Subscribe

Overview

Signals are not a black box.

Every score, every flag, and every commentary note produced by the ATGL engine traces back to a defined set of rules — applied consistently, on a published schedule, against a documented rubric. Nothing about how a stock is evaluated is hidden, arbitrary, or subject to someone’s judgment on a given morning.

The engine works in two layers. An automated workflow — running on a defined schedule — retrieves stock quotes, financial data, and sentiment data for every stock in the evaluation universe. That data is then passed to an LLM, which analyzes it against the published rubric for the relevant strategy and produces a score, a bias label, and a plain-language metrics breakdown.

This is not a fully autonomous AI system making independent decisions. It is a systematic, rules-governed process where automation handles data retrieval and the LLM handles analysis — both operating within the boundaries set by the rubric and the rules set. The rules are the authority. The AI is the analyst.

This page explains how the engine works: the two AI models, the rubric that governs evaluation, the rules that enforce discipline, and the scoring scale that translates everything into a single, actionable number.

Investing with rules isn’t just a tagline. It’s the architecture of everything this platform produces.

The two AI models — Athena and Apollo

The Signals engine runs on two distinct AI models, each serving a different role in the evaluation process. Understanding what each one does — and why both exist — is central to understanding what a Signal actually means.

Athena

Athena is the primary daily scanning model. She runs the morning scan every trading day before market open and produces the weekly trade report for Standard members.

Athena’s focus is speed and breadth: analyzing a wide universe of stocks, identifying setups that match the strategy’s rubric criteria, and producing a clear, ranked output before the trading day begins. Her evaluation covers pattern recognition, momentum signals, volume analysis, and sector context — all calibrated to the strategy and time horizon defined for the job she is running.

Athena’s output for each stock includes a signal score, a bias label (Bullish / Neutral / Bearish), and a metrics breakdown showing exactly which criteria fired and what the analysis found.

Apollo

Apollo is the advanced analytical engine, available exclusively to Premium members. Where Athena covers the market broadly and quickly, Apollo goes deep on the stocks that matter most.

Apollo’s analysis incorporates Athena’s output as one input and layers additional evaluation on top — producing the full ATGL SignalScore™ breakdown with factor-level commentary. Apollo’s report gives Premium members not just a score but a structured explanation of every factor that contributed to it.

The Apollo vs Athena comparative analysis — also exclusive to Premium — shows where the two models agree and where they diverge. A stock that Athena rates as Bullish but Apollo rates as Neutral is itself a signal worth paying attention to.

Why two models?

Athena provides speed and daily coverage. Apollo provides depth and comprehensive analysis. Together, they serve different needs at different points in the decision-making process: Athena for the morning scan and initial candidate selection, Apollo for the detailed evaluation before committing capital.

The comparative output between them also provides something no single model can: a built-in check on its own conclusions. When two independently structured evaluations agree, the signal is stronger. When they diverge, that divergence is information.

The rubric

Every stock evaluated by the Signals engine is scored against a rubric — a defined set of weighted criteria that determines what a good setup looks like for a specific strategy.

Rubrics are not universal. Each strategy on the platform has its own rubric, built around that strategy’s goals, risk parameters, and time horizon. What makes a strong signal for a one-week momentum trade is not the same as what makes a strong signal for a swing trade held over several weeks. The rubric is where those distinctions are encoded, so the engine evaluates each stock in the right context rather than applying a generic template to everything.

What the rubric evaluates

Rubric criteria fall into several broad categories:

Each criterion carries a defined weight in the overall score. The weight reflects how much that criterion matters for the specific strategy being evaluated — momentum may carry more weight for a short-term trade, while sector strength may carry more weight for a longer hold.

Rubric versioning

Rubrics evolve over time as the engine is refined and as market conditions change what makes an effective signal. Every change to a rubric increments its version number, and the change is logged. Historical scores remain permanently tied to the rubric version that produced them, so performance can be reviewed accurately over time — knowing exactly which criteria were in effect when each score was generated.

Rubrics and backtesting

The same rubric used for live scoring can be applied to historical market data to evaluate how the criteria would have performed in past conditions. This backtesting capability is central to validating the rubric’s effectiveness and to refining criteria over time. Results from backtesting studies will be published as the engine matures.

The rules set

Alongside the rubric, each strategy operates under a defined rules set — a list of conditions that govern which stocks are eligible for evaluation in the first place.

Where the rubric scores a stock, the rules set determines whether the stock is even considered. Rules can be hard disqualifiers that automatically remove a stock from output, or soft warnings that flag a stock for caution without removing it entirely.

Hard disqualifiers vs soft warnings

A hard disqualifier is a condition the strategy considers unacceptable regardless of how well the stock scores on other criteria. If the disqualifier condition is present, the stock does not appear in the output — no exceptions, no manual override.

A soft warning is a condition the strategy considers elevated risk. The stock remains in the output but is flagged, and the warning appears in the metrics column so members can weigh it in their own decision-making.

A concrete example — earnings proximity

For the Top Weekly Choice strategy, any stock with an earnings announcement within five trading days is a hard disqualifier.

The reason is straightforward: the Top Weekly Choice is a one-week hold. An earnings announcement within that window introduces binary risk — the kind of sudden, unpredictable move that can instantly reverse even the strongest technical setup. No signal score, however high, changes that risk. So the rule removes the stock entirely rather than leaving it in the output with a flag that a member might ignore under the pressure of a compelling score.

This is what it means for a rule to have teeth: it enforces the strategy’s discipline automatically, at the engine level, before the output ever reaches a member. The judgment has already been made by the rules, so the member does not have to make it under pressure.

Rules prevent the kind of discretionary override that is often well-intentioned but consistently costly. See why in the real-world case study → Why Rules Matter

Trade interval and time horizon

One of the most important — and least obvious — variables in trade signal evaluation is time horizon. The same stock, evaluated against the same rubric, can produce a meaningfully different score depending on how long the trade is intended to be held.

This matters because different signals carry different weights at different time horizons. A stock with strong support, high sentiment, and slowing momentum might be a well-supported trade over three weeks — there is time for momentum to recover. The same stock as a two-day trade is a different proposition: if momentum is slowing now, there may not be time for the setup to fully develop before the position needs to close.

How the engine handles time horizon

Every Signals job runs with a defined trade interval — the expected holding period for the strategy — built directly into the job configuration. This is passed to the AI model alongside the rubric criteria and metric data, so the evaluation and commentary always reflect the relevant time frame.

The result is that scores are not generic. They are calibrated to the strategy’s holding period. A signal scored for a one-week hold is a different evaluation from the same stock scored for a same-day trade, because the engine has been told what kind of trade it is evaluating.

A real example

During development of the Signals engine, a stock was evaluated without a time horizon specified. The engine returned a score of 4 out of 5 — strong support, high sentiment, multiple positive criteria. Then the same stock was re-evaluated with an explicit short-term horizon of one to five days. The score changed to 3 out of 5.

The reason: slowing momentum, which the engine had noted but not heavily weighted without time context, became a more significant concern once the holding period was defined as short. Over one to five days, there is no time to wait for momentum to recover. The engine correctly downweighted the setup once it understood the trade’s time frame.

This is why trade interval is a first-class input in every Signals job — not an afterthought, not a prompt engineering note that someone has to remember to include, but a defined field that governs every evaluation, every run, automatically.

The ATGL SignalScore™ scale

Every stock evaluated by the engine receives a SignalScore™ — a numeric value from 1 to 5 — alongside a plain-language label that translates the number into an actionable description.

Score Label What it means
5 Very Bullish Multiple strong criteria align. High-confidence setup for the strategy and time horizon. The engine sees broad agreement across technical, momentum, volume, and pattern signals.
4 Bullish Most criteria are favorable. A good setup with manageable risk factors. One or two signals may be mixed but the overall picture is positive for this strategy.
3 Neutral Mixed signals. No strong directional conviction in either direction. The setup does not meet the threshold for a confident bullish or bearish evaluation.
2 Bearish More criteria are unfavorable than favorable. Elevated risk for this strategy at this time horizon. Caution is warranted.
1 Very Bearish Multiple negative criteria align. High risk for the strategy and time horizon. The engine sees broad negative agreement across evaluation factors.

The SignalScore™ is not a price target or a buy/sell recommendation. It is a structured, rules-based evaluation of how well a stock currently fits the defined criteria for a specific strategy. Members should use it as one input in their own decision-making process alongside the metrics breakdown and any relevant context the engine flags.

Update schedule

Signals reports run on a defined schedule. Every report page displays a “Last updated” timestamp so members always know exactly how current the data is.

All updates are automated. The engine runs on schedule regardless of market conditions, without manual intervention. When a report runs, the previous version moves to the archive and the new version becomes the active display.

Why rules matter

The Signals engine and the rubric it runs on exist because of a simple, repeatable problem: when human judgment overrides a systematic process — even with good intentions — the cost compounds quickly and quietly.

Here is what that looks like in practice.

A real example

A position was opened under the Top Weekly Choice strategy with a starting balance of $42,656.56, buying 125 shares at $341.06. By the end of the first week, the stock was down roughly 5%. The rules said to close. The position was held instead, because did not want to realize a visible loss.

The stock kept falling. The position was held through a second week, then a third, then a fourth. By the time it was finally reviewed, the stock had fallen to $303 — an 11.2% loss from the entry price — with a total position value of approximately $37,875.

The visible loss was $4,757. But that was only part of the cost.

A weekly rotation strategy compounds across many short cycles. Every week capital is sitting in a position that the system would no longer select is a week that compounding does not happen. Modeling the alternative — take the 5% loss in week one, redeploy into the top-ranked pick for each subsequent week at the strategy’s typical return — produces an ending balance of approximately $43,830 by week four. The gap between the two outcomes was nearly $6,000, or about 14% of the starting balance. Most of that gap was not the original loss. It was three weeks of compounding that never happened.

The discipline is not about any single trade being right or wrong. It is about making sure the strategy’s actual edge — consistent weekly rotation — never gets quietly switched off. Read the full case study →

This is why the rules engine exists. Every disqualifier in the rules set, every scoring criterion in the rubric, and every trade interval parameter in the job configuration is there to prevent exactly this kind of drift — automatically, without requiring anyone to make the right call under pressure.

Our transparency commitment

Transparency is not a feature we add on top of the Signals engine. It is the reason the engine was built the way it was.

Here is what that commitment means in practice:

Investing with Rules means the rules are visible.

⚠️ We're upgrading your experience.

Our site is currently being updated. You may notice some changes over the next week 

What to expect during this period:

  • Some pages may load slowly or look slightly different as updates are rolled out
  • Occasional brief interruptions may occur — we’ll do our best to minimise these
  • Your trading signals and core member resources will remain accessible throughout
 
Please reach out if you need any assistance. Thank you for your patience!