RAPTOR v4.0.0 — Hot Streak Follower (helpers-native)

v4.0.0 (2026-05-12) — plumbing-only migration

NO thesis change. v3.4 quality-first pipeline, whale entry-discipline 5% threshold, nested-positions parser, scoring weights, conviction-tier leverage all preserved verbatim. Migration: mcporter → SenpiClient, scanner-side create_position → producer push_signal(), openclaw cron → producer_daemon, Python state files for daily cap → runtime.yaml risk.guard_rails, MARKET exits → FEE_OPTIMIZED_LIMIT.

Find traders who are already winning big right now, confirm they're quality (ELITE or RELIABLE on TCS), figure out what bet is driving their streak, check the smart money crowd agrees, and piggyback the same trade.

v3.0 — why the complete rewrite

RAPTOR v2.1 traded zero times since deployment. Root cause:

v2.1 used leaderboard_get_momentum_events as the primary signal source and dereferenced three fields on each event — concentration, top_positions, and trader_tags. All three are null on blocked events. And blocked events are ~100% of tier 2 events in normal market conditions (820/820 tier 2 events over a 44-hour window in one recent sample were all blocked — either trader_cooldown_active or system_cooldown_active).

Per the senpi guide for leaderboard_get_momentum_events:

Blocked events are equally valid momentum signals. The blocking only affects notification delivery, not signal quality. Events with top_positions: null (blocked events without position data) will not match any asset filter.

v2.1's filter chain silently dropped every blocked event before ever reaching the SM alignment check. Raptor was mathematically incapable of producing a signal regardless of market conditions.