Idea for Consideration: Trend Analysis on Time Series Metrics (e.g., Team Velocity)

[Sebastiaan127001]

Idea for Consideration: Trend Analysis on Time Series Metrics (e.g., Team Velocity)

In some cases, applying static thresholds to time series metrics (such as team velocity) is too rigid and may lead to false positives or missed signals. For example, a single sprint with lower velocity does not necessarily indicate a problem — however, a statistically significant downward trend over multiple sprints might.

Proposal:

Introduce a trend analysis feature that calculates a p-value for the presence of a statistically significant trend (based on linear regression). This approach quantifies whether a perceived trend is likely to be real or just natural variation.

Why?

• Allows teams to distinguish between noise and structural issues
• This is related to change certain metrics to informative by default #10847 : as we have discussed, some metrics are not eligible to be informative by default, but they also do not apply to be evaluated by a static threshold value. The solution is to measure the trend instead of a fixed calculation like average. This makes the metric evaluation more efficient, having less false positives.

Example (Python):

from scipy.stats import linregress

# Example sprint velocities
sprint_numbers = [1, 2, 3, 4, 5, 6, 7]
velocities = [11, 31, 23, 22, 14, 26, 14]

# Run linear regression to extract trend significance
slope, intercept, r_value, p_value, std_err = linregress(sprint_numbers, velocities)

print(f"Trend slope: {slope}")
print(f"P-value (trend significance): {p_value}")

Interpretation of the p-value (trend figure):

A low p-value (e.g., < 0.05) would indicate a statistically significant trend in the data. This "trend score" could be surfaced in Quality-time alongside raw metric values, or flagged via a derived metric.

P-value (trend figure)	Interpretation
< 0.01	Very strong indication of a trend
< 0.05	Significant trend
0.05 - 0.1	Weak indication of a trend
> 0.1	No evidence of a trend

[Sebastiaan127001]

failed ci-jobs and pipeline duration might also apply for a trend metric (PLUS set to default when creating the metric)