Disclosure

Methodology & Data

Effective September 20, 2026 · v3

GET UP STRATEGIES is an independent equities research environment. Everything it publishes is informational and is not investment advice, not a recommendation to buy or sell any security, and not a solicitation. You are solely responsible for your own decisions. This page explains where our data comes from, how fresh it is, what our signals do and do not claim, and how our optional AI features are bounded.

Data sources

Freshness & latency

• Quotes and snapshots may be delayed, and when a live stream is unavailable the platform falls back to periodic REST polling — labeled as such in the interface.
• Outside regular hours, a value is shown against its own basis (for example a prior close) and timestamp; the platform rejects provider sentinels rather than presenting them as a live price.
• Historical bars are end-of-session and may be adjusted or revised by the provider.
• Macro figures are official values as published on each agency's schedule and can be revised by the source.
• Any figure may be delayed, incomplete, or unavailable. Freshness state is surfaced in the product rather than assumed.

What the signals mean

Scores combine technical indicators (moving averages, multi-horizon returns, RSI), catalyst/news analysis (entity centrality and event typing, so tangential mentions do not drive a ticker's direction), macro regime context, and non-directional dark-pool context. Signals are decision support, not conclusions.

Composite score specification

Each component is normalized to a range from −100 to +100. The current v2 composite is the weighted average below. Positive values are bullish evidence and negative values are bearish evidence; the magnitude is evidence strength, not an expected return or probability of profit.

Formula: 0.35×trend + 0.22×momentum + 0.16×volume/liquidity + 0.17×macro + 0.10×news. For example, component values of +50, +30, −20, +10, and −10 produce a composite of +21.6. That remains Neutral: Strong Bullish begins at +65, Bullish at +25, Bearish at −25, and Strong Bearish at −65. The default discovery universe also requires a price of at least $3 and average dollar volume of at least $5 million unless an explicit ticker request applies a different research path.

Validation and execution firewall

A directional label is not permission to trade. The platform only marks a setup execution-eligible when an exact strategy-version, configuration, direction, setup-family, and horizon cohort passes every validation and live-risk gate.

• The fixed 60-outcome minimum belongs to descriptive research statistics; it is not the canonical execution threshold. A canonical family member requires at least 140 prospectively eligible signals with exact completed coverage.
• Results use forward, next-bar entry and include spread, slippage, market impact, commission, and short-borrow costs. Current validation defaults are 5, 8, 2, and 0.5 basis points respectively, plus 300 annualized basis points for short borrow.
• Promotion requires a positive clustered lower confidence bound, benchmark-relative alpha, a profit factor of at least 1.10, point-in-time provenance, walk-forward evidence, and an untouched holdout.
• Execution promotion uses one prospectively registered, closed cohort family and one frozen look per member, with Benjamini–Yekutieli false-discovery-rate control under arbitrary dependence. Open-batch Benjamini–Hochberg results remain research diagnostics only. Negative out-of-sample evidence explicitly demotes a setup.
• Canonical publication is currently unavailable and fails closed with bootstrap_calibration_required: no approved reproducible null-calibration artifact is committed, and the frozen candidate window is structurally too short. Research summaries continue but cannot authorize execution.
• A range printed beside a validation cohort is a moving-block bootstrap range at the registered level. It is not a calibrated confidence interval. Measured against the platform's own null scenarios, it excludes zero more often than its level allows when there is no edge, most of all at 5- and 20-session horizons, which is one reason promotion stays closed. Historical-replay studies use a different interval: Student's t over groups of consecutive sessions, with the horizon left out between groups so that no two groups share a forward return, widened for skew. At a nominal one-sided 2.5% its measured error is about 2% on average and about 3% at worst across those scenarios when sessions are independent, and about 4% for strongly left-skewed overlapping returns. It is not reliable when daily session averages are themselves strongly autocorrelated, where the measured error reaches 7%.
• Supported measurement horizons are scanner-specific and include 15m, 30m, close, 1d, 5d, and 20d. Dark-pool prints remain context-only because aggressor direction is not observable.
• Live entry is blocked for stale or unverified quotes, excessive spread, closed-session conditions unless explicitly allowed, missing risk inputs, daily-loss limits, portfolio-exposure limits, or an unvalidated evidence cohort.

Version and change policy

Scoring or gate changes produce a new strategy identity or configuration hash. Historical outcomes from a different identity are not silently pooled with the active model. Methodology revisions are dated on this page; material scoring, cost, horizon, or promotion changes require regression tests and a fresh validation cohort before they can regain validated status.

AI briefings & their limits

• AI summaries are optional and rely on third-party models that may be unavailable, delayed, truncated, or wrong.
• They consume the same underlying evidence and inherit the same freshness limits; they do not add new market facts.
• They are a research aid, never advice. Verify anything material against the primary source before acting.

Glossary

The console links each of these terms here. Plain words first; the technical name is in brackets.

Earnings drift (PEAD)

Post-earnings announcement drift: a stock's tendency to keep moving in the direction of an earnings surprise for days or weeks after the report. The Lab measures it after costs. It is research, not a trading signal.

Earnings surprise (SUE)

Standardised unexpected earnings: how far a quarter's earnings per share landed from the same quarter a year earlier, divided by how much this company's surprises usually vary. Above zero means better than that baseline.

Return beyond the market (abnormal)

The stock's return minus the part the market (SPY) explains for it, scaled by the stock's own sensitivity to the market (beta).

Score accuracy (IC)

The information coefficient: how well a score put stocks in order of their next returns, from -1 to 1. Near 0 means no ordering skill. Small positive values, held across many days, already matter.

Likely range (bootstrap)

A range for the average result, built by resampling past outcomes in blocks. It shows how far the average could move by chance. It is not a confidence guarantee, and with few outcomes it looks narrower than it is.

Group (cohort)

The signals judged together: the same scanner, strategy version, horizon and rules. Results are never mixed across groups.

Price data from different times (mixed timing)

The inputs behind a result were not all observed at the same moment, for example today's quote with yesterday's volume. Read it as a lead to check, not as a measurement.

Auction imbalance (NOI)

Net order imbalance: the excess of buy or sell orders waiting for the NYSE opening or closing auction. It needs a separate market-data entitlement.

Attribution

Market data is provided by Massive. Macroeconomic data is retrieved from FRED (Federal Reserve Bank of St. Louis), the U.S. Bureau of Labor Statistics, and the U.S. Bureau of Economic Analysis. All product names, logos, and trademarks are the property of their respective owners and are used for identification only.