How we choose what to feature
We look for clear editorial value: distinctive voices, strong subject matter, thoughtful production, and a sense that the show offers something specific to its audience.
We do not rely on popularity alone. A show can be small, new, or niche and still be an excellent recommendation if its craft or perspective stands out.
Signal Score
What is Signal Score?
Signal Score ranks podcasts by meaningful discovery behavior, not by who is already biggest. It blends scout approvals, saves, shares, repeat engagement, and editorial review into one weighted score.
Signal Score
hourlyPrimary ranking score for discovery surfaces and show cards.
0.20*Scout Approvals + 0.14*Campfire Saves + 0.10*Shares + 0.12*Discovery Activity + 0.10*Repeat Engagement + 0.08*Signal Velocity + 0.06*Freshness + 0.20*Editorial Review
Distinct actor counts, per-actor weight caps, saturating normalization, and negative-feedback penalty prevent brute-force boosting.
Signal Strength
hourlyMeasures current community pull from high-intent actions.
0.45*Scout Approvals + 0.30*Campfire Saves + 0.25*Shares, each normalized with log-style saturation.
Only distinct users/sessions count; action totals are capped before normalization.
Signal Velocity
hourlyDetects acceleration by comparing 7-day intent rate against the 28-day baseline.
50 + 30*(signal_rate_7d / max(signal_rate_28d, floor) - 1) + 20*normalized recent approvals.
Uses rate deltas, minimum baseline floors, and capped recent-volume contribution.
Trail Momentum
dailyRanks how strongly a show is moving within its category lane.
0.55*Signal Velocity + 0.45*category_percentile(Signal Strength).
Category-relative percentile dampens raw-volume advantages from broad categories.
Frontier Index
dailyRewards promising shows still early in their exposure curve.
0.5*Signal Velocity + 0.3*Freshness + 0.2*(100 - audience saturation proxy).
Exposure proxy is capped and cannot increase only from impressions without matching intent.
Discovery Rank
hourlyStable ordering field used for lists and recommendation tie-breaks.
Local helper estimate: 0.65*Signal Score + 0.20*Trail Momentum + 0.15*Frontier Index. Canonical production rank is computed in SQL snapshots.
Derived from normalized score distribution rather than raw event counts.
Community Confidence
dailyEstimates trust that the positive signal is real and not noisy.
Wilson-style lower bound over positive actions vs negative feedback, scaled 0-100.
Needs both action volume and positive ratio; low-volume spikes stay conservative.
Beacon Rating
dailyEditorial + community blended signal for featured placements and badges.
0.6*Community Confidence + 0.25*Editorial Review + 0.15*Signal Strength.
Editorial review is capped, so curator interest cannot fully override weak community response.
Pathfinder Score
weeklyHighlights breakout shows that trusted scouts found before broader adoption.
0.5*Frontier Index + 0.3*Scout Confidence + 0.2*Trail Momentum.
Leans on scout confidence and frontier conditions, not just raw interaction volume.
Scout Confidence
weeklyRates how much weight to give scout actions based on consistency and action quality.
0.35*action quality + 0.25*consistency + 0.2*category expertise + 0.2*integrity.
Confidence only rises with sustained behavior quality over time; low-history scouts stay near baseline.
Ranking notes
- High-intent actions count more than passive traffic.
- Trusted scouts carry more weight than brand-new accounts.
- Recency can matter for discovery, but it is not treated as a proxy for quality.
- Per-user caps and distinct-actor rules limit gaming.
Signal Score is directional, not absolute. New shows can move quickly, established shows can cool off, and no score is paid placement.
What we value
- Clear focus and a compelling point of view.
- Good production quality and a reliable listening experience.
- Useful context for listeners who want to discover something new.
- Shows that feel worth recommending even when they are not the loudest voice online.
The goal
The aim is simple: help listeners find podcasts that feel worth their time. That means balancing discovery, quality, and clarity in every recommendation we publish.