Methodology

How Trailblazer detects, scores, and surfaces emerging Solana ecosystem trends before they become obvious.

Pipeline Overview

Every fortnight, Trailblazer runs an automated pipeline that ingests live data from multiple sources, scores each protocol on momentum and novelty, clusters related signals into narratives, and generates actionable build ideas using AI. The entire process runs in under 5 minutes.

1
Ingest
Fetch live signals
2
Score
Momentum + novelty
3
Cluster
Group narratives
4
Label
AI-powered naming
5
Generate
Build ideas

Data Sources

Onchain Activity

via Helius RPC

Transaction volume per program (current vs baseline)

Unique wallet estimation via transaction sampling

New wallet share and retention heuristics

47 tracked programs

Developer Activity

via GitHub API

Commit velocity (current vs baseline period)

Star growth and fork counts

New contributor detection and release tracking

47 tracked repositories

Twitter / X KOL Signals

via public RSS proxies

Core team: Toly, Raj Gokal, Solana Labs, Foundation

Protocol founders: Mert, Meow, Cindy, Lucas

VCs and research: Multicoin, a16z, Paradigm, Messari

91 KOL accounts monitored

News and RSS Feeds

public RSS/Atom feeds

Solana blog, The Block, CoinTelegraph, CoinDesk

Decrypt, Blockworks, Messari

Protocol-specific mention matching

8 RSS feeds per cycle

Scoring Formula

Each protocol is scored across three dimensions, then combined with novelty and quality adjustments.

// Z-score: measures change from baseline
z(metric) = (current - baseline) / max(baseline, 1) // clamped [-5, 5]
// Momentum: weighted signal combination
momentum = 0.70 x onchain_z + 0.50 x dev_z + 0.35 x social_z
// Novelty: recency bonus
novelty = lerp(1.3, 1.0, age_days / 60)
// Quality penalties
single_wallet_spike = 0.6x
hype_only_social = 0.7x
totalScore = momentum x novelty x quality
70%
Onchain weight
50%
Dev activity
35%
Social weight

Narrative Clustering

Top-scoring candidates are grouped into narratives using agglomerative clustering with cosine similarity on text embeddings.

Text Embeddings
384-dimensional character trigram hashing — deterministic, no external API dependency
Average Linkage
Merge clusters above 0.45 cosine similarity threshold until max 10 clusters
AI Labeling
Kimi K2 analyzes each cluster to generate narrative title and summary

Saturation Analysis

Each build idea is checked against a corpus of 150+ existing Solana projects to estimate market saturation. Cosine similarity between the idea description and existing project descriptions determines the saturation level: low (unique opportunity), medium (some competition), or high (crowded space). This helps founders focus on underserved areas.

AI-Powered Analysis

Kimi K2 (Moonshot AI) powers three key analytical steps with structured JSON outputs validated by Zod schemas.

1
Narrative Labeling
Descriptive titles and one-paragraph summaries for each detected narrative cluster
2
Build Idea Generation
3-5 concrete product ideas per narrative with pitch, rationale, competitive landscape, and difficulty estimates
3
Action Pack Generation
Downloadable spec.md, tech.md, milestones.md, and deps.json for each idea

Tracked Protocols (47)

Jupiter
Jupiter Perpetuals
Drift Protocol
Raydium
Orca (Whirlpool)
Phoenix
Marginfi
Kamino Finance
Zeta Markets
Flash Trade
Marinade Finance
Jito
Sanctum
BlazeStake
Light Protocol
Squads Protocol
Clockwork
Switchboard
Pyth Network
Tensor
Metaplex
Wormhole
Dialect
Sphere Pay
Realms
Star Atlas
Pump.fun
Meteora
Helium
Parcl
Access Protocol
Helius
Backpack
Phantom
Magic Eden
PYUSD
Meta DAO
Genopets
Render Network
Hivemapper
Bonk
Dogwifhat
Moonshot
Robot AI
Ondo Finance
DFlow
Solana Mobile

Refresh Schedule

Reports are generated automatically on the 1st and 15th of each month via GitHub Actions cron. The pipeline can also be triggered manually via the admin API endpoint.