Classic indicators lag. Opaque tools cannot be checked.
Positioning: Institutional standard for assisted trading, built on statistical discipline and complete traceability.
Most retail traders work with indicators that describe the past. Moving averages and oscillators smooth the price after the movement has taken place, and in transition phases generate conflicting signals.
At the opposite extreme, black-box tools promise automation but don't show how the decision was made. With no model version, no risk parameters, and no error history, there is no real basis for capital allocation.
Meritexia starts from a different premise. Trust is not gained by hiding losses, but by systematically documenting them. Each recommendation includes the probability distribution, invalidation threshold and proposed exposure, and the outcome is published in the open journal for community audit.
| Criterion | Conventional approach | AI black box | Audited model Meritexia |
|---|---|---|---|
| Decisional basis | Lagging indicators, manually calibrated | Opaque score, no reproducible explanation | Probability distribution by market regime |
| Risk control | Fixed stop, often discretionary | Unpublished thresholds, modified retroactively | Automatic invalidation and calibrated exposure |
| Traceability | Screenshots, no unit history | Selective results, no model versions | Public log with hash, latency and denouement |
Built for the daily practice of the systematic trader
The platform is used before opening the session, during exposure monitoring and after closing, during the audit stage. The flow remains the same regardless of the asset analyzed.
- Statistical noise filtering before any operational signal
- Complete hypothesis documentation prior to execution
- Post-session review based on immutable log
From raw data flow to recommendation with clear boundaries
The system ingests multi-source data, detects the market regime and converts uncertainty into operational parameters: position size, invalidation level and validity horizon.
Real-time multi-source stream processing
We normalize quotes, volumes and macro context data into a single format. Outliers are flagged, not deleted, to preserve the integrity of the audit.
- 320 ms — median normalization and deduplication latency
- UTC — unique timestamp on each event
- Filter — separation between structural motion and noise
Predictive models of market regime detection
The models estimate the probability of each regime—trend, lateral, volatile transition—and report the uncertainty. When confidence drops below the threshold, the system recommends reducing exposure.
- v4.x — strict versioning, no retroactive rewriting
- Walk-forward — validation on consecutive time windows
- p-regime — explicit probability, not opaque score
Dynamic exposure calibration and automatic override
Each recommendation includes the maximum accepted loss in risk multiples and the exact condition under which the scenario is declared invalid. Discipline is dictated by parameters, not emotion.
- -0.8R — example of standard invalidation threshold
- 0.25–1.0% — typical risk range per idea, configurable
- Car — logical closure of scenario on invalidation
Scalable recommendations and workflow integration
The results are delivered to the workspace and via the programming interface, with the same data structure. Teams can redirect signals to their own execution engine.
- API + webhook — documented and stable JSON format
- CSV / log — full export for reconciliation
- Roles — separate access for analyst and manager
Proof is not a promise. It is a verifiable structure.
We do not publish selective historical returns or present testimonials. We publish the method by which any recommendation can be independently verified, including when the script is invalidated.
| Registry field | Content and operational role |
|---|---|
| UTC timestamp | The timing of issuing and closing the scenario. It allows correlation with market data and measurement of actual latency. |
| Pattern Hash | Unique identifier of the version (eg v4.2.1 + parameter set). Excludes retroactive modification of logic. |
| Risk parameters | Proposed exposure, invalidation threshold in R, validity horizon. The basis for the calculation of capital disciplines. |
| Script vs. denouement | The initially estimated distribution compared with the observed evolution. Deviations are kept and analyzed quarterly. |
| Check status | Confirmed / Invalidated / Pending / Community Audited. Status is updated only once, with time stamp. |
The table describes the structure of a ledger entry, not financial results. Illustrative methodological sample.
Daily routine: four steps, same standard of rigor
The flow is designed for real trading sessions, from morning preparation to evening reconciliation.
Statistical noise aggregation and filtering
Raw data is cleaned, synchronized and evaluated for consistency. Sessions with incomplete data are flagged with caution.
Generating probability distributions
The model estimates possible regimes and attaches a probability and confidence interval to each, not a unique direction.
Issuance of recommendation with invalidation thresholds
Only scenarios that pass the quality filter become recommendations, each with proposed exposure and explicit exit condition.
Registration in the open register
The recommendation and outcome are published for audit. Lessons from invalidated scenarios feed into the next calibration.
Questions from traders and portfolio managers
Concise answers, no unnecessary jargon. For full details, see the methodological register.
How to prevent models from over-optimizing on historical data
We use walk-forward validation and locked out-of-sample windows before calibration. Parameters are frozen across model versions, and any changes generate a new version with a separate history.
Models that perform only on a single regime are relegated to observational status and do not issue operational recommendations.
What exactly does the public log contain and how is it verified?
Each entry contains the UTC timestamp, model version hash, risk parameters, estimated distribution, and observed outcome. The final status is confirmed, invalidated or community audited.
Inputs are immutable. Fixes appear as new versions, and community members can report inconsistencies that are resolved publicly.
Meritexia takes custody of funds or executes orders directly
No. Meritexia is a decision analysis and optimization layer. The funds remain in their own accounts, with the broker chosen by the user, and the execution remains under his exclusive control.
Referrals can be directed via API to its own infrastructure, but the platform does not initiate orders without user action.
How to integrate recommendations into your own infrastructure
Through the web workspace for manual analysis and through APIs and webhooks for algorithmic flows. The data structure is stable and documented, with the same set of fields as in the public log.
Median latency and UTC time format are documented to allow reconciliation with your own execution log.
How the system reacts in extreme macroeconomic volatility
In events with major macro impact, the regime filter automatically reduces the proposed exposure or suspends the issuance of new scenarios. The invalidation thresholds are narrowing and the validity horizon is shortening.
These periods remain visible in the log, precisely to allow the evaluation of the system's behavior under stress conditions.
Evaluate the method on real data at your own pace.
Request access to explore the methodology registry, public journal structure, and risk calibration mode. No commitment, no commercial pressure.