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)

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