The Alert Fatigue Problem
Most stock alert platforms share a fundamental design flaw: they treat every news item with approximately equal urgency. A press release announcing a new VP hire gets the same delivery priority as an FDA drug approval for a company with a single-asset pipeline. A routine earnings announcement with results in-line with estimates arrives with the same formatting as an earnings beat of 40% above consensus. The result is alert fatigue — traders tune out the alerts because 90% of them lead nowhere, and they miss the 10% that matter.
Signal scoring solves this problem by quantifying the probability that a given event will produce a significant price move before the alert is delivered. The TMS (Trade Momentum Score) is TradeAI News's implementation of this principle: a 0–100 composite score calculated by a 6-layer AI engine, assigned to every detected signal, that determines both the delivery tier and the confidence level communicated in the alert.
What the TMS Measures
The TMS is not a price target and not a buy/sell recommendation. It is a probability-weighted estimate of signal quality: the likelihood that this specific catalyst event, detected at this specific moment, in these specific market conditions, will produce a price move of 3% or more in the indicated direction within the next 1–4 hours. Higher TMS scores reflect higher confidence in a significant move; lower scores reflect weaker evidence or conflicting signals.
The score is calculated fresh for every event. There is no manual override, no editorial weighting, and no recency bias — the same algorithm runs on every signal regardless of how exciting or boring the headline sounds.
Layer 1 — Historical Pattern Matching
The foundation of the TMS is a comparison against historical outcomes. When a new signal arrives, the system identifies the most similar events from the 1,860+ labeled historical events in the database, weighted by similarity across 12 dimensions: catalyst type, company size, sector, session timing, options flow state, market regime, and others. The historical outcomes of these analogs — what percentage produced a 3%+ move, what the average move size was, how quickly it developed — form the prior probability that the current event is significant.
This layer answers the most fundamental question: "Has something like this happened before, and what happened?" If the current event closely resembles 20 historical events of which 18 produced significant moves, the baseline TMS contribution from this layer is high. If the analogs are mixed or the event type is rare, this layer contributes less certainty.
Layer 2 — Velocity Score
The velocity score measures how fast information about this event is spreading across monitored sources. When a high-impact event occurs, the number of sources reporting on it and the rate at which new sources pick it up accelerates rapidly. This acceleration is measurable within the first 60–120 seconds of detection and is highly correlated with the magnitude of the eventual market reaction.
A catalyst that fires on a single source and does not propagate to additional sources within 2 minutes is less likely to produce a sustained move than one that appears across multiple primary sources simultaneously. Velocity is not just a measure of publicity — it is a measure of how many market participants are receiving actionable information, which directly determines the demand side of the initial price impulse.
Layer 3 — Options Intelligence
The options layer integrates data from the options market for the affected ticker. Four sub-signals are evaluated: sweep detection (large, aggressive options orders crossing the ask), IV ratio (current implied volatility versus its 30-day historical average), put/call ratio (the balance between bearish and bullish options activity), and open interest changes (new positioning building in specific strike clusters).
When options market activity aligns with the catalyst direction — call sweeps and IV expansion on a bullish catalyst, put sweeps on a bearish one — the score contribution from this layer is high. When options activity contradicts the catalyst or shows no unusual activity, this layer reduces rather than adds to the composite score.
Layer 4 — Market Context
The same catalyst event produces different outcomes in different market environments. This layer evaluates three contextual dimensions: sector momentum (is the sector the stock belongs to in a bullish or bearish regime today?), macro environment (VIX level, current market regime classification), and related ticker behavior (are peers moving in the same direction, confirming the catalyst, or are they contradicting it?).
An FDA approval in a biotech sector that has been in a downtrend for three weeks will produce a weaker and shorter-lived move than the same approval in a sector with positive momentum. The market context layer adjusts the TMS accordingly — it does not eliminate the signal, but it calibrates the expected move magnitude for current conditions.
Layer 5 — Timing Model
Catalyst events that fire at different times of day have systematically different move profiles. Pre-market FDA approvals (before 9:30am ET) produce different patterns than after-hours earnings releases, which differ again from intraday catalysts during the trading session. The timing layer applies historical calibration for when in the trading day this catalyst was detected, adjusting the expected reaction window and move magnitude accordingly.
Pre-market catalysts, for example, have an extended window for price adjustment before the regular session opens, meaning institutional participants can position for 2–4 hours before retail volume arrives. After-hours catalysts often show an initial spike followed by a fade in overnight trading before finding true equilibrium at the open. The timing model captures these structural differences and incorporates them into the score.
Layer 6 — Semantic Pattern Matching
The final layer compares the text of the current signal against the 1,860+ historically labeled events using semantic embedding similarity. Unlike keyword matching (which would classify "FDA grants approval" and "FDA authorizes clearance" differently), semantic matching identifies the conceptual similarity between signal texts regardless of exact wording.
This layer catches nuances that simpler classification misses: a press release that is technically an "analyst upgrade" but uses language closely associated with historical M&A rumors; an earnings release whose text pattern strongly resembles historical "blow-out beat" events versus "in-line beat" events. Semantic similarity adds a final calibration layer that refines the classification from the rule-based catalyst type assignment.
The Three TMS Tiers
WATCH (55–67): Worth monitoring. The signal has meaningful evidence of potential movement but lacks full multi-factor confirmation. Appropriate for watchlist placement, not immediate action. This tier fires frequently — multiple times daily. SEND PREMIUM (68–81): Strong signal with multi-factor confirmation. PRO and ELITE subscribers receive immediate alerts. The balance of evidence favors a significant move. These fire several times per week. SEND NOW (82+): Highest-conviction signals. All subscriber channels alert immediately. Multiple independent signals are converging. These fire on the order of 3–5 times per week on average, with higher frequency during active catalyst periods (earnings seasons, major regulatory decisions).
What TMS Is Not
The TMS is not a guarantee. A TMS 90 signal can and does occasionally produce no move or a move in the wrong direction — markets are probabilistic, and any scoring system has a false positive rate. The score indicates confidence in the signal, not certainty about the outcome. It is also not a price target or duration forecast — it measures the probability of a significant move, not how large that move will be or how long it will last. Those parameters depend on the specific catalyst type, which is communicated separately in each alert.
Frequently Asked Questions
How is the TMS score calibrated?
The calibration is ongoing. Each completed signal — where the outcome (move size, direction, duration) is known — is added to the historical dataset and used to refine the layer weights. The system is retrained periodically as new data accumulates, particularly when the distribution of catalyst types or market regimes shifts significantly.
Why does the same ticker sometimes get different TMS scores on similar news?
Because the market context differs. Options activity, sector momentum, and macro environment change continuously. Two similar news events on the same ticker can produce different TMS scores if the options market is showing unusual activity for one and not the other, or if sector momentum has shifted between the two events.
Can TMS be gamed or front-run?
Not in a way that degrades signal quality. The TMS is calculated from objective, measurable data (options flow, dark pool volume, news velocity, historical analogs). The inputs cannot be fabricated at scale without also creating the real market conditions they represent — which means the signal is genuine if the inputs are genuine.
How does TMS perform across different market regimes?
Higher TMS scores (82+) show relatively consistent performance across regimes because they require multi-factor confirmation — the convergence criterion itself acts as a regime filter. Lower tier signals (55–67) are more sensitive to regime, with lower win rates in high-VIX, risk-off environments where even genuine catalysts fail to sustain directional moves.
What happens when a SEND NOW signal fails?
Failed signals — where the catalyst fired but the expected move did not materialize — are logged, analyzed for common characteristics, and used to refine the model. The most common failure modes are: the catalyst was already priced in (the stock had already moved before the alert), conflicting macro pressure overwhelmed the individual catalyst, or the catalyst was less significant than initially classified. These failure modes inform ongoing calibration.