Methodology
TrendNexus exists to answer a harder question than “what is trending?” — what is rising, why, whether it can be trusted, and what can be built around it. That only works if every number can defend itself. This page is the defense.
Principles
- Explainability over mystery
- Every score exposes its raw inputs, normalized values, weights, calculation date, confidence and missing-data penalties. If we can't show the work, we don't ship the number.
- Signal over vanity
- Stars are an input, not the product. Acceleration, sustained activity, contributor health, responsiveness and ecosystem adoption carry the weight.
- Freshness is visible
- Every analytical surface shows when it was last synced, and whether a value is exact, estimated or inferred. In the current phase, velocity inputs are deterministic estimates — labeled est. wherever they appear — until the full ingestion pipeline replaces them.
- No silent zeroes
- A missing metric never quietly becomes zero. Scores carry confidence, coverage and an explicit missingMetrics list instead.
Score definitions
Momentum Score
How much attention is this gaining right now — and is it accelerating?
Each input is normalized by repository age, size, topic baseline and anomaly likelihood. A repository gaining 500 stars from a 1,000-star base is not treated like one gaining 500 from 200,000.
Project Vitality Index
Will this project still be maintained when you depend on it?
Bands: 90–100 Exceptional · 80–89 Strong · 65–79 Healthy · 50–64 Watch · 35–49 Fragile · 0–34 High risk.
Ecosystem Influence
How much of the ecosystem leans on this project?
Package-registry ingestion arrives in a later phase; until then the input is estimated and the score's coverage reflects it.
Opportunity Score
How much commercial surface area is forming around this trend?
Every opportunity card shows the evidence behind each factor.
Predictions are archived, then graded
The breakout model is deliberately interpretable: star and fork acceleration, contributor growth, mention velocity, release timing and topic momentum, with anomaly penalties. Every forecast is stored with its inputs and scored honestly after the window closes — precision, recall, Brier score and calibration will be published as the archive matures. Until then, prediction confidence is labeled experimental.
What the Security & Maintenance score is not
It is a risk indicator, built from observable maintenance signals — release recency, license clarity, maintenance status, dependency staleness and bus factor. It is not a security audit, and TrendNexus never assigns security certification through automated analysis alone.
Current model version: tn-interpretable-1.0.0 · formulas versioned in source · corrections welcome via GitHub issues.