Short-squeeze lists are dangerous because they hide complexity behind a single number. Two stocks with identical scores can be there for completely different reasons. Daniel Oliveira, writing on DEV.to on October 21, 2026, demonstrates how to use an AI agent connected to the Equibles MCP server to tear apart these scores. By querying FINRA data directly, he reveals that headline numbers often mask critical differences in positioning versus catalysts.

The Illusion of the Top Score

Oliveira starts by pulling the top short-squeeze scores for stocks over $1B in market cap with at least $10M in daily volume. The board shows Choice Hotels (CHH) with a score of 98 and Hasbro (HAS) with 97. On the surface, they look like similar high-risk setups. However, CHH has 14.7% of its shares sold short, while HAS has only 5.9%. The AI agent breaks these scores down into their constituent factors, showing that CHH's high score is driven by heavy positioning and days to cover, whereas HAS's score is inflated by a capped catalyst boost from recent price spikes and volume surges.

Digging Into the FINRA Data Stream

The real power of this workflow is bypassing the summary to see the raw FINRA series. Oliveira queries the agent for eight settlements of short interest data. For CHH, the days to cover have remained stubbornly between 10 and 14 days, indicating a rebound in short interest rather than a new buildup. For HAS, the recent 14.2% increase in short interest is traced to a single jump between September 15 and September 30, with positions roughly returning to mid-June levels. This granular view exposes the lag and noise inherent in aggregated scores.

Why Daily Short Volume Is Misleading

A common trap is misinterpreting daily short volume. The agent retrieves CHH's daily short volume, showing a 77% short volume on October 7. Oliveira clarifies that this data only covers trades reported to FINRA's off-exchange facilities, not the entire tape. Furthermore, 40 to 50% short volume is the normal baseline due to market maker mechanics. A 77% day is elevated, but it does not mean 77% of all trading was new short positions. Without this context, traders can easily misread the intensity of selling pressure.

The Lag Problem and Model Estimates

FINRA data is twice-monthly with a lag, meaning the scores use positions that are already weeks old. Until the next settlement report on October 26, the Equibles server provides model estimates. For CHH, the model suggests a 45% chance of a rise in short interest, which is effectively a coin flip. For HAS, the chance is 54%. Both ranges straddle the previous figure, offering no definitive signal. This highlights a critical limitation: AI agents can process data quickly, but they cannot overcome the fundamental latency of regulatory reporting.

Key Takeaways

  • Aggregate squeeze scores obscure the difference between heavy positioning and temporary catalysts.
  • FINRA short volume data is off-exchange only and requires context regarding market maker baselines.
  • Model estimates for upcoming settlements are probabilistic and often inconclusive when ranges straddle prior data.
  • Connecting AI agents to MCP servers like Equibles allows for real-time dissection of complex financial datasets.

The Bottom Line

Trust the source data, not the dashboard. AI agents are powerful tools for auditing black-box scores, but they are only as good as the underlying data's timeliness. In short squeezes, the lag is the enemy, and the score is just the symptom.

Sources

DEV.to