topic-scorer

Algorithmic topic scoring engine — computes composite scores from freshness, search volume, social buzz, polarization heat, brand fit, and competition gap. Use when ranking topics for content priority or recalibrating scoring weights.

Model	Source
sonnet	pack: content-pumper

Full Reference

┏━ 🎯 topic-scorer ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ┃ Algorithmic engine for topic priority ranking ┃ ┃ — scores, thresholds, and weight recalibration ┃ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

topic-scorer

Computes composite scores for topics in content-topics.json. Reads raw signals, normalizes them to 0-100, applies weights, and writes scores back via topic-memory. Triggers status transitions when scores cross config.autoThreshold.

---

Scoring Formula

composite = Σ(normalized_signal × weight)

Weights come from content-topics.json → config.scoringWeights. Default weights:

Signal	Default Weight
freshness	0.20
searchVolume	0.20
socialBuzz	0.15
polarizationHeat	0.10
brandFit	0.20
competitionGap	0.15

Total weight must sum to 1.0. Validate before scoring — abort with error if sum != 1.0.

---

Signal Normalization

Every signal normalizes to 0-100 before weighting.

freshness

100 = discovered today
decay = 5 pts/day since firstSeen
normalized = max(0, 100 - (daysSinceDiscovered × 5))

Source: history.firstSeen in topic record.

searchVolume

normalized = min(100, log10(volume) × 20)

Volume < 10 → normalized = 0. Volume 100k+ → normalized capped at 100. Source: content-research skill or Google Trends via WebSearch.

socialBuzz

normalized = socialBuzz × 100

Raw value already 0-1. Source: trend-scanner signal output.

polarizationHeat

normalized = polarizationHeat × 100

Raw value already 0-1. Source: sentiment-mapper signal output.

brandFit

normalized = LLM assessment 0-100

Assess topic against brand.json fields: voice, values, verticals. High fit = topic aligns with brand verticals AND voice. Low fit = off-brand subject matter or tone mismatch.

competitionGap

normalized = max(0, 100 - (competitors × 10))

competitors = number of ranking articles found via "topic" site:competitor.com WebSearch. 10+ competitors → normalized = 0.

---

Data Sources

Signal	Source
freshness	history.firstSeen in content-topics.json
searchVolume	content-research skill → WebSearch Google Trends
socialBuzz	trend-scanner output → topic signals
polarizationHeat	sentiment-mapper output → topic signals
brandFit	LLM assessment vs brand.json
competitionGap	WebSearch competitor count

---

Batch Scoring Process

1. Read content-topics.json
2. Filter: topics where status != "archived"
3. For each topic:
   • Read raw signals from topic.signals
   • Normalize each signal (rules above)
   • Compute composite = Σ(normalized × weight)
   • Round to 2 decimal places
   • Call topic-memory update-score with result
4. After all scores updated, check threshold (see below)
5. Output leaderboard: top 10 topics sorted by score descending

---

Threshold Actions

After batch scoring, for each topic where status === "discovered":

if composite >= config.autoThreshold AND queued_count < config.maxQueueSize:
  → set status = "queued"
  → log: "topic queued — score {composite} crossed threshold {autoThreshold}"

queued_count = current count of topics with status === "queued". Never exceed maxQueueSize.

---

Weight Recalibration

Weekly process — correlates historical topic scores with article performance.

Input

• Topics with status in ["published", "monitoring"]
• history.performance[] entries from GA4 or GSC

Performance Signal

performance_score = normalize(sessions × 0.4 + impressions × 0.3 + clicks × 0.3)

Normalize across all published topics to 0-100.

Correlation Step

For each signal, compute Pearson correlation between:

• normalized_signal_at_queue_time (from score history)
• performance_score (from actual results)

Weight Update Rule

new_weight = current_weight × (1 + (correlation - avg_correlation) × 0.1)

Clamp each new weight to [0.05, 0.40]. Normalize all weights to sum to 1.0 after adjustment.

Output

Write updated config.scoringWeights to content-topics.json. Log before/after weights for audit.

Recalibration requires minimum 10 published topics with performance data. Skip if insufficient data — log reason.

---

Integration

Skill	Relationship
topic-memory	Writes scores and status via update-score operation
trend-scanner	Provides socialBuzz signal
sentiment-mapper	Provides polarizationHeat signal
content-research	Provides searchVolume data
content-pumper-pimp	Triggers batch scoring on schedule

---

Output Format

topic-scorer batch complete — 2026-03-01T14:00:00Z

Scored   ✓  42 topics
Queued   ✓  3 new (total: 7)
Skipped  ○  2 archived

Top 5:
  88.4  AI regulation debate         queued
  82.1  Medicare negotiation impact   queued
  79.3  EV tax credit changes         discovered
  71.0  Student loan forgiveness      monitoring
  65.5  Housing affordability index   discovered