Cross-Project Incident Timing Correlation(3.4.10)

When two separate projects both experience issue spikes at the same time, consistently, it may not be a coincidence. Shared deployment pipelines, shared infrastructure components, or shared dependencies can cause correlated failures across multiple projects simultaneously.

This visual quantifies the timing correlation between project pairs — so you can see which pairs of projects spike together and decide where a shared driver is worth investigating.

What the coefficient is measured on

A portfolio has its own rhythm: quiet weeks and busy weeks that lift or drop every project at once. Correlating raw weekly counts would report that shared rhythm as a relationship between every pair, so this chart removes it first. For each week, the average incident count across all projects is subtracted from each project's own count, and the coefficient is the Pearson correlation of those deviations. A pair therefore scores highly only when it moves together beyond whatever the portfolio as a whole was doing that week.

The week in progress is not counted: every coefficient is measured over complete weeks only, so it does not shift as the current week fills. A part-week is short by however much of it has not happened yet, and a single coefficient has nowhere to show that.

A portfolio of only two projects gets no coefficient at all. With two projects the pair IS the portfolio, so one project's deviation from the weekly average is always the exact opposite of the other's, and the calculation returns -1.00 whatever the two projects actually did. That is arithmetic rather than a measurement, so the cell is left blank instead. Three or more projects are needed for the comparison to mean anything.

What you can conclude

• A high coefficient (above 0.7) between two projects means their incident weeks rose and fell together relative to the rest of the portfolio. Shared infrastructure and a shared release process can both produce this; which one applies is a question the coefficient alone cannot answer.
• Investigating highly correlated project pairs together (rather than separately) may reveal a shared root cause that individual project-level analysis would miss.
• A negative coefficient is a reading worth having: one project ran busy in the weeks the other ran quiet, relative to the portfolio. Because the shared rhythm has been removed, a matrix with roughly half its pairs below zero is the ordinary shape of an unrelated portfolio rather than a signal in itself.
• A low coefficient across all project pairs indicates that incidents are arriving independently — project-level root causes are more likely than shared infrastructure failures. This reading applies to the cells that carry a number: 0.00 is a measurement of independent timing, while a blank cell is a pair the matrix could not measure and supports no conclusion either way.

How this chart works

Correlation matrix showing, for each pair of projects, the Pearson correlation of weekly issue-creation counts after that week's cross-project average has been subtracted from each. Higher values indicate incident timing that is synchronized beyond the portfolio's own weekly pattern. A pair needs at least four complete shared weeks to be published at all.

Measured cells print their coefficient to two decimals, and the diagonal is always 1.00. A cell with no number is not placed on the color scale at all: it takes the same pale neutral background as the near-zero band, so the number rather than the shade is what tells you a pair was measured, and hovering a blank cell reads r = n/a.

A project is only compared if it recorded incidents in at least four separate weeks. A project with a handful of incidents spread thinly across the window is mostly empty weeks, and two such projects can appear to move together almost perfectly while resting on two or three incidents each — so they are left out rather than shown as a strong reading. If a project you expect is missing from the matrix, that is why, and it is a statement about how much incident history it has rather than about the project itself.

A blank cell and a cell reading 0.00 say different things. 0.00 is a real reading: both projects were measured, and their incident weeks moved independently of each other. A blank cell means no coefficient exists for that pair — either the two projects share too few complete weeks of history to support one, or one of them sat at the same distance from the weekly portfolio average in every shared week, and a series that never varies gives a correlation nothing to track. Reading a blank cell as evidence that two projects are unrelated is the mistake this distinction exists to prevent.

A whole row and column of blanks points at one project rather than at a gap in the matrix: a project whose weekly position never moves relative to the portfolio leaves every pair it belongs to unmeasured.