Allan's Market Lab

How It Works · Version 2.2

Allan's Market Lab monitors unusual stock declines, investigates possible causes using a two-stage AI research process, and then compares its original recovery forecasts with what actually happens.

01

Market Monitoring

Allan's Market Lab monitors more than 500 stocks in the S&P 500 during regular U.S. market hours. The scanner runs approximately every five minutes and evaluates short-term, hourly, and daily price movement.

Individual stock performance is also compared with SPY. This helps distinguish a stock that is falling with the broader market from one that is underperforming the market by an unusual amount.

The system separately monitors broad market conditions and can identify a Market Selloff, Severe Market Selloff, or Market Crash when SPY and a large percentage of monitored stocks decline together.

02

DEEP Detection

When a stock meets the configured decline and relative-to-market thresholds, a DEEP is recorded immediately. The event stores the detection price, approximate pre-drop price, recent price changes, SPY performance, and the initial event low.

The website immediately displays the event and Telegram sends a DEEP DETECTED notification. At this stage the system does not pretend that the cause is already known.

03

AI Analysis In Progress

A newly detected DEEP initially appears as AI ANALYSIS IN PROGRESS.

This separation is intentional. Detection happens immediately, while cause analysis waits for additional information instead of forcing an explanation during the first moments of a decline.

The first Telegram alert also clearly states that AI analysis is still in progress and that a completed analysis will follow.

04

One-Hour Initial Research Window

After a DEEP is detected, Allan's Market Lab allows approximately one hour for explanatory reporting, company announcements, filings, and other relevant information to begin appearing.

The Initial Research window includes approximately four calendar days before detection and one hour after detection. The lookback helps capture catalysts that may have appeared before the market opened, during after-hours trading, or before a weekend.

The purpose of this delay is not to wait for the stock price to recover. It is to give the system a better information set before producing its first explanation and forecast.

05

Initial News & Luna Research

At the end of the Initial Research window, the system collects the available Alpaca/Benzinga news associated with the event.

Luna then performs an independent web research pass using selected financial, regulatory, company, and established news sources. The research stage looks for evidence that could explain the decline and builds a structured evidence packet.

The packet can include the likely catalyst, evidence quality, key findings, research gaps, and supporting sources. Evidence that falls outside the permitted event window is excluded from the Initial Analysis.

Alpaca/Benzinga articles are also assigned a relevance score based on factors such as company focus, timing, catalyst language, and whether the headline directly describes the stock move. Generic market roundups, historical performance stories, passive-income articles, and similar low-value items receive strong penalties.

The public Related News section only displays articles with a relevance score of 20 or higher. If the news feed contains items but none are sufficiently relevant, the website can show No closely related news has been identified yet. The relevance score is a filtering signal and does not by itself prove that an article caused the decline.

06

Initial Sol Analysis

Sol receives the market data, Alpaca/Benzinga news, and the Luna research packet. Sol does not perform its own additional web search during this stage.

It evaluates the supplied evidence and produces the Initial AI Analysis, including classification, cause confidence, catalyst type, evidence quality, what happened, why the stock may be falling, positive factors, risk factors, and recovery reasoning.

Positive Factors and Risk Factors are displayed directly on the alert card. This makes it easier to see which evidence supports a possible recovery and which evidence could prevent or weaken it, without changing the underlying frozen Initial Analysis.

Individual stock classifications include TEMPORARY, FUNDAMENTAL, DANGER, and UNCLEAR. Broad market declines are handled separately as MARKET events.

07

Initial 3D / 7D / 20D Recovery Forecast

The Initial Analysis also produces separate recovery forecasts for 3, 7, and 20 trading days.

These percentages measure recovery of the original decline rather than the future percentage return of the stock. A value of 0% represents no recovery from the event low, while 100% represents a return to the approximate pre-drop price.

Values below 0% are possible if the model expects the stock to fall further, while values above 100% represent a forecast beyond the original pre-drop level.

08

Initial Analysis Is Frozen

Once the Initial AI Analysis is marked COMPLETE, its classification, reasoning, and 3D / 7D / 20D forecasts are permanently preserved.

Later headlines, price movements, and hindsight do not rewrite the original assessment. This creates a fixed historical prediction that can later be measured against both the Deep Review and the actual market outcome.

At this point the website updates the original DEEP card and Telegram sends the AI ANALYSIS COMPLETE notification.

09

24-Hour Extended Evidence Collection

Research does not stop after the Initial Analysis. For the remainder of the first 24 hours following detection, the system continues collecting newly available Alpaca/Benzinga news and preserves that material for the later Deep Review.

This collection is incremental. The system checks periodically for new material rather than repeatedly downloading the entire original research window.

At the 24-hour cutoff, the accumulated news is combined with a separate Luna Deep Research pass. Luna searches beyond the Alpaca/Benzinga feed for additional financial reporting, company disclosures, regulatory information, and other relevant evidence available within the permitted research window.

The frozen Initial Analysis is never changed during this process. The expanded evidence set is reserved for the separate second-stage Deep Review performed after the 24-hour research window is complete.

10

24-Hour Luna Deep Research

Once the event reaches its 24-hour cutoff and the final news collection is complete, Luna performs a second independent research pass.

This Deep Research has a larger evidence window and can examine information that became available after the Initial Analysis, while still rejecting information that falls after the 24-hour cutoff.

Luna remains the research layer. It gathers and structures evidence but does not decide the final classification or recovery forecast.

11

Sol 24-Hour Deep Review

Sol receives the frozen Initial Analysis together with the Initial Research and the completed 24-hour Deep Research packet.

It then produces one separate Deep Analysis. This can confirm or change the classification, confidence, catalyst assessment, evidence quality, and 3D / 7D / 20D recovery forecasts.

The Deep Review does not overwrite the Initial Analysis. Both versions are preserved so the website can show what the system believed after the first research window and what changed after a full day of additional evidence.

Initial vs Deep
Example: Initial 3-Day Forecast 15% → Deep 3-Day Forecast 30%. Both numbers remain stored. The later actual result can then be compared with both forecasts.
12

Live Recovery Tracking

Price recovery is tracked independently from both AI analyses. While the market is open and the event remains active, the scanner updates the current price and Live Recovery approximately every five minutes.

Live Recovery measures how much of the original decline has already been recovered from the recorded event low toward the pre-drop price.

It is an observed market measurement, not an AI forecast, and can move higher or lower as the stock price changes.

13

3 / 7 / 20 Trading Day Actual Results

Official recovery checkpoints are recorded after 3, 7, and 20 trading days.

These checkpoints provide the actual outcome that can be compared with the frozen Initial Forecast and the later 24-Hour Deep Forecast.

This creates a measurable history of whether the early model, the Deep Review, both models, or neither model correctly estimated the subsequent recovery.

14

Recovery Milestones, Telegram & Archive

Historical events remain available in the Alert Archive together with their Initial Analysis, Deep Review when available, and actual recovery results.

Telegram can send separate notifications for the original DEEP detection, completion of the Initial AI Analysis, completion of the 24-Hour Deep Review, and qualifying recovery milestones.

Current recovery milestone levels are 25% recovery at 3 trading days, 60% at 7 trading days, and 80% at 20 trading days.

15

No Trading Access

Allan's Market Lab is an educational market-monitoring and research project. It does not place trades, control brokerage accounts, or automatically buy or sell securities.

Its purpose is to detect unusual market events, research possible causes, preserve time-stamped AI assessments, and measure those assessments against what actually happens afterward.