Mindforge Intelligence

Validation & Methods

Current product methodology, deployed configuration, evidence boundaries, and known limitations. Not investment advice.

Market Risk Classifier Space Weather (SEWS)

Product

Market Risk Classifier (MRC)

1. Current Product and Evidence Status

The Market Risk Classifier publishes a rules-based market-backdrop report before the United States market opens, using market information available through the prior completed session. Its intended product framework defines five base states. The deployed composer can still emit WHIPSAW as a competing state, so exact five-state production emission is not claimed. The dated record is preserved for operational review.

Product component	Current production status	Historical performance status
STABLE	Base state, rule version 2.3.0.	Withheld.
SHIFTING	Base state, rule version 5.2.0.	Withheld.
ELEVATED	Base state, rule version 6.0.0.	Withheld.
SHOCK	Base state, rule version 2.0.0.	Withheld.
CRISIS	Base state, rule version 6.0.0.	Withheld.
WHIPSAW	Watch rule version 1.0.0 is deployed. Its intended product role is contextual.	Withheld.
RESOLVED	Context added after a return to STABLE. It has no base-state rule file.	Withheld.

Production configuration verified from the live rule bucket on August 12, 2026.

1.1 What Has Been Verified

The production-faithful decision replay is sealed. It reproduces the deployed target-date path over 8,789 historical sessions from January 23, 1990 through December 30, 2024.
Live receipts separately establish operation and traceability. Retained March 2026 receipts preserve delivery timestamps, displayed performance-manifest versions, delivery identifiers, and alert-hash prefixes. They do not prove the rule bytes that produced each classification.
The replay does not validate performance. Precision, recall, false alarms, lead time, recovery timing, state shares, and economic value remain withheld.
CRISIS 7.0.0 is inconclusive, and SHIFTING 6.0.0 failed. Neither candidate is deployed or describes the public product.

Historical aggregates remain withheld until a production-faithful outcome scorer is approved. A decision-replay seal proves exact classification-path reproduction, not predictive accuracy or buyer value.

2. Product Taxonomy

The intended product framework defines five base states and two contextual report behaviors. The deployed composer still allows WHIPSAW to compete with those base states.

Component	Functional role	Product status
CRISIS	Highest-severity base state.	Deployed.
SHOCK	Acute-dislocation base state.	Deployed.
ELEVATED	Elevated-backdrop base state.	Deployed.
SHIFTING	Changing-backdrop base state.	Deployed.
STABLE	Intended stable-backdrop base state.	Deployed.
WHIPSAW	Context when visible market stress diverges from STABLE.	Contextual in the intended product ontology.
RESOLVED	Context after a non-STABLE episode returns to STABLE.	Derived in the report layer.

The intended product framework defines five base states: STABLE, SHIFTING, ELEVATED, SHOCK and CRISIS. WHIPSAW and RESOLVED are intended contextual report overlays, not sixth and seventh base states. The deployed composer still allows WHIPSAW to compete with base states. EARLY_RECOVERY or RE_ENTRY is separate provisional research, not RESOLVED and not a production capability.

3. Methodology and Known Limits

How the production record is reconstructed and what remains unvalidated.

3.1 Publication Clock

The classifier runs before the market opens on the publication date.
The target-date market row is not yet available, so production carries forward the prior completed session's market values.
Each historical publication date is reconstructed independently through the complete feature calculation and rule engine.
The historical decision record is sealed only when clean regeneration matches byte for byte.

3.2 Scope

What MRC does	What MRC does not do
Classifies market risk with fixed, rules-based definitions	Provide price targets or trade recommendations
Combines environmental inputs with market confirmation and context	Operate without market variables in its confirmation and context layers
Preserves dated classifications and rule artifacts	Present archive creation as proof of contemporaneous delivery
Supports a buyer-scored live pilot	Promise a market, portfolio, or economic outcome

3.3 Current Validation Boundary

The exact deployed decision path is reproducible across 8,789 historical target sessions.
The independent-event state-quality ledger is sealed for internal candidate comparison.
No historical aggregate is approved outward for precision, recall, misses, false alarms, lead time, recovery, state frequency, or state share.
No portfolio or economic-value result is approved because no named buyer supplied the action policy being simulated.

3.4 Environmental Input Distinction

The primary environmental inputs do not use S&P 500 Index or Cboe Volatility Index values. The complete classifier separately uses market variables for confirmation and context. This supports an input-layer distinction. It does not extend that distinction to the complete classifier.

3.5 Reviewed State-Machine Research

Same-trigger continuation is a validated research correction and is not deployed.
WHIPSAW is contextual in the retained research composer and must not compete with base states.
SHIFTING is ordered above STABLE in the retained research composer.
Cooldown-independent full-window entry failed and remains rejected.

3.6 Data Sources

The classifier uses S&P 500 Index and Cboe Volatility Index market data alongside exogenous environmental signals. Detailed environmental-source disclosure is reserved for controlled diligence. References to third-party providers do not imply endorsement.

3.7 Recovery Research Status

RESOLVED is contextual report behavior after a return to STABLE. Historical recovery and re-entry benchmarking motivates a separate EARLY_RECOVERY or RE_ENTRY hypothesis, but it does not prove that such a production alarm exists.

Recovery timing, re-entry timing, and portfolio-value claims remain withheld from outward use.

3.8 Candidate Rule Status

CRISIS 7.0.0 was inconclusive against CRISIS 6.0.0, and SHIFTING 6.0.0 failed against SHIFTING 5.2.0. Neither candidate is deployed, and their candidate metrics are not product claims.

4. Event Risk Briefs

An Event Risk Brief is issued when an Escalation, Stand-Down, or Divergence warrants a report. The brief records the dated classification and market context. Historical precedent percentages and outcome verdicts are currently suppressed because those scorers are not approved for outward use.

4.1 Event Library

Event	Operational description	Evidence boundary
Escalation	The system enters a more severe state than the prior session	The dated report is reviewable. Aggregate success rates are withheld.
Stand-Down	The system returns to STABLE after a non-STABLE episode	RESOLVED may be added as context. Recovery performance is withheld.
Regime Active	A non-STABLE state persists (reported with day count)	The current state is descriptive, not a forecast.
Drawdown Divergence	S&P 500 down 4%+ over 10 sessions while STABLE	WHIPSAW may be added as context. Historical performance is withheld.
Volatility Divergence	VIX up 25%+ over 5 sessions while STABLE	WHIPSAW may be added as context. Historical performance is withheld.
Routine	None of the above	No Event Risk Brief is required.

4.2 Archive and Receipt Provenance

Production receipts prove delivery. A retained receipt can establish the delivery time, displayed performance-manifest version, delivery identifier, and alert-hash prefix for that specific report. It does not prove the engine rule bytes.
Archive pages provide context. An archive page created after an event does not prove that the page existed before the outcome.
Historical reconstruction is labeled. A reconstructed classification is not described as a live alert unless independent delivery evidence exists.
Market values name their source and as-of date. External market context is kept separate from the exact value read by production.

Event Risk Briefs are research documents. Archive history is not a forecast, and no public aggregate currently establishes accuracy, lead time, recovery timing, or portfolio value.

Review the evaluation pilot Or email sales@mindforge.tech

Space Weather

Space Weather Early Warning System (SEWS)

SEWS is a separate product from MRC. It classifies space weather risk, not market risk.

5. What SEWS Does

SEWS classifies space weather risk using upstream Forbush Decrease detection. It provides probabilistic risk context days ahead of many downstream storm-time indicators and public alerts.

SEWS is a complementary upstream scenario planning layer, not a replacement for official space weather forecasts. It does not predict specific events or their timing.

Not affiliated with NOAA, ESA, NASA, or any government space weather service.

6. SEWS Performance

Full-period backtest (2010–2024, 15 years, 31 cataloged events).

6.1 Full-Period Backtest

Tier	Precision	Events Caught	Avg Lead	Alerts/Year
ELEVATED	23%	58%	~7d	~7/yr
CRITICAL	46%	29%	~8d	~2/yr

ELEVATED

Wider net. More false alarms. Use for early scenario planning.

CRITICAL

Selective. Higher confidence. Worth escalating monitoring posture.

6.2 Walk-Forward Validation (Out-of-Sample)

9-fold walk-forward (train on 2010–2016 base + expanding window, test on each subsequent year 2016–2024). 14 events in walk-forward test periods.

Tier	Precision (fold-avg)	Events Caught	Lead (event-weighted)
ELEVATED	17% ± 17%	79% (11/14 events)	~6.3 days
CRITICAL	75% ± 14%	57% (8/14 events)	~7.6 days

Validates: Multi-day lead time holds out-of-sample (6–8 days). CRITICAL precision strong (75%). ELEVATED recall strong (79%).

Known limitations: High fold-to-fold variance on ELEVATED precision (sparse events per year). Small sample (14 events in walk-forward test periods).

7. SEWS Methodology

7.1 Detection Approach

SEWS uses environmental stress indicators and public space weather data streams to classify Forbush Decrease risk. Validation is performed against a pre-published event catalog (2010–2024) using a fixed 14-day scoring horizon.

7.2 Data Sources

Public space weather and geomagnetic datasets (NOAA SWPC)
Public CME and solar event catalogs (NASA DONKI)
Environmental monitoring networks for event cataloging and cross-checks
Mindforge Signal Platform (composite indices)

7.3 Scope

What SEWS does	What SEWS does not do
Classifies environmental risk states (NORMAL / ELEVATED / CRITICAL)	Predict specific events or their timing
Delivers probabilistic upstream context days before official alerts	Replace NOAA/ESA confirmed-event forecasts
Walk-forward validates with expanding-window methodology	Claim affiliation with government space weather services
Publishes tier performance transparently (including false alarm rates)	Guarantee future performance based on historical validation

7.4 Overfitting Controls

Out-of-sample walk-forward validation (9 folds, 2016–2024)
Separate reporting of full-period backtests (descriptive) vs. walk-forward results (out-of-sample)
Event-weighted lead time for out-of-sample headline lead time
Transparent tier tradeoffs: ELEVATED prioritizes coverage, CRITICAL prioritizes confidence

View SEWS Product Page

Informational research only. Not investment, operational, or safety advice. Past performance does not guarantee future results. For complete reproduction procedures, request an auditor-friendly validation packet via the form above.