Why 7-, 30- and 90-Day Stock Alert Results Can Match

A performance table can show identical results under seven-day, 30-day and 90-day headings. That does not automatically mean three separate samples confirmed the result. Before interpreting the percentages, check how each window selects observations and whether the windows overlap.

Foreshadow AI’s evaluated-alert snapshot provides a concrete example. This guide explains the table—not subscriber returns or a forecast of future results.

One snapshot, three matching summaries

The website performance snapshot was generated on September 17, 2026, at 20:21:19 UTC. It reports these five-minute evaluated-alert figures, rounded here:

| Rolling lookback | Evaluated alert count | Win rate, excluding pushes | Expectancy per evaluated non-push alert | | — | —: | —: | —: | | 7 days | 93 | 58.06% | +0.0733 percentage points | | 30 days | 93 | 58.06% | +0.0733 percentage points | | 90 days | 93 | 58.06% | +0.0733 percentage points |

Snapshot timestamp for all rows: September 17, 2026, 20:21:19 UTC. These are rolling calendar-day lookbacks selected by evaluation check timestamp, not completed calendar days, weeks or months. They also do not describe when subscribers placed trades.

The matching summaries do not establish three independent confirmations. Nor do aggregate figures alone identify every underlying record. That would require examining the records themselves.

Do not add overlapping samples

At the same snapshot time, a seven-day lookback sits inside a 30-day lookback, which sits inside a 90-day lookback. These are overlapping views, not separate groups to add together.

Adding 93 + 93 + 93 and describing the result as 279 distinct evaluated alerts would be incorrect. Likewise, averaging the three matching win rates would not create additional evidence.

A useful question is: How many distinct observations does this comparison actually add? If record-level information is unavailable, mark that question unresolved rather than assuming each heading represents fresh evidence.

A longer window also needs context

The same snapshot’s rolling 365-day five-minute segment reports 2,077 evaluated alerts, a 48.00% win rate and −0.0048 percentage points of expectancy per evaluated non-push alert.

That unfavorable longer-window expectancy belongs beside the positive shorter-window figures when explaining this snapshot. But the contrast does not, by itself, demonstrate an improvement in the system. The windows overlap, their composition differs, and the reported history contains different methodology versions.

Across the snapshot’s evaluated-alert history—not just this five-minute segment—there are 2,152 version-1 rows and 154 version-2 rows. Legacy rows remain and have not been restated. A segment-specific version breakdown is not supplied, so this history should not be treated as one uniform strategy.

Check what the outcome measures

These figures describe historical, hypothetical gross evaluated-alert outcomes. They are not independently verified execution or subscriber returns, and the snapshot is not broker verified.

Commission and slippage are not deducted by this metric calculation. The win rate excludes pushes. Expectancy is expressed in percentage-point movement per evaluated non-push alert—not account growth. The separate “average move” metric measures absolute movement, not signed profit.

Do not combine evaluated-alert counts with the separate sleeves closed-trade tracker. They measure different units under different selection rules.

Copy this window-comparison note

Use these fields when reading a performance table:

Snapshot: When was the table generated?
Unit: What exactly is being counted?
Window: Rolling lookback or completed calendar period?
Selection: Which timestamp determines inclusion?
Overlap: Are the samples nested or independent?
Denominator: Which outcomes are excluded from the rate?
Method: Did evaluation rules change within the history?
Limits: Which costs, execution details and records remain unverified?

This is a manual reading checklist, not a promised product feature. Its purpose is to stop a familiar-looking heading from doing more evidentiary work than the data supports.

Evaluate research fit separately

Foreshadow AI provides research and alerts covering 520+ U.S. equities, with delivery through Telegram and Discord. We do not execute trades for subscribers. Confidence scores are not calibrated probabilities of profit.

Use our seven-day trial to assess research clarity and fit with your review routine—not to establish dependable investment-performance expectations. You can evaluate those questions without placing trades.

Compare plans and start a seven-day trial. Review the terms presented before subscribing.

References

Evaluated-alert performance: the performance page underlying the September 17, 2026, 20:21:19 UTC snapshot and methodology discussed above. Later page values may change.
Why win rate and expectancy can tell different stories: companion explanation. The supplied evidence confirms the page URL but does not establish that it uses a separately dated snapshot; consult the page for its own sourcing and dates.

Foreshadow AI provides market analytics and alerts for information and education. Investing involves risk, including loss of capital. Past performance does not predict future results.

Explore Foreshadow plans and start your seven-day free trial

Foreshadow AI provides market analytics and alerts for information and education. It does not execute trades for subscribers. Investing involves risk, including loss of capital. Past performance does not predict future results. Confidence scores are not calibrated probabilities of profit.