Methodology

How Our Intelligence Stack Works

Full transparency into the production analysis layers that transform consumer reviews into investor-ready chain intelligence. No black boxes, no fake engine counts.

⚙
Neural Core
Anomaly Detection & Embeddings Engine

Neural Core is the foundation of ReviewSignal's intelligence stack. It transforms raw review text into 384-dimensional semantic embeddings using MiniLM, then applies Isolation Forest anomaly detection to identify statistically significant deviations in rating, sentiment, and review volume across all tracked locations.

✓ MiniLM 384-dim Embeddings — all-MiniLM-L6-v2 captures semantic meaning far beyond keyword matching. Each review is encoded into a dense vector space where similar opinions cluster together.
✓ Isolation Forest Anomaly Detection — Trained on 8,700+ real samples, the model identifies locations exhibiting statistically unusual rating drops, sentiment shifts, or volume spikes.
✓ Welford's Online Statistics — Incremental algorithm tracks running means, variances, and z-scores per entity without reprocessing historical data. Scales to millions of data points.
✓ Zero External API Cost — The entire inference pipeline runs locally on our infrastructure. No OpenAI, no cloud NLP services. Every embedding computed in-house at zero marginal cost.
384
Vector Dimensions
<1s
Inference Latency
$0/mo
API Cost
8,700+
Training Samples
♬
Echo Engine
Sentiment Propagation & Confidence Scoring

Echo Engine models how consumer sentiment cascades across geographic and brand networks. Using sparse matrix propagation and live confidence scoring, it generates BUY/HOLD/SELL style operating signals by analyzing how sentiment at one location predicts movements at related locations and competing brands.

✓ Sparse Matrix Propagation — Sentiment signals propagate through a weighted adjacency matrix connecting locations by geography, brand affiliation, and competitive relationships.
✓ Confidence-weighted runtime — Each signal carries explicit strength only when the live propagation payload supports it, instead of promising a decorative simulation layer.
✓ Distance-Decay Weighting — Propagation strength decreases with geographic distance following an exponential decay function, ensuring local signals carry appropriate weight.
✓ BUY/HOLD/SELL Signal Generation — Final output is a discrete trading recommendation with confidence interval, direction magnitude, and temporal horizon.
79
Live Chains
21,090
Live Locations in Scope
0.78
Avg Confidence
3
Signal Types
★
Singularity Engine
7-Level Causal Analysis & Semantic Resonance

Singularity goes beyond correlation to establish causation. Its 7-level causal analysis framework traces sentiment shifts back to their root causes through temporal manifold analysis, semantic resonance mapping, and causal archaeology. It classifies events as STRUCTURAL (permanent operational change) or EPISODIC (temporary fluctuation), giving investors critical context for position sizing.

✓ 7-Level Causal Depth — From surface symptoms to root causes: event detection, pattern matching, temporal alignment, semantic clustering, causal chain reconstruction, structural classification, and impact projection.
✓ Temporal Manifold Analysis — Maps sentiment trajectories in multi-dimensional time space, separating seasonal patterns from genuine structural shifts in brand perception.
✓ Semantic Resonance — Detects when review language patterns across locations converge on similar themes, indicating systemic rather than isolated issues.
✓ STRUCTURAL vs EPISODIC Classification — Automatically determines whether a sentiment shift represents a permanent operational change or a temporary disruption, directly informing position duration.
7
Causal Levels
2
Event Classes
90-day
Lookback Window
Enterprise
Tier Required
✎
Cortex AI
Institutional Narrative Generation

Cortex AI transforms quantitative signals into written analysis for the exact dashboard or report context. It uses section-specific prompts and a constrained institutional research voice rather than one generic market summary.

✓ Claude Foundation Model — Built on Anthropic's Claude for concise institutional commentary tied to the selected section or report block.
✓ Section-Specific Prompts — Supported dashboard and report sections use dedicated prompt templates tuned for that analytical context, including map and competitor views.
✓ Institutional Research Voice — Writing style is constrained toward a colder, desk-grade tone instead of generic assistant language.
✓ Scoped Runtime Context — Commentary is grounded in the current dashboard aggregates, chain state, or export payload available to the runtime.
16
Prompt Templates
Section-aware
Context Mode
Report + dashboard
Output Surfaces
Institutional
Voice Target
⚠
Higgs Nexus
Market Phase & Field Dynamics

Higgs Nexus is the enterprise overlay that interprets market-wide signal balance, consensus, and instability. It classifies the current phase state from recent BUY/HOLD/SELL distributions and field dynamics, giving portfolio teams a higher-level read on whether the system is orderly, chaotic, or breaking in one direction.

✓ Phase State Detection — Labels the current state as symmetric, chaotic, bullish break, or bearish break based on recent signal agreement and dispersion.
✓ Consensus vs Diversity Readout — Measures whether the market is aligning behind one direction or fragmenting into mixed cross-chain signals.
✓ Swarm Snapshot — Tracks how many chains are active in the current signal field and how BUY/SELL/HOLD pressure is distributed.
✓ Enterprise Overlay — Sits above chain-level analysis to help investors judge whether a single-chain move is happening inside a stable or unstable broader environment.
4
Phase States
7d
Signal Window
BUY/HOLD/SELL
Field Inputs
Enterprise
Tier Required
Get started

See our engines in action with a
free 14-day trial

Start with the core chain intelligence operating surface in your free 14-day trial. No credit card required.

Start Free Trial View Pricing