Data storage
Market data, forecasts, evaluations, news and analysis results are stored in PostgreSQL.
DATA LIVE
Loading ...
Transparency
StockMarketScore combines market data, forecasts, deterministic scoring rules, historical evaluation and selected model-based analysis. This page explains what the system measures, how the daily processing flow works and how the published values should be interpreted.
The scores and analyses are informational tools. They are not guarantees, not personalized investment advice and not a promise of future returns. Every model and every dataset has limitations.
System overview
The public website does not calculate the scores itself. A Go worker prepares and stores the data in PostgreSQL. The Go API reads those stored results and the Next.js frontend displays them.
Daily pipeline
The daily pipeline is ordered deliberately. Older forecasts are evaluated before new forecasts and final stock analyses are created.
The worker imports Open, High, Low, Close and Volume data. These time series are the foundation for technical indicators, forecasts and historical evaluations.
Older forecasts are compared with later observed market data once the target date is available. This creates historical evaluation data without using future information too early.
Recent news articles are imported and linked to the assets they mention or affect.
New article-asset combinations are analyzed by a local Qwen model and then validated before being stored.
Macroeconomic and company-level data are imported before the final stock analysis is created.
Technical indicators are imported or calculated from the latest available market data.
Chronos receives historical time series and produces model-based forecast data for later scoring and evaluation.
The worker builds the full analysis context, applies deterministic scoring logic and stores the final analysis values.
Processing types
The system uses both deterministic code and model-based outputs. The distinction matters because a score produced by rules is different from a forecast or language-model assessment.
Rules, filters, score normalization, aggregation, validation and database constraints are implemented in code. With identical input data and identical methodology, these parts are reproducible.
Some steps use trained models. Their outputs are not guarantees. They are structured model outputs based on the data available at the time.
Model outputs are not simply copied into the product. They are parsed, validated, constrained and then converted into structured values.
Chronos and forecasts
Chronos is a time-series model. It receives historical OHLCV data and produces a model-based forecast. The forecast is not a certainty and does not know future events.
Scores and interpretation
The published values are analytical metrics, not guarantees. Exact component formulas and weights must be documented from the current scoring implementation before they are published as formulas.
A combined internal analysis score. It is not a direct probability and does not mean that a score of 70 equals a 70% chance of a price increase.
How to read it
Use it as a relative analytical score within the StockMarketScore methodology.
Important limitation
The exact component weights must be documented from the current scoring code before publication.
A deterministic score derived from forecast-related information. It is not the raw Chronos forecast itself.
How to read it
It summarizes forecast-related signals into a normalized scoring value.
Important limitation
It should not be described as expected return unless the API explicitly exposes expected return separately.
A separate analysis value for risk-related information.
How to read it
Risk should be read according to the documented scale direction in the implementation.
Important limitation
No single risk value can fully represent all real-world investment risks.
An internal assessment of the quality or completeness of the analysis basis.
How to read it
It can help users understand how strong the available analytical basis is.
Important limitation
Confidence is not a guaranteed probability of a future price move.
A diagnostic value based on previously evaluable forecasts.
How to read it
It describes historical forecast quality where enough completed evaluations exist.
Important limitation
It is backward-looking and is not part of the current Overall Score contract.
Point-in-time logic
Historical analyses must only use information that was already known at the selected date. Later forecast evaluations are not allowed to improve an earlier historical score.
Forecast created
A forecast is generated with only the data available at its reference time.
Future value observed
The real later market value becomes available only after the target date.
Evaluation allowed
The forecast evaluation may only be used in analyses whose as-of date is after the real value was known.
Sentiment & News
Sentiment & News is an independent feature. It evaluates recent news article-asset combinations and currently does not change the Overall Score.
| News value | Range | Meaning |
|---|---|---|
| Sentiment Score | -100 to 100 | Negative values indicate negative asset-specific interpretation, positive values indicate positive interpretation and 0 is neutral. |
| Impact Score | 0 to 100 | Estimates how strongly the described event could affect the specific asset. |
| Analysis Confidence | 0 to 100 | Expresses how confident the model is in the article assessment based on the available text. It is not a market probability. |
| Relevance Score | 0 to 100 | Measures how directly and materially the article relates to the specific asset. |
News Sentiment is currently a separate analysis area. It does not automatically affect the Overall Score, Forecast Score, Risk, Confidence or Historical Reliability.
Missing data
A core principle of the system is that missing or unfinished analysis should not be hidden behind artificial neutral values.
Limitations
The platform is designed to be structured and transparent, but it cannot remove uncertainty from financial markets.