Methodology

3-Layer Signal Architecture

SOLIS detects emerging Solana narratives by triangulating signals across three independent data layers, each with different lead times relative to market price action.

Layer 1: Leading (GitHub)

Developer activity that precedes market movement by 2-4 weeks. When devs start building, the market hasn't noticed yet.

  • - Star velocity on 200+ tracked Solana repos
  • - Commit frequency surges (z-score > 2.0)
  • - New repo clusters by topic
  • - Contributor count deltas
  • - Fork rate anomalies

Layer 2: Coincident (DeFi Llama + Helius)

Real-time capital and onchain activity. When TVL and transaction volume align with dev signals, the narrative is materializing.

  • - Solana chain TVL history and protocol-level changes
  • - DEX volumes and fees
  • - Stablecoin flows (net inflows/outflows)
  • - Program activity (transaction volume per program)

Layer 3: Confirming (CoinGecko)

Market price and volume data. When tokens in a narrative cluster start moving, the signal is confirmed but alpha is diminishing.

  • - Solana ecosystem token prices and volumes
  • - Category market cap movements
  • - Trending coins

Signal Stage Classification

StageLayers ActiveAlpha Potential
EARLYLayer 1 onlyHighest — devs building before market notices
EMERGINGLayer 1 + 2High — builders and capital moving together
GROWINGAll 3 layersModerate — gaining mainstream traction
MAINSTREAMAll 3, high confidenceLow — likely already priced in

Anomaly Detection

SOLIS uses z-score analysis to identify statistically significant deviations from baseline activity. A z-score above 2.0 (2 standard deviations) flags a metric as anomalous. No machine learning is involved — it's pure statistical math applied to time-series data.

Two-LLM Pattern

SOLIS uses two LLMs with different roles:

  • Claude (Agent SDK) — Orchestrates the pipeline, uses tools, handles errors. Runs the overall decision-making.
  • GLM-4.7 (via OpenRouter) — Handles bulk analysis: narrative clustering, idea generation, signal summarization. 10x cheaper than Claude for text analysis.

Total LLM cost per report: ~$2-5. Reports are generated fortnightly.

Data Sources

SourceAccessRate Limit
GitHub APIFree (PAT)5,000 req/hr
DeFi LlamaFree (no key)No hard limit
HeliusFree (1M credits/mo)~1M RPC calls/mo
CoinGeckoFree (10K/mo)~330 calls/day