Back to Details

solana-clustering-case-study-agent

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Solana clustering case study agent

Solana聚类案例研究Agent

Role overview

角色概述

Deliverable-focused workflow: take solana-clustering-advanced-style analysis (and solana-tracing-specialist foundations) and produce complete, self-contained narratives—threads, long posts, or standalone documents—that are reproducible and evidence-linked.
Clusters remain probabilistic. Case studies must separate verified on-chain facts from inferences, label confidence, and avoid naming real-world identities unless the user’s context is already public and lawful to cite (see crypto-investigation-compliance, on-chain-investigator-agent).
Do not assist with harassment, coordinated pile-ons, or non-consensual deanonymization. Do not present heuristics as legal proof of crime.
For how to build graphs, score bundles, and run community detection, use solana-clustering-advanced—this skill focuses on selection, story, packaging, and publication shape.
以交付成果为核心的工作流：承接solana-clustering-advanced风格的分析（以及solana-tracing-specialist的基础能力），生成完整、独立的叙事内容——推文线程、长文或独立文档——且内容具备可复现性和证据关联性。
聚类结果仍具有概率性。案例研究必须将已验证的链上事实与推论分开，标注置信度，除非用户提供的上下文已公开且引用符合合法性要求，否则不得提及真实世界身份（详见crypto-investigation-compliance、on-chain-investigator-agent）。
不得协助骚扰、协同围攻或未经同意的去匿名化操作。不得将启发式方法作为犯罪的法律证据呈现。
如需了解如何构建图谱、对分组评分以及运行社区检测，请使用solana-clustering-advanced——本技能专注于案例筛选、叙事构建、内容打包以及发布形式优化。

1. Case selection and seed identification

1. 案例筛选与种子识别

  • Prioritize high-signal events the user specifies: launches with unusual volume, liquidity events, coordinated sells, or public tips tied to Pump.fun-class, Raydium, Jupiter, or similar—verify each claim against chain data.
  • Strong seeds — One signature, token mint, suspected dev or early buyer, or program-derived account; document why the seed is anomalous (timing, size, program path).
  • Rapid triage — Before a deep dive, check: Jito bundle overlap (where visible), tight timing bands, PDA/authority reuse—abort or narrow scope if the graph is too noisy or ambiguous.
  • 优先处理用户指定的高信号事件：异常交易量的项目上线、流动性事件、协同抛售，或与Pump.fun类平台、Raydium、Jupiter等相关的公开线索——需针对每条线索验证链上数据。
  • 优质种子——单个签名、代币铸币地址、疑似开发者或早期购买者钱包，或程序派生账户；需记录该种子为何异常（时间点、规模、程序路径）。
  • 快速分类——在深入分析前，检查：Jito bundle重叠情况（若可见）、紧凑的时间区间、PDA/权限复用——若图谱过于嘈杂或模糊，则终止分析或缩小范围。

2. Multi-layer graph construction and clustering (summary)

2. 多层图谱构建与聚类（概述）

  • Build temporal directed graphs: nodes = resolved owner wallets (and ATAs/programs when needed); edges = transfers, relevant CPIs, bundle co-participation, ATA create/close—slot/time on every edge.
  • Layer heuristics (apply in documented order; tune windows per case):
    • Temporal coordination (e.g. sub-5s bands—context-dependent).
    • Jito bundle siblings and tip patterns (weak alone).
    • Launch-window density (e.g. first 60s—tune per protocol).
    • PDA derivation and authority lines.
    • Behavioral fingerprints (CU bands, swap route shapes, peel-like hops).
    • Optional ML features from exports (entropy, burstiness, program diversity)—validate against seeds.
  • Community detection (Louvain, Leiden, etc.) → ranked clusters with 0–100 or tiered confidence from heuristic overlap and density—document weights and cutoffs.
Full methodology lives in solana-clustering-advanced; reuse its reporting tables and falsification criteria.
  • 构建时序有向图谱：节点 = 已解析的所有者钱包（必要时包含ATA/程序）；边 = 转账记录、相关CPI、bundle共同参与、ATA创建/关闭——每条边需标注slot/时间。
  • 分层应用启发式规则（按记录的顺序应用；根据案例调整时间窗口）：
    • 时序协同（例如：5秒内的时间区间——视上下文而定）。
    • Jito bundle关联项与小费模式（单独使用时置信度较低）。
    • 上线窗口期密度（例如：上线后前60秒——根据协议调整）。
    • PDA派生与权限链路。
    • 行为指纹（CU区间、兑换路径形态、类似剥离的跳转）。
    • 可选：导出文件中的ML特征（熵值、突发度、程序多样性）——需与种子进行验证。
  • 社区检测（Louvain、Leiden等算法）→ 根据启发式规则重叠度和密度生成带0–100分或分级置信度的排名聚类——需记录权重和截断阈值。
完整方法论详见solana-clustering-advanced；可复用其报告表格和证伪标准。

3. Narrative and storyline development

3. 叙事与故事线开发

  • Turn clusters into chronological arcs with neutral section labels where useful: e.g. launch / accumulation / high-coordination window / large moves / post-event flows—avoid criminal verdicts in headings.
  • Quantify carefully: volumes and counts from parsed transfers; “victim” counts only with clear definitions (e.g. wallets receiving from a contract—state as approximate if sampled).
  • Evidence moments — Anchor the story on signature links, bundle IDs where available, and explorer URLs (Solscan, SolanaFM, etc.); optional annotated screenshots from public explorers/visualizers (verify licensing for republished images).
  • Counterfactuals / alternatives — Brief “what if this were organic?” and which observations would argue against coordination—strengthens credibility.
  • 将聚类结果转化为按时间顺序的叙事框架，必要时使用中性章节标题：例如，上线/积累/高协同窗口/大额异动/事件后资金流向——避免在标题中做出刑事判决类表述。
  • 谨慎量化：从解析后的转账记录中提取交易量和交易次数；“受害者”数量仅在有明确定义时列出（例如：从合约接收资金的钱包——若为抽样统计需标注为近似值）。
  • 证据锚点——将故事与签名链接、可用的bundle ID以及浏览器URL（Solscan、SolanaFM等）绑定；可选择使用来自公开浏览器/可视化工具的带注释截图（需验证重发布图片的许可）。
  • 反事实/替代解释——简要说明“如果这是有机行为会怎样”，以及哪些观测结果会反驳协同行为——这能增强内容可信度。

4. Visualization and evidence packaging

4. 可视化与证据打包

  • Visuals (choose what fits the medium): cluster graphs with communities; timeline strips of key txs; Sankey-style flow summaries; heatmaps of heuristic strength per wallet—embed or link to live explorers for every critical hop.
  • Export bundle — Include:
    • CSV of cluster members, roles (if any), and key metrics.
    • Query scripts or saved SQL (Dune/Flipside) with parameters and run date.
    • Version notes for RPC/indexer queries (method names change—cite docs snapshot or date).
  • Reproducibility — Enough detail that a third party can re-fetch the same txs and rebuild a similar graph (filters, time range, mint/program IDs).
  • 可视化内容（根据媒介选择合适形式）：带社区标注的聚类图谱；关键交易的时间线条；桑基图风格的资金流向摘要；每个钱包的启发式规则强度热图——每个关键跳转都需嵌入或链接至在线浏览器。
  • 导出包包含：
    • CSV文件：聚类成员、角色（若有）以及关键指标。
    • 查询脚本或已保存的SQL（Dune/Flipside），需包含参数和运行日期。
    • RPC/索引器查询的版本说明（方法名称可能变更——需引用文档快照或日期）。
  • 可复现性——需提供足够细节，使第三方能够重新获取相同交易并重建相似图谱（筛选条件、时间范围、铸币/程序ID）。

5. Output formats

5. 输出格式

  • Thread — Numbered posts: hook → seed → method (short) → timeline → cluster summary → evidence links → limitations → disclaimer (not legal/financial advice; probabilistic clustering).
  • Standalone doc — Executive summary, methodology appendix, full evidence table, glossary of heuristics, changelog if updated after feedback.
  • 推文线程——编号帖子：钩子→种子→方法（简短）→时间线→聚类摘要→证据链接→局限性→免责声明（非法律/财务建议；聚类结果具有概率性）。
  • 独立文档——执行摘要、方法论附录、完整证据表格、启发式规则术语表、若经反馈更新则需包含变更日志。

6. Ethical and professional guardrails

6. 伦理与专业准则

  • Educational and defensive framing; no call to vigilante action.
  • Precision over viral certainty—weak clusters belong in an appendix, not the headline.
  • Illicit framing: use suspected coordination, reported incident, or cite public charges only when the user supplies citable sources—do not invent legal conclusions.
  • Cross-check on-chain-investigator-agent for evidence style and defi-security-audit-agent if token/contract risk is part of the same story.
Goal: Polished, verifiable community education and fraud awareness—built from immutable public signals, with humility about what clustering can and cannot prove.
  • 采用教育性和防御性框架；不得号召私刑行动。
  • 优先精准性而非博眼球的确定性——置信度低的聚类应放在附录中，而非标题。
  • 关于非法行为的表述：使用“疑似协同”、“已报告事件”，或仅在用户提供可引用来源时提及公开指控——不得自行做出法律结论。
  • 如需证据风格参考，可交叉核对on-chain-investigator-agent；若代币/合约风险是故事的一部分，可参考defi-security-audit-agent。
目标：打造经过打磨、可验证的社区教育与欺诈意识内容——基于不可篡改的公开信号，同时对聚类方法的能力与局限性保持谦逊态度。