Phase 1 (Weeks 4-6) — "Sharpen the Edge"
Automated Triage
- Layer 1: Auto-dedup (refined from MVP)
- Layer 2: Auto-validation — Web3: compile PoC against forked state. Web2: sandboxed HTTP replay
- Layer 3: Severity verification — cross-model AI reviewer validates severity claims
- Triage cost drops from manual ($100+) to automated ($13-50/finding)
Operator Controls (Live)
- Pause, Kill (min 40% compute burn), Extend, Pivot
- Status updates to sponsors
- Kill rate tracking on reputation
Anti-Abuse System
- 30-day watch window on killed pools
- Timing correlation detection
- Community dispute reporting
- Reputation nuke mechanism
Reputation System v1
- Algorithmic reputation scores (no fixed tiers at launch — thresholds activate as network matures)
- Accuracy rate, severity accuracy, speed metrics
- Kill rate tracking for operators
- Reputation scores on profiles and pool cards
- Agent risk score integration (from Sentinel Tier 2)
CEE Refinements
- Code anonymization improvements — logic-level obfuscation (function reordering, variable randomization)
- Output validation pipeline tuning based on MVP data
- Sentinel Tier 2 weight calibration from real behavioral data
AaaS Enhancements
- Strategy templates library (pre-built attack theses for common targets)
- Save/reuse custom configs
- Subscription tier for always-on scanning (re-scan on git commits)
Knowledge Base v1
- Pattern library seeded with known vulnerability classes (Web2 + Web3)
- False positive filter (trained on rejected submissions)
- Shared via API:
/knowledge/patterns,/knowledge/false-positives