pool while ensuring topic diversity. It supports two modes: auto-select (algorithm picks) and explicit (editor picks, validated against criteria).

Core Value: Consistent, defensible story selection with documented rationale — eliminating "gut feel" curation that can't be reviewed or improved.

Core Capabilities

• 5-Criterion Evaluation — Score each story on audience relevance, newsworthiness, topic diversity, engagement potential, and narrative strength
• Weighted Scoring — Configurable weights per criterion with sensible defaults
• Diversity Constraint — Prevent same-topic clustering (no more than 2 stories from the same domain)
• Auto-Select Mode — Rank all candidates, pick top N with diversity constraint applied
• Explicit Mode — Validate editor pre-selections against criteria, flag concerns
• Selection Rationale — Every pick and every exclusion includes a documented reason

Quick Start

1. Gather candidate stories (from script-to-article segmentation or other sources)
2. Choose mode: auto-select (pick best N) or explicit (validate pre-selections)
3. Apply the 5-criterion evaluation to each candidate
4. In auto-select: rank and pick top N with diversity constraint
5. In explicit: validate pre-selections and flag any concerns

Key Workflows

1. Auto-Select Mode

1. Receive candidate stories — Each with headline, summary, and body
2. Score each story on 5 criteria (1-10 scale per criterion):
   • Audience relevance (weight: 30%) — Does the target audience care about this topic?
   • Newsworthiness (weight: 25%) — Is this timely? Does it contain new information?
   • Topic diversity (weight: 15%) — Does this add a different topic to the edition?
   • Engagement potential (weight: 15%) — Will readers share, discuss, or act on this?
   • Narrative strength (weight: 15%) — Is the story well-told with a clear arc?
3. Calculate weighted score for each story
4. Rank by weighted score
5. Apply diversity constraint — If top N has >2 stories from the same domain, swap the lowest-scoring duplicate for the next-highest story from a different domain
6. Output: Selected stories (ordered) + excluded stories, each with scores and rationale

2. Explicit Mode (Editor Picks)

1. Receive pre-selected stories — Editor has already chosen
2. Score all candidates (selected and unselected) on 5 criteria
3. Compare — Are the pre-selections the top N by score? If not, flag the gap
4. Check diversity — Do the pre-selections cluster on one topic?
5. Output: Validation report — confirms good selections or flags concerns with data

Best Practices

• Score before reading the editor's picks (in explicit mode) to avoid anchoring bias
• When two stories score within 1 point of each other, prefer topic diversity
• Document the "almost made it" stories — they're candidates for next edition
• Review scoring weights quarterly based on reader engagement data

Integration

Consumed by newsletter-producer agent as step 2 of the 7-step pipeline. Receives input from script-to-article segmentation output.