Crypto Twitter is useful only when you can separate a durable market narrative from a one-hour engagement spike. A token can trend hard on CT and still fail if the claim came from weak accounts, died after one refresh cycle, or immediately attracted a credible opposing cluster.

The Solana Twitter Token Trading Agent fetches recent tweets for tracked Solana symbols, models author credibility and source diversity, and then asks a Claude agent to decide whether each narrative is durable, contested, or fading. The output is a ranked board of narrative signals with action hints and explicit durability context. It is intentionally skeptical of narratives that look large only because the same claim is being echoed in one cluster.

FETCH -> AGGREGATE -> SCORE DURABILITY -> FLAG CONTESTED -> RANK

Research Board • Durability Model • At a Glance • Operating Surfaces • How It Works • Example Output • Technical Spec • Risk Controls • Quick Start

• Use case: Solana narrative ranking across Crypto Twitter windows
• Primary input: credibility, source diversity, persistence, contradiction
• Primary failure mode: mistaking hype loops for durable narratives
• Best for: operators who want to know which stories last beyond the first engagement spike

Live research board: tracked narratives, durability scores, credibility markers, contradiction flags, and action hints ranked by whether the story is actually holding.

How the agent ranks a narrative: raw mentions are grouped into claim clusters, scored for source quality and persistence, checked for contradictions, and promoted only when the claim survives beyond the first engagement spike.

• Research Board: shows which narratives are emerging, confirmed, contested, or decaying
• Durability Model: explains how source quality, persistence, and contradiction checks turn noisy mentions into ranked narrative signals
• Contradiction Lens: makes opposing evidence visible before a narrative gets promoted
• Ranking Output: promotes durable, confirmed, or contested claims with action context

The problem with narrative tracking is that engagement is easy to see and hard to interpret. A claim can look huge on CT and still be completely fragile once you inspect who is repeating it and whether any credible counter-argument is growing at the same time.

This agent is built to make those hidden differences explicit. It is less interested in who is shouting loudest and more interested in whether the same claim survives across time, sources, and pushback.

The agent uses a disciplined narrative loop:

1. fetch fresh mentions for the tracked Solana tokens
2. cluster similar claims into one narrative lane
3. score the lane on credibility, diversity, persistence, and contradiction
4. downgrade claims that are viral but narrow, repetitive, or heavily contested
5. rank the remaining narratives into a board the operator can act on

The point is to identify what lasts, not what spikes for one refresh cycle.

• repeated by more than one quality source cluster
• still alive across multiple windows
• not immediately met by a strong contradictory lane
• supported by evidence instead of only mood

That usually matters more than raw engagement totals.

SOLANA TWITTER TOKEN TRADING AGENT // NARRATIVE BOARD

lead narrative     SOL fee compression
durability         0.71
credibility        0.78
source diversity   0.68
contradiction      low
state              confirmed

operator note: claim is persisting across multiple desks without strong pushback

The agent does not equate engagement with quality. Each token narrative is scored on four dimensions:

Durability = 0.35 * credibility + 0.30 * source_diversity + 0.20 * persistence - 0.15 * contradiction_ratio

Where:

• credibility blends author reach with observed engagement quality
• source_diversity rewards narratives repeated across different accounts and reply chains
• persistence increases when the same claim survives multiple windows instead of one spike
• contradiction_ratio rises when bearish and bullish evidence clusters stay active together

Operational rules:

• viral posts with weak persistence stay in hype, not bullish
• narratives above the contradiction threshold are marked contested
• score ordering is durability-adjusted, not raw engagement-only
• action hints should cite whether the edge comes from persistence, credibility, or a decaying claim

• credibility weighting: prevents weak accounts from overpowering the board through raw volume
• contradiction penalty: stops contested narratives from being ranked like clean ones
• persistence requirement: downgrades stories that only live for one short window
• cluster skepticism: filters cases where one social pocket keeps repeating itself

This is designed to be skeptical because hype is abundant and durable narratives are rare.

Twitter fetch
  -> mention aggregation
  -> durability + contradiction scoring
  -> Claude narrative review
  -> ranked signal board
  -> Redis-backed state (optional)

git clone https://github.com/tradinglabpremium/solana-twitter-token-trading-agent.git
cd solana-twitter-token-trading-agent && bun install
cp .env.example .env
bun run dev

ANTHROPIC_API_KEY=sk-ant-...
TWITTER_BEARER_TOKEN=...
REDIS_URL=redis://localhost:6379
AUTHOR_CREDIBILITY_WEIGHT=0.35
SOURCE_DIVERSITY_WEIGHT=0.30
PERSISTENCE_WEIGHT=0.20
CONTRADICTION_PENALTY_WEIGHT=0.15
NARRATIVE_HALF_LIFE_MINUTES=180

• Planned commit sequence: docs/commit-sequence.md
• Draft engineering issues: docs/issue-drafts.md

• Runbook
• Changelog
• Contributing
• Security

MIT

read what lasts, not what trends for ten minutes.