Glossary | Vyreon Labs

Vyreon Labs uses a recurring set of terms to describe options-market structure, probabilistic expectations, uncertainty, validation, signal processing, research experiments, and production data systems.

This glossary is written for readers who may not have a background in finance, statistics, machine learning, or software engineering. The same terms are used across the weekly reports, the Methodology, How To Read The Charts, the FAQ, and the Research And Engineering page.

Some entries describe the current public SPY telemetry system. Other entries describe historical or active research. A research term is not automatically a production signal or a validated trading method.

The glossary does not disclose proprietary model coefficients, feature weights, covariance states, operational thresholds, private source code, or execution rules.

Quick Index

A to C

Adaptive Model · Amplitude Target · API · Append-Only Event Stream · Approximation Energy · Average Error · Backwardation · Baseline · Bifurcation · Bucket · Calibration · Call Option · Canonical Payload · Causal · Centroid · Coherence · Confidence Interval · Constraint Field · Contango · Continuous Wavelet Transform, CWT · Convolutional Neural Network, CNN · Covariance Matrix · Coverage

D to H

Data Lineage · Days To Expiry, DTE · Delta · Derived Layer · Detail Energy · Dominant Eigenstructure · Effective Horizon · Effective Matured Date · Eigenstructure Instability · EMA Z-Score · Event Study · Excess Return · Expected Mean · Expected Range · Expected Return · Expiry · Expiry Boundary · Exponential Moving Average, EMA · Feature · Field Center · Field Width · Forecast Ledger · Gamma · Gamma Share At 0DTE · Hazard Signal · Horizon · Horizon-Averaged Forward Return

I to M

Idempotent Ingestion · Implied Volatility, IV · Impulse Response Function, IRF · Initial Coefficient Covariance, P0 · Innovation · Innovation Dispersion · Joseph Covariance Update · Kalman Filter · KOI, Open-Interest Centroid · K-Star, K* · Live · Loading Rotation Speed · Market State · Market Structure · Maturation · Maturity Surface · Mixed State · Model Uncertainty · Moneyness, M-Space

N to R

Observation Variance, R · Open Interest, OI · Options Chain · Options Matrix · Out-Of-Sample · Parkinson Realized Volatility · Payload · Payload Hash · Phase Locking · PI95 · Positioning Center · Prediction Interval · Predictive Covariance · Principal Component Analysis, PCA · Prior Estimate · Probability Mass · Process Covariance, Q · Put Option · Realized Return · Realized Volatility · Recurrent Neural Network, RNN · Recursive State Estimation · Residual · Ridge Regression · Risk Overlay · Root Mean Square, RMS · R-Squared

S to Z

Savitzky-Golay Filter · Sharpe Ratio · Spectral Bandwidth · Spectral Entropy · Spot Price · Standardization · State Estimation · State Vector · Stationary Wavelet Transform, SWT · Step Response · Strike Price · Structural Alignment · Structural Repricing · Telemetry · Template Matching · Uncertainty Band · Vega · Volatility Center · Volatility Regime · Volatility Signal · Volatility Term Structure · Walk-Forward Validation · Warmup · Wavelet · Wavelet Energy · XGBoost · Z-Score

Adaptive Model

An adaptive model updates its internal state as new information becomes available.

A static model keeps the same learned relationship until it is trained again. An adaptive model can change gradually without requiring a full retraining process after every new observation.

Vyreon begins with a learned statistical relationship between options-derived features and forward return structure. A recursive state estimator then updates that relationship as matured realized outcomes become available.

Adaptive does not mean that the model changes without limits. The update speed is controlled. New observations influence the model, but they do not automatically replace everything learned before them.

Related Terms: Kalman Filter, Recursive State Estimation, Prior Estimate.

Amplitude Target

An amplitude target measures how large a market response became, without requiring the model to predict its direction.

Examples from the expiry research include:

Maximum absolute underlying move over days 1 to 3.
Mean absolute move over days 1 to 3.
Sum of absolute moves over days 1 to 3.

These targets answer a different question from a signed return target. They ask how much movement the market was capable of expressing after a known event.

Amplitude targets were more learnable than directional targets in the documented expiry experiments. This does not make them production forecasts. They remain research targets that require chronological and symbol-holdout validation.

API

An API, or application programming interface, is a defined way for software systems to request and exchange data.

The Vyreon backend exposes read-only API routes. Internal routes can return detailed payload, prior, matured, and derived records. Customer routes can return a smaller and more stable data contract.

An API is not the model itself. It is a delivery layer between stored data and another program, notebook, dashboard, or service.

The API uses scoped keys for access control. The raw secret is not stored in the database. A hash is stored instead.

Related Terms: Derived Layer, Canonical Payload, Data Lineage, Payload Hash.

Append-Only Event Stream

An append-only event stream preserves new records without silently replacing the earlier source records.

Vyreon treats the telemetry mailbox as an event stream. Each email records that a payload for a node and date arrived at a particular time.

Earlier messages remain available even when a later correction becomes canonical. This supports audit, replay, backfill, and debugging.

Append-only does not mean every record is equally current. The canonical-selection rule determines which valid payload is the present answer for a node and date.

Related Terms: Canonical Payload, Data Lineage, Idempotent Ingestion.

Approximation Energy

Approximation energy is the lower-frequency energy measured from the approximation coefficients of a wavelet decomposition.

In the eigenstructure research, approximation energy was calculated from the slower component of loading-rotation instability.

It was interpreted as a description of slower structural reorganization or regime consolidation. It tended to rise after more localized detail energy had expanded.

Approximation energy is a descriptive research measurement. It is not a directional price forecast and is not part of the current public production model.

Average Error

Average error summarizes the typical distance between realized returns and the expected mean.

In the validation chart, the error is measured in return space. An average error of 1.3% means the realized horizon-averaged return differed from the expected mean by about 1.3 percentage points on average under the displayed calculation.

Average error does not show direction by itself. A small average error can still hide a consistent positive or negative bias if signed errors are not also examined. It should be read together with the validation lines, interval coverage, and the visible shape of the errors through time.

Average error is not the same as trading profit or loss. It measures agreement between a model estimate and a later realized outcome.

Related Terms: Calibration, Realized Return, Coverage.

Backwardation

Backwardation describes a volatility term structure in which shorter-horizon volatility is higher than longer-horizon volatility.

A simple example is a short-dated volatility measure rising above a longer-dated measure during stress.

Backwardation often appears when near-term uncertainty is unusually high. It does not specify whether the underlying price will rise or fall.

Post-launch research found that telemetry transition events were associated with a higher frequency of later backwardation in the studied sample. That is a probability shift, not a guaranteed outcome.

Related Terms: Contango, Volatility Term Structure, Hazard Signal.

Baseline

A baseline is a reference level used to separate broad market drift from the part of return the model is trying to estimate.

The current production target uses a causal rolling mean of previously matured bucket returns. The baseline is calculated before the newest matured return is added to the history. This prevents the current outcome from influencing its own reference level.

The model estimates excess return relative to this baseline. The baseline is then added back to create the public total expected return.

A baseline is not a promise that the future will resemble the recent average. It is a reference frame. It helps the model focus on deviations from the market's recent horizon-specific return behavior.

Related Terms: Excess Return, Expected Return, Causal.

Bifurcation

A bifurcation is a change in the organization or stability of a system.

In Vyreon's historical intraday research, the term described a detected transition in multiscale market structure. The detector attempted to recognize when the prior state was breaking down.

A bifurcation signal does not determine which direction the new state will take. The retired 1DTE system repeatedly demonstrated that transition timing and directional resolution are different problems.

The term is retained for the historical research program. It is not a current production trade signal.

Bucket

A bucket is a group of option contracts with similar time remaining until expiration.

The current public system uses four buckets:

Near-term, 8 to 30 calendar days.
Short-term, 31 to 60 calendar days.
Medium-term, 61 to 120 calendar days.
Long-term, 121 to 365 calendar days.

Each bucket has its own feature state, regression model, recursive estimator, expected return, and uncertainty.

Buckets preserve time as a structural dimension. A contract expiring in two weeks does not describe the same market process as a contract expiring in nine months.

The labels used in public reports may describe the same buckets as approximately 1 month, 2 months, 3 months, and 8 months. Those labels refer to effective average horizons, not the exact upper boundary of the bucket.

Related Terms: Horizon, Effective Horizon, Expiry.

Calibration

Calibration measures whether the model's stated uncertainty agrees with realized outcomes over many observations.

A calibrated 95% expected range should contain realized outcomes close to 95% of the time under the validation method being used. It should not contain every outcome. A model that never misses may be using ranges that are too wide.

Calibration is different from accuracy. A model can have a central estimate that is not very precise while still representing uncertainty honestly. A model can also have a strong average prediction while using uncertainty bands that are too narrow.

Vyreon publishes calibration information because uncertainty is part of the model output, not an optional decoration.

Coverage should always be interpreted together with the exact scoring method. Adaptive filter coverage and strict issue-date forecast coverage answer related but different questions.

Related Terms: Coverage, Prediction Interval, Model Uncertainty.

Call Option

A call option is a contract connected to the right to buy an asset at a specified strike price before or at expiration, depending on the contract type.

In the Vyreon structure chart, call open interest shows the number of outstanding call contracts at each expiry.

Call open interest is not automatically bullish. A call can be bought or sold. It can be used for speculation, hedging, income, spread construction, or risk transfer. Open interest alone does not reveal which side initiated the position or why it exists.

The call share in the report is therefore a composition measurement. It describes inventory, not trader intent.

Related Terms: Put Option, Open Interest, Options Chain.

Canonical Payload

A canonical payload is the valid payload currently selected as the main answer for a node and date.

The backend can retain several payload versions for the same logical date. This can happen because of duplicate messages, migration replay, backfill, or an intentional correction.

Vyreon's current rule is that the latest valid payload by received_at becomes canonical. Earlier versions remain stored for audit.

Canonical does not mean historically first. It means currently selected under the defined resolution rule.

Related Terms: Append-Only Event Stream, Payload Hash, Data Lineage.

Causal

A calculation is causal when it uses only information that was available at the time of the estimate.

For example, a causal smoothing filter can use today's value and earlier values. It cannot use tomorrow's value to improve today's estimate.

Causality is essential in financial modeling. A historical result can look excellent if future information accidentally enters a feature, a normalization window, or a validation calculation.

Vyreon's production transformations use endpoint-based smoothing, exponential updates, delayed target maturation, and prediction-first recursive updates. Each calculation is intended to preserve the order in which information becomes available.

Causal does not mean correct. A causal model can still make poor estimates. It means the estimate was not built using future data.

Related Terms: Prior Estimate, Walk-Forward Validation, Maturation.

Centroid

A centroid is a weighted average location.

For an options distribution across strikes, the centroid summarizes where the selected weight is concentrated.

Different weights produce different centroids. Open interest produces a positioning centroid. Volatility or option sensitivities can produce other structural centers.

A centroid is a compact summary. It does not prove that price must move toward the centroid. It does not reveal who opened the positions or why they exist.

Coherence

Coherence describes how consistently the different horizons are behaving.

High coherence means several horizons show similar direction, similar evolution, or mutually supportive states. Low coherence means the horizons disagree, move at different speeds, or provide only partial confirmation.

For example, a positive long-term state with negative near-term behavior is not fully coherent. It may describe a short pullback inside a stronger long-term structure. It may also describe a transition that has not resolved.

Coherence is not a single price forecast. It is a summary of agreement across the maturity structure.

The weekly report may describe coherence as aligned, fragmented, conditional, or unresolved. These terms indicate how much confidence can be placed in a market-wide interpretation rather than one isolated horizon.

Related Terms: Horizon, Structural Alignment, Market State.

Confidence Interval

The public charts currently use the phrase confidence interval for the shaded expected ranges.

In strict statistical language, a confidence interval usually describes uncertainty around an estimated parameter or mean. A prediction interval describes the range in which a future realized outcome may fall.

The Vyreon bands are intended to describe plausible future realized horizon outcomes. For that reason, prediction interval or expected range is the more precise interpretation.

The glossary preserves the chart language because readers will see it in the legend. When reading the chart, treat the 68% and 95% bands as expected outcome ranges around the central mean.

A 95% interval is not a guarantee. Outcomes outside the range are expected to occur occasionally.

Related Terms: Prediction Interval, Expected Range, Coverage.

Constraint Field

Constraint field is Vyreon's term for the structural influence implied by the distribution of options inventory, volatility, and sensitivity across strikes and expiries.

The term does not mean that price is physically trapped inside a fixed boundary. It does not mean one strike must act as support, resistance, or a future target.

Instead, the term describes how options-related hedging and positioning can make some price movements easier, harder, more stable, or more volatile while the structure exists.

The field changes through time. Contracts expire. Open interest moves. Volatility is repriced. Gamma and vega concentrations shift. The market can therefore gain or lose degrees of freedom as the options matrix changes.

The production system measures parts of this structure through centers, widths, relative positions, and their evolution.

Related Terms: Options Matrix, Field Center, Field Width, Market Structure.

Contango

Contango describes a volatility term structure in which longer-horizon volatility is higher than shorter-horizon volatility.

It is often associated with calmer conditions than backwardation, but it is not a guarantee of market stability.

The exact definition depends on which volatility instruments or horizons are being compared.

Contango and backwardation describe the shape of the volatility curve. They do not directly provide price direction.

Related Terms: Backwardation, Volatility Term Structure.

Continuous Wavelet Transform, CWT

A continuous wavelet transform decomposes a time series across time and scale.

It can show when short, medium, or longer oscillatory structures become active. Unlike one global Fourier summary, the CWT preserves local timing.

Vyreon's historical bifurcation research used wavelet-derived measurements such as spectral bandwidth, entropy, phase behavior, and higher-order derivative structure.

CWT output can be sensitive to boundaries, wavelet choice, scale definitions, and noise. A visually interesting scalogram is not proof of predictive value.

Related Terms: Wavelet, Spectral Bandwidth, Spectral Entropy, Bifurcation.

Convolutional Neural Network, CNN

A convolutional neural network is a machine-learning model that searches for local patterns in structured arrays or sequences.

Vyreon's historical bifurcation detector used a lightweight one-dimensional CNN on an eight-channel time-series fingerprint.

The model was designed to recognize transition shapes such as curvature change, spectral expansion, and derivative sign structure.

The CNN was live-tested, but it did not solve the separate problem of direction. It is not part of the current public telemetry model.

Covariance Matrix

A covariance matrix describes uncertainty and co-movement across several variables.

The diagonal entries contain the variance of each variable. The off-diagonal entries describe how pairs of variables vary together.

In the production Kalman filter, the coefficient covariance is a full matrix. This allows uncertainty in one coefficient to be related to uncertainty in another coefficient.

In the eigenstructure research, a rolling covariance matrix describes how the four normalized horizon states move together.

Coverage

Coverage is the percentage of scored realized outcomes that fall inside a stated expected range.

PI95 coverage refers to the 95% range. If 98 of 100 scored outcomes fall inside the range, coverage is 98%.

Coverage above 95% can mean the model is conservative. It can also mean the intervals are wider than necessary. Coverage below 95% can mean the intervals are too narrow, the mean is biased, the residual distribution is not represented well, or the market changed faster than the model adapted.

Coverage should not be judged from a very small sample. It becomes more meaningful as observations accumulate across different conditions.

The validation method matters. A range scored immediately before an adaptive update is not identical to a range archived on the original issue date and scored after the full horizon matures.

Related Terms: Calibration, PI95, Prediction Interval.

Data Lineage

Data lineage records where a value came from and how it changed through the system.

For Vyreon, lineage can connect:

raw email
-> parsed payload
-> prior and matured rows
-> canonical selection
-> derived outputs
-> API response

Good lineage makes it possible to answer which model output existed for a date, whether it was corrected later, and which record was used in validation.

Data lineage is essential for strict issue-date validation. Without it, a later reconstruction can accidentally replace the forecast that was actually available at the time.

Days To Expiry, DTE

Days to expiry, or DTE, is the number of calendar days remaining before an option expires.

A contract with 0DTE expires on the current trading day. A 30DTE contract has about 30 calendar days remaining.

DTE affects option sensitivity, decay, and the role of the contract inside the maturity surface.

Near expiry, gamma can become highly concentrated around the current price. Longer-dated options usually carry more vega and less immediate gamma concentration.

DTE is a time measurement. It does not provide direction by itself.

Related Terms: Expiry, Gamma, Vega, Maturity Surface.

Delta

Delta measures how much an option's value is expected to change for a small change in the underlying price, all else equal.

A call delta is usually positive. A put delta is usually negative.

Delta also changes as price, time, and implied volatility change. Gamma measures how quickly delta changes when the underlying price moves.

Delta is a local sensitivity, not a guaranteed option-price change. Large moves, volatility repricing, and time decay can make realized option behavior differ from a simple delta estimate.

Related Terms: Gamma, Call Option, Put Option, Spot Price.

Derived Layer

A derived layer stores stable values calculated from canonical source records.

The Vyreon backend converts canonical priors into return-space and price-space outputs. Examples include expected mean, uncertainty, 95% range, expected price range, and innovation RMS.

The API can serve these derived values without recalculating the full model during each request.

Derived does not mean raw. It also does not mean untraceable. Each derived record should remain linked to the canonical source payload that produced it.

Related Terms: API, Canonical Payload, Data Lineage.

Detail Energy

Detail energy is the higher-frequency energy measured from the detail coefficients of a wavelet decomposition.

In the eigenstructure research, detail energy was applied to loading-rotation instability rather than directly to price.

It was interpreted as localized turbulence inside the changing cross-horizon structure. In several observed sequences, detail energy expanded before slower approximation energy.

Detail energy is a research diagnostic. It does not provide reliable direction by itself.

Dominant Eigenstructure

Dominant eigenstructure is the strongest shared organization mode inside a covariance matrix.

A covariance matrix can be decomposed into eigenvalues and eigenvectors. The largest eigenvalue identifies the direction that explains the greatest amount of joint variation. Its eigenvector describes that dominant direction.

In Vyreon's research, the covariance matrix was built from four normalized horizon states. The dominant eigenvector summarized how those horizon states were jointly organized inside a rolling window.

This is a structural description. It does not directly predict market direction.

Related Terms: Covariance Matrix, Eigenstructure Instability, State Vector.

Effective Horizon

The effective horizon is the average future time represented by a horizon-averaged target.

Vyreon samples several future calendar dates across each bucket and averages the returns to those dates. The average of those sampled dates gives the effective horizon.

The current approximate effective horizons are:

Near-term, 19 days.
Short-term, 46 days.
Medium-term, 90 days.
Long-term, 243 days.

The effective horizon is not the same as the maturation delay. The complete target cannot be calculated until the final sampled date in the bucket exists. A target with an effective horizon of 19 days still needs the 30-day sample before its full average is known.

Related Terms: Horizon-Averaged Forward Return, Maturation, Bucket.

Effective Matured Date

The effective matured date is a display date used on the validation chart.

It is calculated as the forecast issue date plus the average forward horizon represented by the bucket. This places the realized horizon-averaged outcome near the economic center of the future period it represents.

The effective matured date is not necessarily the date when the full target became available to the system. Full availability requires the upper boundary of the bucket to pass.

For example, the near-term target has an effective horizon near 19 days, but its final sampled return occurs at 30 days. The point can be displayed near day 19 while the complete label is only known after day 30.

Related Terms: Effective Horizon, Maturation, Validation.

Eigenstructure Instability

Eigenstructure instability describes rapid change in the dominant organization of the multi-horizon state.

The research measures this through loading rotation, which compares the dominant eigenvector with its previous value.

High instability means the strongest cross-horizon organization is changing quickly. Low instability means the organization is comparatively stable.

Historical event studies associated larger instability events with elevated later movement magnitude over several days. Direction remained inconsistent.

Research status: retained as a structural diagnostic, not deployed as a directional or execution signal.

EMA Z-Score

An EMA Z-score standardizes a value using an exponentially updated mean and variance.

The current value is compared with the statistical state that existed before the value was incorporated. The mean and variance are then updated for the next observation.

This creates a causal normalized signal. It answers a question such as:

How unusual is this measurement relative to its recent exponentially weighted history?

EMA smoothing and EMA Z-scoring are different operations. The first EMA smooths the measurement. The EMA Z-score tracks the evolving center and scale used for normalization.

Vyreon uses both because financial measurements can change in level and variance through time.

Related Terms: Z-Score, Standardization, Causal.

Event Study

An event study aligns many observations around a defined event and compares what happens before and after it.

For example, expiry research can set the expiry boundary as time zero, then calculate the average path of KOI, price, or another variable from several days before to several days after.

Event studies can show timing, average direction, persistence, and variation across events.

They do not automatically prove causality. Results can also be affected by overlapping events, dependence across symbols, changing regimes, and the event definition itself.

Excess Return

Excess return is the realized or expected return after subtracting the causal baseline.

The purpose is to remove part of the market's recent horizon-specific drift so the model can focus on structural deviation from that reference level.

The Ridge model and recursive estimator operate on excess return. The baseline is later added back to produce the total expected return shown publicly.

Excess return in this system is not necessarily the same as return above a risk-free rate, return above an index, or investment alpha. It is defined relative to Vyreon's own causal rolling bucket baseline.

This distinction is important because the same phrase can have different meanings in finance.

Related Terms: Baseline, Expected Return, Ridge Regression.

Expected Mean

The expected mean is the central estimate of the model's current return state for a horizon.

It is shown as the main line in the forward-expectations chart. It is also shown as the orange line in the validation chart.

A positive expected mean places the center of the current estimate above zero. A negative expected mean places it below zero.

The mean does not determine the final state classification by itself. The full 95% expected range matters.

A positive mean can still be Mixed if the lower edge of the expected range falls below zero. A negative mean can also be Mixed if the upper edge remains above zero.

The expected mean is not an exact target price. It is the center of a probability-aware return estimate.

Related Terms: Expected Return, Expected Range, Mixed State.

Expected Range

An expected range is the interval around the expected mean that represents uncertainty about a future outcome.

The forward-expectations chart shows a darker 68% range and a lighter 95% range.

A wider range means the model sees more possible outcomes. A narrower range means the estimate is more concentrated.

The width of the range is not a direct measure of bullishness or bearishness. A wide positive range can still be uncertain. A narrow negative range can be more concentrated but still negative.

The range is also used to classify a horizon as Positive, Negative, or Mixed.

Expected ranges should be evaluated over many matured observations. One outcome inside or outside the band does not establish calibration by itself.

Related Terms: Prediction Interval, Uncertainty Band, Calibration.

Expected Return

Expected return is the model's estimate of the average future return associated with the current market state and a specific horizon.

The public value combines two parts:

A causal rolling baseline based on previously matured returns.
An excess-return estimate produced from the current options-derived feature state.

Expected return is not the return that must occur. It is the center of an uncertain distribution.

The estimate is horizon-specific. A near-term expected return and a long-term expected return describe different future periods and can have different signs.

Expected return should not be interpreted as a recommendation to buy or sell. It is one component of the financial telemetry report.

Related Terms: Baseline, Excess Return, Horizon.

Expiry

Expiry, also called expiration, is the date when an option contract ends.

Time to expiry is a major dimension of the options matrix. Contracts with different expiries can represent different event windows, risk horizons, and hedging conditions.

Vyreon groups expiries into four production buckets. Weekly expiries are included because they improve time resolution in the shorter horizons.

As expiry approaches, option sensitivity can change quickly. Gamma can become concentrated near the current price. Expiration can also remove a large layer of open interest from the matrix.

For end-of-day research, the expiry-day snapshot can still contain the expiring contracts. The first observable snapshot without them is usually the next trading day. That distinction matters when measuring an expiry-driven change in the options field.

Expiry does not dictate price direction. It changes the structure and can increase or reduce the constraints acting on price.

Related Terms: Expiry Boundary, Bucket, Options Matrix, Gamma.

Expiry Boundary

The expiry boundary is the point between the final snapshot that still contains expiring contracts and the first snapshot that no longer contains them.

In the daily research data:

t = final snapshot containing the expiry
t+1 = first trading-day snapshot without it

This distinction corrected an important timing problem in the KOI research. Movement into the expiry-day close was not the same as the structural change caused by removal of the expiring inventory.

The observable post-expiry step is:

Delta KOI_post = KOI(t+1) - KOI(t)

This observed step is not fully pre-event because the t+1 snapshot can include new positioning.

Related Terms: Expiry, KOI, Open-Interest Centroid, Step Response.

Exponential Moving Average, EMA

An exponential moving average is a causal smoothing method that gives more weight to recent observations.

Unlike a simple moving average, an EMA does not assign equal weight to every value in a fixed window. Older information fades gradually.

Vyreon uses EMA smoothing after the causal Savitzky-Golay transformation. This reduces noise while preserving responsiveness to recent structural changes.

The current production feature pipeline uses the same EMA alpha across the four active buckets. This was selected empirically and reduces the number of free parameters.

The volatility-regime chart also uses an EMA to show the trend of raw innovation dispersion. In that chart, the raw line shows the newest movement and the EMA shows the broader direction.

Related Terms: Savitzky-Golay Filter, EMA Z-Score, Volatility Regime.

Feature

A feature is a numerical measurement used as an input to a statistical model.

Vyreon features are derived from the options matrix and recent price behavior. Examples include weighted strike centers, field widths, volatility-surface measurements, relative positions, trend coefficients, and normalized changes through time.

A feature is not automatically a trading signal. It can be useful only in combination with other measurements and the learned model state.

The production feature list is frozen in a specific order. The same order and preprocessing must be used in training and live operation. A mismatch would change the meaning of the model coefficients.

Exact feature weights and the complete internal feature specification remain proprietary.

Related Terms: Options Matrix, Ridge Regression, Standardization.

Field Center

A field center is the weighted center of an options sensitivity distribution across strike prices.

Vyreon uses gamma and vega to form a combined structural field. Contracts with more relevant sensitivity have more influence on the center.

The field center describes where this sensitivity mass is concentrated. It is not necessarily the current price. It is also not a guaranteed price target.

Changes in the field center can show that the options structure is moving relative to price or relative to the open-interest positioning center.

The center is most useful when read together with field width, current price, volatility center, and changes through time.

Related Terms: Field Width, Gamma, Vega, Constraint Field.

Field Width

Field width measures how spread out an options sensitivity distribution is across strikes.

A narrow field means sensitivity is concentrated in a smaller strike region. A wide field means the structure is distributed across a broader range.

Field width is not the same as an expected price range. It describes the geometry of the options sensitivity field, not the model's direct prediction interval.

A wide field can indicate that risk sensitivity is distributed across many strikes. A narrow field can indicate stronger concentration. Neither state is automatically bullish or bearish.

Field width is used as one part of the model's structural description.

Related Terms: Field Center, Market Structure, Expected Range.

Forecast Ledger

A forecast ledger is an immutable record of the exact forecast that existed on its original issue date.

A useful ledger stores the issue date, model identity, horizon, expected mean, uncertainty, and any later correction history.

The ledger allows a future realized outcome to be scored against the forecast that was actually available before the outcome occurred.

This is required for strict issue-date coverage. Reconstructing an old forecast from current state can produce a different answer.

Related Terms: Data Lineage, Coverage, Prior Estimate, Canonical Payload.

Gamma

Gamma measures how quickly an option's delta changes when the underlying price changes.

Delta describes directional sensitivity. Gamma describes how rapidly that sensitivity can change.

Gamma is often strongest for options near the current price and near expiration. This can increase hedging activity because a dealer's delta exposure can change quickly as price moves.

Vyreon uses gamma as a structural weight. It helps identify where directional sensitivity is concentrated across strikes and expiries.

Gamma does not reveal whether the market will rise or fall. It describes sensitivity and potential hedging responsiveness.

Related Terms: Vega, Field Center, Expiry.

Gamma share at 0DTE measures how much of the selected gamma structure belongs to contracts expiring on the current trading day.

A high share means a large part of measured gamma sensitivity is concentrated in the nearest expiry.

This can describe how much near-term hedging sensitivity may disappear across the expiry boundary.

The expiry research found that 0DTE gamma share was not a reliable linear amplitude predictor by itself. It may act as a gate or context variable, but larger values did not always produce larger underlying moves.

Related Terms: Gamma, Days To Expiry, DTE, Expiry Boundary, Amplitude Target.

Hazard Signal

A hazard signal indicates that the probability of a future condition has changed.

It does not necessarily predict the direction or exact timing of the outcome.

In post-launch research, large changes in the volatility signal were associated with a higher frequency of later stress conditions, including backwardation and short-horizon volatility stress.

That supports a hazard interpretation:

The distribution of possible future conditions changed.

It does not support a statement such as:

Price must fall next.

Related Terms: Volatility Signal, Backwardation, Structural Repricing.

Horizon

A horizon is the future time period represented by an estimate.

Vyreon reports four horizons rather than one combined forecast. Each horizon has a different bucket of option expiries and a different effective duration.

The horizons can disagree. Near-term structure can be negative while long-term structure is positive. This is not necessarily a contradiction. Different parts of the options market can represent different phases of the same broader transition.

A horizon should be matched to the question being asked. A near-term state is more relevant to the next few weeks. A long-term state describes a broader multi-month structure.

Related Terms: Bucket, Coherence, Effective Horizon.

Horizon-Averaged Forward Return

A horizon-averaged forward return is the average of several future returns sampled across one bucket.

For an origin date, the system measures return to several future calendar dates between the lower and upper bucket boundaries. Those returns are averaged into one target.

This reduces dependence on one exact terminal date. It describes the broader return path represented by the bucket.

For example, the 31 to 60 day bucket is not evaluated only at day 60. It averages several returns across the full range.

The complete average cannot be known until the final sampled date exists. This is why the maturation delay equals the bucket's upper boundary even though the effective average horizon is shorter.

Related Terms: Effective Horizon, Maturation, Realized Return.

Idempotent Ingestion

An ingestion process is idempotent when repeating the same input does not create an incorrect duplicate result.

The Vyreon backend can receive the same logical payload more than once. A payload hash helps identify exact duplicates. Alternate valid versions can still be retained.

Idempotency is important because live systems need retries, replays, and backfills. A safe retry should not corrupt the database.

Idempotent does not mean earlier source records are deleted. The append-only raw layer can preserve them while the parsed and canonical layers resolve duplicates correctly.

Related Terms: Payload Hash, Append-Only Event Stream, Canonical Payload.

Implied Volatility, IV

Implied volatility is the level of future uncertainty embedded in an option's market price under an option-pricing framework.

It is called implied because it is inferred from the option price rather than measured directly from past price movement.

Implied volatility can differ across strikes and expiries. This creates a volatility surface.

Vyreon measures the level, slope, fit quality, and structural center of this surface within each bucket.

Implied volatility is not a direct probability forecast and does not determine price direction. Higher implied volatility means the option market is pricing more movement or uncertainty, but it does not say whether that movement will be upward or downward.

Related Terms: Volatility Center, Realized Volatility, Options Matrix.

Impulse Response Function, IRF

An impulse response function shows the average path of a system after a defined input or event.

In the expiry research, the event was a scheduled change in options inventory. Series were aligned around the expiry boundary and measured relative to a pre-event baseline.

The IRF can show:

Delay before the response.
Direction of the average response.
Peak timing.
Persistence or decay.
Variation across events.

The expiry input is more accurately described as a step change than a mathematical impulse, but IRF remains common shorthand for the event-aligned response plot.

An IRF is not an individual-outcome prediction.

Related Terms: Step Response, Event Study, Expiry Boundary.

Initial Coefficient Covariance, P0

P0 is the covariance matrix used to describe uncertainty in the initial regression-coefficient state.

The coefficient values come from the trained Ridge model. P0 describes how uncertain those trained coefficients are and how that uncertainty is related across features.

The August 2026 audit replaced an arbitrary identity initialization with a covariance derived from the Ridge training fit.

A sensible P0 matters most during startup. It affects the first predictive intervals, Kalman gains, and early covariance evolution.

P0 is not a forecast range by itself. The current feature vector, process covariance, and observation variance are also required.

Innovation

An innovation is the difference between what the model expected before an observation and what was later observed.

In simple form:

innovation = realized outcome - prior expected outcome

A small innovation means realized behavior remained close to the prior model state. A large innovation means the market departed more strongly from that state.

Innovations are used in two ways. They update the adaptive model, and their magnitude becomes a measurement of current repricing instability.

An innovation can be positive or negative. The volatility signal focuses mainly on magnitude. A large positive innovation and a large negative innovation can both indicate strong departure from prior expectations.

Related Terms: Prior Estimate, Volatility Signal, Kalman Filter.

Innovation Dispersion

Innovation dispersion summarizes the recent magnitude of innovations across the system.

The Current Volatility Regime chart shows a raw innovation-dispersion line and a smoothed EMA line.

Raw dispersion reacts quickly to new surprises. The EMA shows whether those surprises are becoming more or less persistent.

High dispersion means realized behavior is departing more strongly from prior expectations. Low dispersion means the market is behaving more consistently with the current model state.

Innovation dispersion does not indicate direction. It measures alignment and repricing intensity.

The weekly report may describe the regime as expanding, compressing, elevated, or stable based on the relationship between raw dispersion and its trend.

Related Terms: Innovation, Volatility Regime, Volatility Signal.

Joseph Covariance Update

The Joseph covariance update is a numerically stable way to update a Kalman covariance matrix.

It uses both the state-update matrix and observation variance:

P_new = A P_prior A^T + R K K^T

where K is the Kalman gain and A represents the remaining state uncertainty after the observation is applied.

This form helps preserve symmetry and positive semidefiniteness under floating-point arithmetic.

It does not change the model's objective. It improves numerical reliability.

Related Terms: Kalman Filter, Covariance Matrix, Predictive Covariance.

Kalman Filter

A Kalman filter is a recursive method for updating an estimate and its uncertainty as new observations arrive.

Vyreon uses a multidimensional Kalman-style adaptive regression filter. The state contains a vector of regression coefficients. The covariance matrix describes uncertainty and relationships across those coefficients.

The current production filter begins from:

Ridge coefficients as the initial state.
A trained coefficient covariance called P0.
A process covariance called Q.
An adaptive observation or innovation variance called R.

For each matured observation, the filter follows prediction-first order:

Form prior covariance from the current covariance plus process covariance.
Read the observation variance that existed before the new outcome.
Calculate the prior expected value.
Calculate the innovation and full predictive variance.
Score and assimilate the observation.
Update the innovation-variance estimate for the next observation.

This order prevents the current error from deciding how surprising that same error was.

The filter does not make the model infallible. It provides temporal continuity, controlled adaptation, and uncertainty tracking.

KOI, Open-Interest Centroid

KOI is Vyreon's research shorthand for an open-interest-weighted strike centroid.

It summarizes where open option inventory is concentrated inside a selected expiry or group of expiries.

KOI_all refers to an aggregate centroid across a broader part of the options matrix.

KOI is a structural measurement. It is not proof of dealer direction, trader intent, support, resistance, or a guaranteed price attractor.

Expiry research studies how the aggregate centroid changes when known inventory expires. The observed post-expiry change is called Delta KOI_post.

Research status: KOI is retained as a structural feature. The expiry step-response program remains research and is not a validated trading strategy.

Related Terms: Centroid, Open Interest, Expiry Boundary, Positioning Center.

K-Star, K

K-star, written k*, is a research term for a center or basin derived from the implied-volatility structure across strikes.

The exact calculation belongs to the proprietary structural pipeline. Conceptually, it describes where the fitted volatility surface is centered under the selected weighting and aggregation.

In the expiry research, kstar_all behaved more like a smooth state descriptor than a clear forcing variable. It often moved contemporaneously with broader market repricing.

K-star is not a guaranteed equilibrium price, fair value, support level, or target.

Related Terms: Volatility Center, Implied Volatility, Centroid.

Live

Live refers to operation after the public production system began issuing daily measurements in real time.

The Live marker on a chart identifies the public launch boundary. Its meaning depends on the chart's x-axis.

On an issue-date chart, points after the marker were issued during live operation. On an effective-maturity chart, points to the right may represent earlier forecasts whose represented horizon extends beyond the launch date.

Live does not automatically mean that a full horizon has matured. A long-term forecast issued today may require up to one year before its complete realized target exists.

Related Terms: Maturation, Out-Of-Sample, Effective Matured Date.

Loading Rotation Speed

Loading rotation speed measures how quickly the dominant eigenvector changes from one observation to the next.

The research definition is:

loading_rotation_speed = norm(v1_t - v1_(t-1))

where v1 is the dominant eigenvector of the rolling covariance matrix across four normalized horizon states.

A high value means the dominant cross-horizon organization is changing quickly. A low value means it is changing slowly.

Historical tests associated larger rotation events with a higher probability of later movement, but not a reliable direction.

Market State

Market state is the system's current estimate of the structural conditions associated with price, options positioning, volatility, expected return, and uncertainty.

A market state is broader than one directional prediction. It can include:

A positive or negative central expectation.
Wide or narrow uncertainty.
Expanding or compressing innovation dispersion.
Agreement or disagreement across horizons.
Positioning and volatility centers relative to spot.

The state changes as new options data and matured return observations arrive.

The weekly report translates this multidimensional state into plain-language sections. The translation remains conditional. It does not claim that one state guarantees one future path.

Related Terms: State Estimation, Market Structure, Coherence.

Market Structure

Market structure is the organized pattern beneath the current price.

In Vyreon, it includes how open interest, implied volatility, gamma, vega, strikes, and expiries are distributed across the options matrix.

Two markets can have the same price and different structures. One may have concentrated near-term inventory and narrow risk sensitivity. Another may have dispersed long-term positioning and wider uncertainty.

Market structure helps explain the conditions under which price is moving. It does not provide a complete causal account of every market move. News, liquidity, macro events, and other forces can change price independently of the measured options structure.

Related Terms: Options Matrix, Constraint Field, Positioning Center.

Maturation

Maturation is the process by which a forward-looking target becomes fully observable.

When the model issues a forecast for an origin date, the future returns required to evaluate it do not yet exist. The system waits until the upper boundary of the bucket has passed.

For the four production buckets, the minimum full maturation delays are approximately 30, 60, 120, and 365 calendar days.

After maturation, the realized grid-mean return can be calculated. The result is then used to evaluate the prior state and update the adaptive model.

Maturation is different from the effective horizon. The effective horizon is the average represented time. Maturation waits for the final sample.

Related Terms: Horizon-Averaged Forward Return, Effective Horizon, Causal.

Maturity Surface

The maturity surface is the options structure observed across many expiration dates at the same time.

Vyreon's four production buckets are not independent markets. They are selected projections through one coupled maturity surface.

Each bucket has different sensitivity, inertia, noise, and repricing speed. All buckets still exist inside the same options matrix.

The public filter stops at 365 days, although some listed contracts extend farther. The active bucket boundaries were selected because they produced useful signal-to-noise behavior in the research.

Related Terms: Options Matrix, Bucket, Coherence, Horizon.

Mixed State

A Mixed state occurs when the 95% expected range crosses zero.

This means both positive and negative realized returns remain plausible within the model's current uncertainty range.

Mixed does not mean the expected mean is exactly zero. The center can be positive or negative while the wider range still includes both sides.

Mixed also does not mean the model has failed. It means directional certainty is not strong enough for the full 95% range to remain on one side of zero.

A Mixed state can become more informative when combined with trend direction, interval width, other horizons, and the volatility regime.

Related Terms: Expected Range, Expected Mean, Coherence.

Model Uncertainty

Model uncertainty describes how uncertain the estimator is about its current expected outcome.

In the production system, predictive uncertainty combines:

Uncertainty in the adaptive coefficient state.
Process noise that allows the coefficient state to evolve.
An adaptive estimate of innovation or observation variance.

This uncertainty is projected through the current feature vector to produce a predictive standard deviation.

Model uncertainty is not the same as every possible market risk. The model does not directly know future news, policy decisions, geopolitical events, data errors, or liquidity failures.

The uncertainty bands describe uncertainty under the model's current information and assumptions.

Related Terms: Predictive Covariance, Prediction Interval, Kalman Filter.

Moneyness, M-Space

Moneyness describes the position of an option strike relative to the current underlying price.

A strike can be in the money, at the money, or out of the money.

For cross-asset analysis, relative distance is more useful than raw dollars. One common m-space definition is log-moneyness:

m = log(strike / spot)

M-space makes strike distance more comparable across assets with different prices.

A movement across one strike boundary has different dollar size for different symbols. Relative moneyness helps normalize that geometry.

Related Terms: Spot Price, Options Matrix, Strike Price.

Observation Variance, R

R is the observation or innovation variance used by the adaptive Kalman filter.

It represents the amount of outcome variation not explained by the projected coefficient state.

Vyreon estimates R adaptively from innovations. The value used to score the current observation is frozen before that observation updates the variance tracker. The current innovation affects R only for the next observation.

A larger R produces wider predictive uncertainty and a smaller Kalman gain, all else equal.

R is not the same as realized volatility. It is variance inside the model's observation equation.

Open Interest, OI

Open interest is the number of outstanding option contracts that remain open.

It measures contract inventory. It is different from trading volume, which measures how many contracts traded during a period.

Vyreon uses open interest to calculate positioning centers and expiry composition.

Open interest does not reveal trader intent. A large put position can be a hedge, a speculative bet, part of a spread, or the other side of a volatility trade. A large call position can also have several meanings.

Open interest is therefore treated as a structural weight, not as a direct probability, sentiment score, or guaranteed price attractor.

Related Terms: Positioning Center, Call Option, Put Option.

Options Chain

An options chain is the list of available option contracts for one underlying asset.

Each row normally includes a strike price, expiration date, call or put type, and market measurements such as price, open interest, volume, implied volatility, and Greeks.

The chain is a snapshot. It shows the structure available at one time.

Vyreon filters the chain for valid contracts, groups it by horizon, and converts it into a smaller set of structural measurements.

The chain should not be read contract by contract as a collection of independent predictions. Its value comes from the organized distribution across strikes and time.

Related Terms: Options Matrix, Expiry, Feature.

Options Matrix

The options matrix is the options chain organized across strike price and expiration date.

Strike is the price dimension. Expiry is the time dimension. Call and put measurements add another structural distinction.

Each part of the matrix can contain open interest, volume, implied volatility, gamma, vega, and related information.

Vyreon measures this matrix as a field rather than treating one contract as a complete signal.

A single end-of-day matrix is a snapshot. A sequence of matrices through time shows how the field is building, decaying, migrating, and being repriced.

Related Terms: Options Chain, Constraint Field, Market Structure.

Out-Of-Sample

Out-of-sample refers to data that was not used to fit the model parameters being evaluated.

The Out-of-Sample marker on the validation charts identifies the boundary after which the displayed observations belong to the evaluation period rather than the original model-fitting period.

Out-of-sample testing reduces the risk of judging a model only on data it already saw during training.

It does not remove every risk. Results can still be influenced by overlapping horizons, regime concentration, target construction, repeated test windows, or later implementation changes.

The strongest evidence comes from several layers: historical out-of-sample testing, strict causal replay, live operation, and future issue-date cohorts that fully mature.

Related Terms: Walk-Forward Validation, Live, Calibration.

Parkinson Realized Volatility

Parkinson realized volatility estimates historical volatility using each day's high and low prices.

Close-to-close volatility uses only closing prices. Parkinson volatility uses more information from the daily trading range.

Vyreon's volatility validation chart compares the model-based volatility signal with both close-to-close and Parkinson realized volatility.

The two realized-volatility measures can disagree because they summarize price movement differently.

A strong correlation with both measures supports the interpretation that innovation magnitude is tracking changes in realized market variability rather than one narrow price calculation.

Payload

A payload is one structured telemetry emission from a model node for a specific date.

It can contain metadata, current priors, newly matured observations, symbol information, and a model identity.

The backend stores the payload as a durable record and links its child rows to it.

A payload is not automatically canonical. Several payload versions can exist for the same node and date.

Related Terms: Canonical Payload, Payload Hash, Data Lineage.

Payload Hash

A payload hash is a short digital fingerprint calculated from payload content.

If two payloads have the same content, they should produce the same hash under the same hashing method.

The backend uses hashes to identify exact logical duplicates while still preserving raw source messages.

A hash is not encryption and should not be treated as a secret. Its purpose here is identity and deduplication.

Phase Locking

Phase locking describes two oscillatory series maintaining a stable phase relationship.

In historical intraday research, phase-locking measurements were explored to test whether price and structural series were moving in a coherent rhythm.

A high phase-lock value can indicate synchronization. It does not prove that one series causes the other or that the relationship will continue.

Phase locking produced useful event context but was not retained as the current regime oracle.

Related Terms: Wavelet, Spectral Bandwidth, Coherence.

PI95

PI95 is a binary score for whether a realized outcome falls inside a 95% expected range.

For one observation:

PI95 equals 1 when the outcome is inside the range.
PI95 equals 0 when the outcome is outside the range.

Coverage is the average of these scores across many observations.

PI95 is meaningful only when the range was calculated before the outcome was used to update the uncertainty model. The production estimator now uses a prediction-first order for this scoring.

PI95 does not say whether the expected mean was accurate. An outcome can be far from the mean and still remain inside a wide interval.

Related Terms: Coverage, Calibration, Prediction Interval.

Positioning Center

The positioning center is an open-interest-weighted strike center.

Contracts with more open interest have more influence on the result. The center summarizes where outstanding contract inventory is concentrated across a selected group of strikes and expiries.

The structure chart displays a positioning center for each expiry and an overall positioning center across the classified matrix.

The positioning center is not a guaranteed support level, resistance level, fair value, or destination for price. It is a cross-sectional structural measurement.

Its relationship to spot and to the volatility center can still be informative. Large distances describe displacement inside the current options structure.

Related Terms: Open Interest, Volatility Center, Spot Price.

Prediction Interval

A prediction interval is a range intended to contain a future realized outcome with a stated frequency over many observations.

A 95% prediction interval is designed so that realized outcomes fall inside it close to 95% of the time under the model and validation method.

The interval combines the expected mean with predictive uncertainty.

A prediction interval differs from a confidence interval around a fitted parameter. The public charts currently use confidence interval in the legend, but expected range or prediction interval is the more precise reading for future realized returns.

Prediction intervals depend on model assumptions. They can fail during structural breaks, unusual shocks, data problems, or persistent bias.

Related Terms: Confidence Interval, Coverage, Model Uncertainty.

Predictive Covariance

Predictive covariance represents uncertainty in the coefficient state before a new observation is assimilated.

In the adaptive regression filter:

predictive covariance = current coefficient covariance + process covariance

The current feature vector projects this multidimensional uncertainty into return space. Observation variance is then added to produce the full predictive variance.

Using predictive covariance is important because the coefficient state may evolve between observations. The process-covariance term allows for that change.

The August 2026 estimator audit also showed why initialization matters. An arbitrary identity matrix was not a meaningful trained coefficient covariance. The production initialization was replaced with a covariance derived from the Ridge training fit.

Predictive covariance is used for scoring innovations, constructing expected ranges, and determining how strongly the filter responds to new information.

Principal Component Analysis, PCA

Principal component analysis is a method that transforms correlated variables into a smaller set of orthogonal components.

The first component explains the largest amount of variation. Later components explain additional independent variation.

Vyreon explored PCA in forecasting and state-analysis experiments. It also uses eigen decomposition conceptually in the cross-horizon eigenstructure research.

PCA can reduce dimensionality, but it can also hide the meaning of original features. Clean-looking PCA results did not consistently improve walk-forward behavior, so PCA was not retained as a universal solution in the public production path.

Related Terms: Covariance Matrix, Dominant Eigenstructure, Standardization.

Prior Estimate

A prior estimate is the model's prediction before the newest realized outcome is used to update the state.

The term prior refers to information order. It does not necessarily mean the estimate was created on the original forecast issue date unless the record explicitly comes from that date.

For one recursive update, the prior mean is calculated from the current coefficient state and the frozen feature vector. The innovation is the difference between the realized outcome and that prior mean.

After assimilation, the model has a posterior state for the next step.

Keeping prior and posterior quantities separate is essential for honest uncertainty scoring.

Related Terms: Innovation, Predictive Covariance, Kalman Filter.

Probability Mass

Probability mass is weight assigned to discrete possible states or locations.

In Vyreon's options research, strikes and expiries can be treated as a discrete weighted field. Open interest, implied volatility, gamma, vega, or other measurements can supply the weights.

After normalization, the weights can be summarized through centroids, width, curvature, and relative placement.

This is a modeling frame. It does not mean every options weight is a literal risk-neutral probability. The meaning depends on the selected weight.

Related Terms: Centroid, Options Matrix, Field Width, Constraint Field.

Process Covariance, Q

Q is the process covariance in the adaptive Kalman filter.

It represents how much the coefficient state is allowed to change between observations.

The current production system uses a diagonal Q, which adds a small amount of variance to each coefficient before a new observation is scored.

A larger Q allows faster adaptation but increases predictive uncertainty. A smaller Q produces a more stable state but can react slowly to genuine change.

Q is not fitted anew from each observation. It is a model hyperparameter derived from the chosen adaptation scale.

Put Option

A put option is a contract connected to the right to sell an asset at a specified strike price before or at expiration, depending on the contract type.

In the Vyreon structure chart, put open interest shows outstanding put inventory by expiry.

Put-heavy open interest does not automatically mean the market is bearish. Puts are widely used for protection, portfolio insurance, spreads, volatility trading, and regulatory or institutional risk management.

The call and put shares in the chart describe composition. They do not identify who bought or sold the contracts or what outcome they expect.

Related Terms: Call Option, Open Interest, Options Chain.

Realized Return

Realized return is the return that actually occurred after a forecast was issued.

In Vyreon validation, realized return is not one isolated end-date return. It is a horizon-averaged grid return calculated from several future dates inside the bucket.

The realized return becomes available only after the final sampled horizon has passed.

It is then compared with the expected mean and expected range.

Realized return is an observed outcome. Expected return is a model estimate made before the outcome exists.

Realized Volatility

Realized volatility measures how much price actually moved over a historical period.

It is calculated from observed prices. This differs from implied volatility, which is inferred from option prices and concerns expected future uncertainty.

The volatility validation chart uses 21-day close-to-close realized volatility and 21-day Parkinson realized volatility.

Vyreon's model-based volatility signal is standardized before comparison so the chart focuses on shape and timing rather than raw units.

Realized volatility describes magnitude of movement, not direction.

Recurrent Neural Network, RNN

A recurrent neural network is a machine-learning model designed for sequential data.

RNN variants such as LSTM and GRU maintain an internal memory that can represent relationships through time.

Vyreon's earlier forecasting program tested recurrent, convolutional, attention-based, and hybrid models.

Many produced plausible historical forecasts but did not generalize reliably under walk-forward testing. They are not part of the current public production model.

Recursive State Estimation

Recursive state estimation updates a model one observation at a time.

The estimator carries forward its previous state instead of refitting the entire model from the beginning after every new observation.

Vyreon's recursive state includes regression coefficients, their covariance, and an adaptive estimate of innovation variance.

When a matured return arrives, the estimator compares it with the prior prediction. The size of the surprise and the current uncertainty determine how much the coefficients change.

This creates temporal continuity. The model can adapt to changing relationships while remaining anchored to its learned structure.

Related Terms: Kalman Filter, Adaptive Model, State Estimation.

Residual

A residual is the difference between an observed value and a model estimate.

residual = observed value - estimated value

In a static regression, residuals describe training or test errors. In the adaptive filter, the prediction-first residual is called an innovation.

Residual analysis can reveal bias, changing variance, heavy tails, or model failure.

A residual is not automatically random noise. Persistent structure in residuals can indicate that the model is missing information.

Related Terms: Innovation, Average Error, Calibration.

Ridge Regression

Ridge regression is a linear regression method that penalizes very large coefficients.

The penalty reduces sensitivity to noisy or correlated features. This usually makes the model more stable than an unrestricted linear regression.

Vyreon trains a separate Ridge model for each horizon. The learned coefficients initialize the recursive estimator.

Ridge is linear. It does not directly learn every nonlinear market interaction. Its strength in this system is stability, interpretability, and compatibility with online coefficient updates.

The public methodology explains the role of Ridge, but exact coefficients remain proprietary.

Related Terms: Feature, Kalman Filter, Excess Return.

Risk Overlay

A risk overlay changes exposure based on a separate risk-state measurement.

It does not necessarily replace the main strategy or forecast. It modifies position size, leverage, or participation when conditions become more or less hazardous.

Post-launch research tested the innovation-based volatility signal as a modifier to a realized-volatility regime model. The documented improvement was small and exploratory.

A risk overlay can add value by reducing losses during unstable conditions, even when it does not predict direction.

Related Terms: Hazard Signal, Volatility Signal, Sharpe Ratio.

Root Mean Square, RMS

Root mean square summarizes the magnitude of values while treating positive and negative values symmetrically.

For values x_1 through x_n:

RMS = sqrt(mean(x_i squared))

Large positive and negative innovations both increase RMS.

Vyreon's volatility research uses RMS or closely related magnitude summaries to describe innovation dispersion across horizons.

RMS is not a directional measurement. It answers how large the recent deviations were, not whether they were upward or downward.

Related Terms: Innovation Dispersion, Volatility Signal, Hazard Signal.

R-Squared

R-squared measures how much variation in a target is explained by a model relative to a simple mean baseline.

An R-squared of 1 means perfect explanation in the evaluated sample. Zero means no improvement over predicting the sample mean. A negative value means the model performed worse than that baseline.

R-squared must be interpreted with the split used to calculate it. In-sample R-squared can be misleading. Out-of-sample R-squared is more relevant for generalization.

The expiry research reported promising held-out R-squared values for amplitude targets. Those results remain research and require stricter chronological and symbol-holdout tests.

Related Terms: Out-Of-Sample, Amplitude Target, XGBoost.

Savitzky-Golay Filter

A Savitzky-Golay filter smooths a signal by fitting a small polynomial inside a moving window.

It is useful when the shape of a signal matters. It can reduce noise while preserving local slopes, peaks, and turning behavior better than some simple averaging methods.

Vyreon uses an endpoint-based version. The polynomial is evaluated at the newest point in the window. This makes the filter causal because it does not use future observations.

The smoothed output is then passed through EMA smoothing and EMA Z-score normalization.

The filter requires a warmup window before it can produce a valid output.

Related Terms: Causal, Exponential Moving Average, Warmup.

Sharpe Ratio

The Sharpe ratio compares average excess return with return volatility.

A higher Sharpe ratio indicates more return per unit of measured variability under the selected sample and calculation.

Sharpe does not describe tail risk, liquidity, trading costs, or the stability of the result. A small sample or one unusual period can distort it.

Vyreon's research uses Sharpe only as one comparison metric in exploratory risk-overlay studies. It is not evidence that a strategy is production-ready.

Related Terms: Risk Overlay, Realized Volatility, Out-Of-Sample.

Spectral Bandwidth

Spectral bandwidth describes how broadly signal energy is distributed across frequencies or scales.

A narrow bandwidth means energy is concentrated in a smaller range of modes. A wider bandwidth means the structure is more dispersed across scales.

Vyreon's historical bifurcation work used bandwidth expansion as part of a transition fingerprint.

Bandwidth is sensitive to window length, preprocessing, and transform choice. It is a structural descriptor, not a guaranteed event predictor.

Spectral Entropy

Spectral entropy measures how evenly signal energy is spread across frequencies or wavelet scales.

Low entropy means energy is concentrated in a smaller number of modes. High entropy means energy is more broadly distributed.

In historical intraday research, changes in entropy were used to help describe the transition from organized structure to broader multiscale disorder.

Spectral entropy is not thermodynamic entropy. It is a statistical summary of an energy distribution.

Spot Price

Spot price is the current market price of the underlying asset.

In the public SPY report, spot is the current SPY price associated with the options snapshot.

The structure chart compares spot with positioning centers, volatility centers, and expected price ranges.

A center above or below spot describes relative placement. It does not by itself predict that spot will move toward or away from the center.

Spot is also used to convert return expectations into approximate price ranges for display.

Related Terms: Positioning Center, Volatility Center, Expected Return.

Standardization

Standardization converts measurements with different raw units into a common scale.

The production model uses frozen StandardScaler artifacts after the causal feature pipeline. Each input is transformed using the mean and scale learned during training.

This allows features such as strike centers, volatility measurements, widths, and trend coefficients to enter the same regression model without raw unit differences dominating the result.

Standardization must match training exactly. Using a different feature order, mean, scale, or preprocessing method would change the model input space.

Standardization does not make a feature predictive. It only creates a consistent numerical representation.

Related Terms: Z-Score, Feature, Ridge Regression.

State Estimation

State estimation is the process of inferring an underlying condition that cannot be observed directly as one simple number.

The market state is not equal to spot price, open interest, or volatility alone. It is inferred from several measurements and their relationships through time.

Vyreon estimates horizon-specific return states, uncertainty, coefficient states, and innovation behavior.

State estimation is common in engineering systems where observations are noisy and the true process changes gradually.

The output is always an estimate. It should be judged by calibration, stability, and usefulness rather than treated as direct access to a hidden truth.

Related Terms: Market State, Recursive State Estimation, Telemetry.

State Vector

A state vector is an ordered collection of measurements used to describe the current state of a system.

In the eigenstructure research, the state vector contains four horizon values:

[near mu/sigma, short mu/sigma, medium mu/sigma, long mu/sigma]

The vector preserves the simultaneous condition of the maturity surface.

A state vector is not one final market score. Its components can agree, disagree, or change at different speeds.

Related Terms: Maturity Surface, Dominant Eigenstructure, Coherence.

Stationary Wavelet Transform, SWT

A stationary wavelet transform decomposes a series into detail and approximation components without downsampling at each level.

Because the output keeps the original time alignment, SWT is useful for comparing local changes across time.

Vyreon's eigenstructure research applied an SWT with a Coiflet wavelet to loading-rotation instability.

The detail component described localized turbulence. The approximation component described slower reorganization.

The method remains a research diagnostic, not a production trade signal.

Related Terms: Detail Energy, Approximation Energy, Wavelet.

Step Response

A step response describes how a system reacts after an input changes to a new level.

In the expiry research, removal of known option inventory changes the aggregate options field between two daily snapshots. The input is therefore closer to a step change than a brief mathematical impulse.

A step-response analysis can measure delay, direction, peak response, persistence, and settling behavior.

The research found a persistent KOI reconfiguration and a smaller delayed underlying response. This remains system-identification research, not a completed trading rule.

Strike Price

The strike price is the price level specified by an option contract.

A call option is connected to the right to buy at the strike. A put option is connected to the right to sell at the strike, subject to the contract terms.

The options matrix organizes contracts across strike price and expiry.

Strike spacing differs by asset, expiry, and price region. This is one reason relative moneyness is useful for cross-asset comparison.

Related Terms: Call Option, Put Option, Moneyness, M-Space, Options Matrix.

Structural Alignment

Structural alignment describes agreement among different measurements or horizons.

Examples include:

Several horizons moving in the same direction.
Positioning and volatility centers changing together.
Realized behavior remaining close to prior expectations.
Innovation dispersion falling while expected states remain stable.

Fragmentation is the opposite. It means different horizons or measurements are providing conflicting information.

Alignment can increase confidence in a broad interpretation. It still does not guarantee a future market move.

Related Terms: Coherence, Market Structure, Innovation Dispersion.

Structural Repricing

Structural repricing is a material change in how the market's measured state is organized.

It can involve changes in expectations, uncertainty, volatility, cross-horizon relationships, or options geometry.

Structural repricing is broader than one large price candle. A transition can begin through fragmentation or instability before a large displacement becomes visible.

The term does not specify direction. It describes reorganization of the measured market environment.

Related Terms: Bifurcation, Eigenstructure Instability, Market State.

Telemetry

Telemetry is the remote measurement and reporting of a changing system.

Vyreon uses the term because the system behaves more like an instrument panel than a traditional price-tip service.

It measures:

Options market structure.
Horizon-specific expected return states.
Predictive uncertainty.
Innovation magnitude.
Calibration and model alignment.

Telemetry describes what the system is measuring. It does not guarantee that a reader can convert every state into a profitable trade.

The weekly report is a human-readable translation of the telemetry.

Related Terms: Market State, State Estimation, Volatility Signal.

Template Matching

Template matching compares a live signal shape with a predefined pattern.

The historical bifurcation system used Gaussian pulse templates across a transformed signal and its first and second derivatives.

The deterministic template gate was combined with a learned CNN classifier. This reduced some false detections by requiring several representations to agree.

Template matching can improve selectivity, but it can also miss events that do not resemble the chosen pattern.

Related Terms: Bifurcation, Convolutional Neural Network, CNN, Feature.

Uncertainty Band

An uncertainty band is a shaded region around an expected mean.

The width of the band is determined by predictive uncertainty. The forward-expectations chart shows both 68% and 95% bands. The validation chart shows the 95% band used for coverage assessment.

A narrow band means the current model distribution is concentrated. A wide band means more outcomes remain plausible.

Uncertainty bands are not hard boundaries. Price and realized returns can move outside them.

The band should be read together with the expected mean, zero line, horizon, and calibration history.

Related Terms: Expected Range, Prediction Interval, Model Uncertainty.

Vega

Vega measures how strongly an option's value responds to a change in implied volatility.

Options with more vega are more sensitive to volatility repricing.

Vyreon uses vega to help measure the center and shape of volatility-sensitive structure across strikes and expiries.

Vega is usually larger for longer-dated options and contracts near the current price, although the exact pattern depends on the option and market conditions.

Vega does not indicate price direction. It describes sensitivity to implied-volatility change.

Related Terms: Gamma, Implied Volatility, Volatility Center.

Volatility Center

The volatility center is a strike-level summary of where the implied-volatility-sensitive structure is concentrated.

The public structure chart displays a volatility center for each expiry and an overall volatility center.

This center is derived differently from the positioning center. It represents volatility and sensitivity structure rather than open-interest inventory alone.

The distance between spot, the positioning center, and the volatility center describes current structural placement.

The volatility center is not a guaranteed fair value or future price target. It should be interpreted as one part of the current options geometry.

Related Terms: Positioning Center, Implied Volatility, Field Center.

Volatility Regime

Volatility regime describes the current pattern of market variability relative to the model's recent state.

Vyreon's current regime chart is based on innovation dispersion rather than only price volatility or VIX.

The report commonly distinguishes:

Expanding, raw or smoothed innovation magnitude is rising.
Compressing, current innovation magnitude is falling relative to its trend.
Elevated, the smoothed level remains high even if the newest raw value has declined.
Stable, innovations remain relatively contained and consistent.

A volatility regime does not provide direction. Expansion can occur during a rally or a selloff. Compression can accompany a calm rise, a calm decline, or sideways movement.

Volatility Signal

The volatility signal is a model-based measurement of innovation magnitude and alignment between realized market behavior and prior expectations.

It is built from model innovations after filtering and state estimation. A root-mean-square aggregation is one way the research summarizes innovation magnitude across horizons.

The signal rises when the market produces larger surprises. It falls when realized behavior stays closer to the current estimated state.

The validation chart compares the smoothed volatility signal with realized volatility. The series are standardized so the comparison focuses on timing and shape rather than raw units.

The volatility signal is not VIX. It is not implied volatility. It is not a directional forecast. It is a measurement of repricing intensity within the model.

Research after launch found that the level behaves like a volatility-state measurement. Changes in the level may contain transition or hazard information, but they have not produced a validated stand-alone trading strategy.

Volatility Term Structure

The volatility term structure compares volatility across different future horizons.

For example, short-dated implied volatility can be compared with three-month implied volatility.

The curve can be upward sloping, downward sloping, or uneven.

Contango and backwardation are two common descriptions of its shape.

The term structure measures how uncertainty is distributed through time. It does not directly predict price direction.

Related Terms: Backwardation, Contango, Implied Volatility.

Walk-Forward Validation

Walk-forward validation simulates repeated historical deployment through time.

A model is trained on an earlier window. It is then evaluated on a later window. The process moves forward and repeats.

This is more realistic than randomly mixing all dates because financial relationships change through time.

A valid walk-forward process must preserve causality. Feature transformations, scalers, baselines, model fitting, and recursive updates must use only information available before each scored outcome.

Walk-forward results are one layer of evidence. They should be combined with live operation and strict issue-date validation as future cohorts mature.

Related Terms: Out-Of-Sample, Causal, Calibration.

Warmup

Warmup is the historical period used to populate stateful transformations and adaptive model states before public output begins.

The causal Savitzky-Golay filter needs a full input window. EMA and EMA Z-score states need time to stabilize. The rolling baseline needs enough matured returns. The recursive estimator needs observations to adapt from its Ridge prior.

Vyreon performs a fresh warmup when the node launches under the current operating configuration.

During early warmup, some features, baselines, or matured scores may be unavailable. Zero outbound telemetry rows during this phase do not necessarily indicate a failure.

A successful warmup should produce populated feature states, matured updates, coverage records, and stable current priors.

Related Terms: Savitzky-Golay Filter, EMA Z-Score, Maturation.

Wavelet

A wavelet is a localized mathematical function used to analyze structure across time and scale.

Wavelets can separate short-lived high-frequency changes from slower background movement while retaining information about when those changes occurred.

Vyreon used wavelets in several historical research programs, including direct forecasting, intraday bifurcation detection, and eigenstructure-instability analysis.

Wavelet results depend on wavelet family, scale, boundary handling, preprocessing, and causal implementation. The presence of a visible pattern does not establish predictive value.

Wavelet Energy

Wavelet energy summarizes the squared magnitude of wavelet coefficients over a selected time, scale, or level.

Higher energy means more signal activity is concentrated in that component.

Detail energy emphasizes faster local variation. Approximation energy emphasizes slower structure.

Wavelet energy is scale-dependent. Values from different transforms or normalizations should not be compared without care.

Related Terms: Detail Energy, Approximation Energy, Wavelet.

XGBoost

XGBoost is a gradient-boosted decision-tree model.

It builds many small trees in sequence. Each tree attempts to reduce errors left by the previous trees.

XGBoost can represent nonlinear relationships and interactions without requiring a hand-built formula.

Vyreon's expiry research used XGBoost to test whether pre-event state could estimate movement amplitude. Promising held-out results were reported for amplitude targets, while directional targets performed poorly.

This work remains research. It requires chronological, symbol-holdout, stability, and execution testing before deployment.

Related Terms: Amplitude Target, R-Squared, Out-Of-Sample.

Z-Score

A Z-score expresses how far a value is from a reference mean in units of standard deviation.

A Z-score near zero means the value is close to the reference center. A positive Z-score is above the reference mean. A negative Z-score is below it.

A value near plus 2 is about two standard deviations above the reference mean under the chosen normalization. This does not automatically mean the event is rare in financial data because market distributions often have heavy tails and changing variance.

Vyreon uses causal EMA Z-scores for feature normalization and standardized series in the volatility validation chart.

Z-scores make shapes comparable across measurements with different raw units.

Related Terms: EMA Z-Score, Standardization, Innovation.

A Final Reading Rule

No single term should be interpreted alone.

A positive expected mean can still be Mixed. A low volatility signal can exist beside wide long-term uncertainty. Put-heavy open interest can exist inside a positive return state. A strong long-term horizon can coexist with weak near-term behavior.

The purpose of the system is to combine these measurements into a coherent market-state report while preserving uncertainty and disagreement.

For the complete workflow, continue to the Methodology. For visual examples, see How To Read The Charts. For common practical questions, see the FAQ.