SignalGrid
Institutional NLP Intelligence
Market OverviewCompany MonitorSector IntelligencePortfolioAlertsThemesEventsAskReportsMethodologyAdmin
Coverage
503 companies monitored
12,842 documents processed
97% source coverage
Last updated May 19, 2026 9:35 AM ET

Methodology & Trust

Transparent scoring framework, source coverage, normalization logic, confidence construction, limitations, compliance notes, and auditability.

Last model version
NLP-Risk-2026.05
0.0 pts
Mock model registry reference for prototype governance.
Source coverage
97%
+0.4 pts
Coverage across monitored SPY constituents.
Average latency
22 min
-4.0 pts
Median mock ingestion-to-scoring latency.
Historical availability
7 years
0.0 pts
Mock backtest and audit trail availability.
Audit trail status
Enabled
0.0 pts
Every AI-style summary links to supporting evidence.
Entity mapping
99.2%
+0.1 pts
Ticker, company, subsidiary, and sector mapping quality.
Update frequency
15 min
0.0 pts
Mock scoring cadence for high-priority sources.
Data latency
Low
-1.0 pts
Latency varies by source entitlement and document type.

Score definitions

Sentiment, risk, volume, novelty, confidence, and narrative-shift scores are normalized indicators. They are designed to expose language change, not to produce investment recommendations.

Normalization method

Sentiment scores are normalized within company, sector, and index contexts. A score of 80 does not mean 80% positive; it indicates the company's current language environment is strongly positive relative to its historical and peer baselines.

Confidence score

Confidence blends source reliability, source dispersion, evidence count, entity mapping quality, recency, and agreement across document types.

Source quality weighting

Earnings transcripts, filings, regulatory documents, and high-reliability news receive higher evidence weights than low-frequency or lower-verifiability sources.

Percentiles and z-scores

Percentiles compare the current score to company, sector, and SPY baselines. Z-scores show distance from a trailing baseline and are used for alert thresholds.

Known limitations

This prototype uses local mock data. Production usage would require entitlement-aware ingestion, source validation, audit trails, and documented model governance.

Compliance notes

The system avoids buy/sell recommendations and presents source-backed signal context for research workflows. Output should be reviewed by qualified investment professionals.

Auditability

Every summary should retain links to contributing sources, timestamps, topic tags, and signal contribution metadata so analysts can inspect the evidence chain.