WhaleScoutTranscriptScanner / FLTSPerplexitySentimentServiceOpenBB Adapter

Intelligence & OSINT

Four specialized intelligence modules feed the LangGraph orchestrator with institutional flow data, earnings transcript signals, AI-curated sentiment bundles, and microstructure analysis — all processed before any ML or LLM inference.

Intelligence Data Flow

SuperIntelOrchestrator / LangGraph

↑ intelligence inputs

├─── MicrostructureAgent (Databento L2 — applied FIRST, before any LLM)

├─── WhaleScout (institutional flow composite scorer)

│ ├── InstitutionalFilings (SEC EDGAR)

│ ├── MarketIntelligence (OpenBB Adapter)

│ ├── OnChainIntelligence

│ ├── CorporateIntelligence

│ └── TranscriptScanner (FLTS triggers)

└─── PerplexitySentimentService (sonar-pro, 5 intel bundles)

WhaleScout — Institutional Flow Scorer

WhaleScout aggregates five data source ports and produces a single WhaleSignal in the range [−1, +1]. It is consumed directly by the research_agent and debate_agent nodes.

Composite Scoring Weights

40%

FLTS Tone Shift Score

(TranscriptScanner output)

30%

Dark Pool Score

(OpenBB dark pool prints)

30%

Filings Score

(SEC EDGAR 13-F / Form 4)

WhaleSignal Output Structure

@dataclass
class WhaleSignal:
    signal: float           # composite score [-1.0, +1.0]
    recommendation: str     # STRONG_BUY / BUY / NEUTRAL / SELL / STRONG_SELL
    tone_shift_score: float  # FLTS component
    dark_pool_score: float   # dark pool component
    filings_score: float     # SEC filings component
    on_chain_score: float    # on-chain activity (crypto only)
    confidence: float        # confidence in composite [0, 1]
    sources_available: list[str]  # which ports returned data

TranscriptScanner — FLTS Trigger Categories

Forward Language Tone Score (FLTS). Parses earnings call transcripts looking for 6 linguistic trigger patterns. The output is the tone_shift_score (0–1) that feeds into WhaleScout at 40% weight.

Trigger Category	Examples	Signal
Hedging Language Increase	"if conditions deteriorate", "risks to our outlook"	BEARISH
Certainty Language	"we are confident", "we expect significant growth"	BULLISH
Question Deflection	Repeating question, pivoting to guidance	BEARISH
Analyst Tone Shift	More pointed questions, follow-up skepticism	BEARISH
Management Tone Shift	Defensiveness increase vs prior quarter	BEARISH
Forward Guidance Specificity	Quantified guidance with exact figures	BULLISH

OpenBB Adapter — Dark Pool & Institutional Data

Data Feed	Description
Dark Pool Prints	Off-exchange block trade activity and dark pool volume spikes
Short Interest	Days-to-cover ratio, short interest % of float, trend change
Institutional Ownership	13-F ownership %, QoQ change in major holder positions
Insider Transactions	Form 4 SEC filings — director and officer buys/sells

PerplexitySentimentService

Configuration

model: sonar-pro
cache_key: sentiment:{ticker}
cache_ttl: 3600s (1 hour)
circuit_breaker_threshold: 5 failures
recovery_timeout: 300s

Called By

•SuperIntelOrchestrator (step 2 — sentiment enrichment)
•UnifiedSentiment model (feature input)
•research_agent node (LangGraph deep dive)
•ResearchService (cost-gated on-demand research)

5 Intel Bundle Types

news

Breaking market news, analyst note highlights, earnings surprises

social

Aggregated retail sentiment from social media conversations

macro

Central bank commentary, macro data releases, geopolitical signals

earnings

Earnings beat/miss, guidance revision, management color

corporate_actions

M&A, spinoffs, buybacks, dividend changes

MicrostructureAgent

Applied FIRST in the SuperIntelOrchestrator pipeline — before any ML inference or LLM calls. Veto-capable.

Data Source

Databento Level 2 order book — real-time bid/ask depth with full queue reconstruction.

OBI = (Bid_Vol − Ask_Vol) / (Bid_Vol + Ask_Vol)
# OBI range: [-1.0, +1.0]
# Positive: buying pressure
# Negative: selling pressure

Veto Conditions

VETOOBI direction contradicts signal at |OBI| > 0.7
VETOIceberg order wall detected at ≥5× average bid/ask volume — large hidden resistance

ResearchService — Cost-Gated Deep Dive

On-demand research triggered when the deep_research LangGraph node fires. Expensive operation (Perplexity sonar-pro + multiple API calls) — only triggers when all three cost guards pass:

Volatility

30-day IV above threshold for the ticker

Position Size

Research cost < 0.1% of expected position value

Dissent Score

AgentState dissent_score above DEEP DIVE threshold