Methods & Data

Factor model cards

Purpose, implemented transformation, output, and limitation for each grade input.

Definitions

Eligible issues

NYSE issues equal advances plus declines plus unchanged issues. High, low, advance, decline, and unchanged counts must all be present and non-negative; issues must be positive; and highs and lows cannot exceed issues.

Raw trigger

A completed session qualifies when both new highs and new lows are at least 2.2% of eligible issues, new highs are no more than twice new lows, the NYSE McClellan Oscillator is below zero, and the NYA close is above its close 50 trading observations earlier.

Confirmation event

A raw trigger becomes a confirmation event when at least one earlier raw trigger occurred in the preceding 35 trading observations. This is the second qualifying trigger in a causal 36-observation cluster.

Active cluster

The active state is true when a confirmation event occurred in the current or preceding 35 trading observations. Raw trigger, trailing trigger count, confirmation event, and active state are stored separately.

Causality and availability

Every rolling rule uses current and prior observations only. Missing, invalid, or misaligned breadth, oscillator, or actual NYA price data fails closed and withholds the affected output rather than treating the condition as false evidence.

HLLI distinction

The High Low Logic Index is a separate breadth-fracture measure: 100 times the smaller of the new-high and new-low shares, smoothed with a 50-session exponential moving average. It is not an additional Hindenburg trigger condition.

Definitions

Volatility Tsunami static condition

Calculate 20-session population standard deviations of daily VIX and VVIX closes. The published static condition is true only when VIX dispersion is at or below 0.86 and VVIX dispersion is at or below 3.16.

Adaptive extension

RankAlpha separately compares each 20-session dispersion with its 15th percentile over up to 2,520 prior sessions, requiring 756 prior observations. Thresholds use a one-session shift, so the current observation cannot set its own threshold.

Tsunami events

The raw joint condition, its false-to-true cross, and a de-clustered event are distinct fields. A de-clustered event requires no joint condition in the prior 10 sessions. The adaptive variant is a labeled RankAlpha extension.

Canary active peak

For SPX and DJIA closing prices, a 52-week high is the 252-session rolling closing high. The latest equal high resets the active peak, which remains in force until the first 5% decline or a new high.

Canary first breach

The first close at least 5% below the active peak is classified once per peak cycle. A breach within 15 sessions is CANARY. A slower breach is BUY_THE_DIP only when SMA50 is strictly above SMA200; otherwise it records no uptrend or unavailable trend evidence explicitly.

Canary confirmation

A CANARY confirms after two consecutive closes below SMA200 from the signal through session +42. If no confirmation occurs, it expires only after the complete observable window; missing moving-average evidence produces INCOMPLETE_DATA instead of an unsupported negative conclusion.

Canary clustering

Every first-breach event is retained. Events within 21 trading sessions of the preceding event for the same index are marked non-primary rather than deleted.

Role and attribution

These indicators are informational market-risk observations. They do not alter grades, portfolio construction, sizing, risk limits, or broker execution. RankAlpha is not affiliated with or endorsed by Andrew Thrasher or the CMT Association.

Definitions

Rotation score

A 0-100 cross-sectional score equal to 100 times earned points divided by available points. The current point budget is relative strength 30, breadth 25, acceleration 20, and directional volume 5.

Score coverage

The component point budget available for that sleeve, not the percentage of symbols with prices. The currently supported components can supply at most 80 coverage points.

Relative strength

A sleeve return relative to SPY. Industry sleeves must also establish leadership relative to their parent-sector ETF. Sectors and industries are ranked in separate peer sets.

Breadth

When the constituent census is eligible, the model averages the shares above the 20-, 50-, and 200-day moving averages and the shares with positive 5- and 20-session returns, then ranks that result within the peer set.

Acceleration

A blend of percentile-ranked relative trend (35%), relative acceleration (25%), rank improvement (25%), and persistence (15%).

Rank change

Prior rank minus current rank. A positive value means the sleeve moved closer to rank 1. Comparisons are withheld when the peer universe or basket construction is not like-for-like.

Participation

A breadth label describing whether gains are broadly shared, moderate, narrow, or dominated by a few large constituents.

Confidence

A descriptive blend of data-quality confidence (60%) and the strongest qualified or top-ranked sleeve's distance from a neutral 50 score (25%). No additional component fills the remaining 15%, so the current measure cannot exceed 85 and is not a probability of gain.

Status

Leading is 70 or above; emerging is 55-69.99; neutral is 40-54.99; deteriorating is 25-39.99; breakdown is below 25.

Report classification

Confirmed or emerging rotation requires qualified recipient or donor evidence. Broadening means participation is widening without a former-leader break; narrow leadership means concentrated strength; defensive risk-off and temporary mean reversion are separate guarded states; no material rotation means none qualified.

Definitions

Breakout pivot

The prior closing high from the longest qualifying 120-, 60-, or 20-session window. A detected event remains above its pivot through the current observation, and the pivot is frozen while that event lineage is active.

Price gate

The latest adjusted close is above the frozen closing-price pivot, with no intervening closing loss inside the active lineage.

Relative-strength gate

The sleeve-to-SPY ratio is above its frozen relative pivot. An industry sleeve must also clear the equivalent ratio against its parent-sector ETF.

Participation gate

At least eight constituents and 70% constituent coverage, at least 55% above the 50-day average, positive median and equal-weight 20-session returns, and moderate or broad participation.

Persistence gate

At least two consecutive closes above the price pivot.

Breakout score

A percentile score normalized over available evidence. Point budgets are absolute price 25, relative strength 20, breadth 20, volume 5, verified flows 5, volatility 10, fundamentals 10, and regime 5. The score is withheld below 50 available points.

Confidence

A report-level blend of data-quality confidence (70%) and average score coverage among sector and industry sleeves with a published breakout score (30%). Data-quality confidence itself blends proxy coverage (65%) and breadth availability (35%). It is not a probability of gain.

Pivot distance

Latest close divided by the frozen pivot minus one. ATR distance scales that gap by 14-session adjusted true range.

Extension state

At pivot is below 0 ATR; early stage is 0-0.99; confirmed-not-extended is 1-2.49; extended begins at 2.5. Divergence can mark potential exhaustion from 3 ATR or the 95th distance percentile; severe extension begins at 4 ATR or the 97.5th percentile.

Base building / no actionable signal

Fallback presentation states when no higher-priority state qualifies. Base building requires its separate consolidation predicate; otherwise the sleeve is no actionable signal.

Failed breakout

A lineage-level state. Price must first lose its frozen pivot; relative strength must also break and breadth must deteriorate before the event is classified as failed.

Definitions

Rotation axis

The independent sector-rotation status and score for the sleeve.

Breakout axis

The independent breakout stage, score, frozen event lineage, and validity state for the sleeve.

Joint compatible

Both axes are usable and refer to the same stable sleeve design and economic coordinate. Matching dates or similar labels alone are not sufficient.

Confirmed leadership

Rotation is leading and breakout is confirmed breakout or confirmed leadership.

Joint emerging

Rotation is emerging or leading and breakout is emerging, confirmed breakout, or confirmed leadership, without meeting the stronger confirmed-leadership rule.

Aligned setup

Rotation is emerging or leading while breakout is base building or a pre-breakout setup.

Rotation only / breakout only

Only one axis is in a positive state. The label does not treat the other axis as confirming evidence.

Extended / deteriorating

Extended means the breakout is mature-extended or at exhaustion risk. Deteriorating includes weak rotation, a failed breakout, or a terminal or pending-failure lineage.

MFE / MAE

Maximum favorable excursion and maximum adverse excursion during the measured outcome window. Excess versions compare the path with SPY or the parent-sector ETF.

Evidence status

Insufficient is below the early thresholds. Early requires at least 20 completed observations, 10 unique signal dates, and 10 independent windows. Established requires 50, 20, and 20 respectively.

Origin and scope

Live forward contains signals observed after a registered activation. Reconstructed current manifest is a separately scoped diagnostic replay. Origins are never pooled, and retired or pending scopes remain distinct from the current active scope.

Confidence interval

A mean-outcome interval from a date-clustered circular block bootstrap. The resampling unit is the actionable-session cross-section and block width equals the outcome horizon, reducing false precision from overlapping observations.

Definitions

Effective dimensions

The source-published spectral estimate of how many independent patterns remain in the observed correlation structure. Higher values indicate more differentiated movement; lower values indicate more concentration.

Leading common mode

The fraction of total modeled movement represented by the largest eigenvalue. It measures concentration, not expected return.

Correlation communities

Groups of eligible nodes that the source study identifies as moving together. The count is withheld when sampling and community-identification requirements are not met.

Positive / negative density

The share of possible network edges classified as positive or negative by the source study.

Adjusted Rand index (ARI)

Agreement between two community partitions after adjusting for chance. 1 means identical grouping, values near 0 are chance-like, and negative values indicate less agreement than expected by chance.

Edge Jaccard

Intersection divided by union for two edge sets. 1 means identical topology and 0 means no shared edges.

Stable-node fraction

The share of eligible nodes placed consistently across the named perturbation or sampling replications.

Membership coverage

Source-declared coverage of applicable assets observed for the graph. Modeled coverage is the source-declared share of the membership represented by eligible modeled nodes. RankAlpha validates bounds and count consistency but does not recompute either fraction.

Perturbation / sampling stability

Perturbation checks vary the graph around the fixed observation; sampling checks vary the underlying observations. They test different sources of repeatability and are reported separately.

Node-set similarity

Intersection divided by union for the previous and current eligible-node sets. It is not the retained percentage of either set.

Membership churn

The source-published fraction of node-to-community membership that changed between identified adjacent observations.

Sealed observation

An immutable publication tied to source manifest, payload, generation, and source-revision hashes. Corrections append an explicit superseding observation.

Definitions

Sector Shock Reproduction Number

The source-published expected number of secondary sector shocks associated with one new sector shock in the fitted directed network. It is an expected reproduction count, not a probability.

Directed edge

A non-negative fitted source-to-target weight. The source sector is the shock origin and the target sector is the next-session shock recipient represented by that edge.

Criticality

A source-published diagnostic describing an edge's contribution to the fitted network. RankAlpha displays the value without treating it as statistical significance or recomputing it.

Quiet-cross variant

A separately identified model variant whose exact market-state conditioning is sealed in the observation configuration. It must not be pooled with the all-state-cross variant.

Weakly identified source

A source-level warning that the fitted outgoing relationships do not meet the source study's preferred identification quality. The network remains descriptive and the warning is retained.

Configuration scope

Variant and configuration hash jointly identify a comparable series. History is never mixed across configuration hashes.

Non-computable observation

An observation whose required price history, shock evidence, or optimizer result was insufficient. SRN, nodes, and edges are withheld instead of manufacturing a zero.

Sealed lineage

An immutable publication tied to source generation, manifest, payload, price receipt, and source-revision hashes. Corrections append an explicit superseding observation.

Definitions

Recovery asymmetry

Negative-shock restricted mean time to retracement over 20 sessions minus the matched positive-shock value. Positive values mean negative shocks retained more unrecovered time; negative values mean they recovered faster.

Overlap support

A regularized sign-propensity model uses sector and continuous shock magnitude. Overlap weights emphasize comparable negative and positive events, then sign-balance each sector before equal-sector aggregation.

T50 gap

Negative-shock median recovery time minus positive-shock median recovery time. It is withheld when either cohort does not reach 50% recovery inside the observed horizon.

Fast persistence gap

Negative minus positive probability that the same-sign residual remains persistent five sessions after the shock.

Fragility candidate

A research label requiring a positive recovery gap, an elevated historical percentile, the configured fast-persistence minimum, and the configured number of qualifying sessions in its persistence window. It is not an alert.

Lane support

The 20-session and five-session lanes qualify independently as adequate, degraded, or insufficient. A partial observation retains the supported lane and withholds only unsupported metrics.

Sector breadth

The count of supported sector sleeves with positive or negative 20-session recovery asymmetry. Sector results retain per-lane support and effective event counts.

Configuration scope

Methodology, topology, origin variant, and configuration hash jointly identify a comparable series. History and percentiles never mix incompatible configurations.

Sealed lineage

An immutable publication tied to source generation, manifest, payload, price receipt, source revision, and locked historical authority. Corrections append an explicit superseding observation.

Definitions

Drawdown Absorption Budget

The market-weighted average of constituent absorption log ratios after the frozen winsorization rule. Positive values mean the modeled historical-volume absorption proxy exceeds modeled stress demand overall; negative values mean demand exceeds the proxy.

Historical-volume-memory proxy

A model derived from completed daily price and volume history. It is not live order-book depth, executable size, a bid, or a forecast of transaction costs.

Stress bands

The frozen 1%, 2.5%, and 5% market drawdown scenarios. Each band carries its own absorption, expected-demand, net-budget, deficit-share, and severe-deficit-share diagnostics.

Deficit weight share

The final universe weight whose constituent absorption log ratio is below zero. Severe deficit uses the source contract threshold below -1.

Coverage weight

The selected capped universe weight represented by included constituents. It is not a confidence probability.

Risk percentile

A source-published historical rank for the observation under its exact configuration. RankAlpha displays it without treating it as a probability of drawdown or loss.

Configuration scope

The configuration hash identifies a comparable series. Observations from different hashes are discovered separately and are never pooled in the chart.

Sealed lineage

An immutable publication tied to source generation, manifest, payload, bar receipt, universe receipt, and source-revision hashes. Corrections append an explicit superseding observation.

Signals and screeners

These Ideas workflows answer different questions and retain distinct ordering contracts.

Data and freshness

Freshness is scoped to a subject. Current, delayed, stale, and unavailable states describe the named module, not the entire application. Pipeline cadence is an expectation, not proof that a particular run succeeded.