Real statistics, not just dashboards
A dashboard can show that cycle time changed. Gitrevio helps you understand why, how much it changed, and whether the change is meaningful. Gitrevio uses proven statistical methods to look at engineering data. It can detect important changes, compare groups, estimate the impact of a change, find the factors behind a result, and forecast what may happen next. The methods are tested and calibrated using your own engineering history.
The methods, explained
BOCPD
Finds important changes in your engineering data.
Bayesian Online Change Point Detection looks for shifts in how your metrics behave. It can help identify when something changed without relying on manually chosen thresholds.
Cox proportional hazards + Kaplan-Meier
Helps understand how long work takes to finish.
Gitrevio uses survival analysis to look at time to merge and time to resolve. You can compare teams and see which factors are linked to faster or slower completion.
Doubly-robust cohort lift
Measures the impact of a change between groups.
Gitrevio compares cohorts while accounting for differences between them. It also provides confidence intervals, so you can see how certain the result is.
Shapley attribution
Shows which factors contributed to a result.
When a score or metric changes, Gitrevio breaks down the change across its contributing factors. The attribution follows tested Shapley properties rather than using arbitrary weights.
Kalman filter
Makes noisy metrics easier to understand.
Gitrevio smooths metric data to separate normal noise from meaningful movement. The model adjusts to your own historical data.
Lognormal cycle-time forecasting
Forecasts when work is likely to be completed.
Gitrevio uses your team's delivery history to estimate p50, p75, and p90 completion dates. Forecasts are updated as new data comes in.
Bayesian network
Connects different engineering outcomes.
Gitrevio is building a model that connects delivery, quality, attrition, and other engineering outcomes. It will allow you to ask questions about how one outcome may change when another factor changes.
LinUCB policy search
Helps find which recommendations work best.
Gitrevio tracks recommendations and their results over time. This helps identify which actions are more useful in different situations.
Anomaly to root cause
Helps trace a change back to the signal behind it.
When Gitrevio detects something unusual, it follows the relationships between metrics to help identify what may have caused the change.
A worked example
Lead time dropped 18% in March. BOCPD locates the change point, Shapley attribution partitions the drop across its factors, and survival analysis reports the shift in time-to-merge hazard.
Decisions, not just correlations
Seeing two things change at the same time does not tell you what caused the change.
For example, cycle time may have dropped while the team grew. That does not automatically mean the larger team made delivery faster.
Gitrevio uses statistical methods to go further. It can estimate the size of a change, compare different groups, identify meaningful shifts, and show the factors behind a result.
Every attribution is tested.
When Gitrevio uses Shapley attribution, its core properties are checked through automated tests. This helps make sure the results are consistent and can stand up to review.
Calibrated to your data
The models use your own engineering history.
Gitrevio does not rely only on fixed assumptions about how engineering teams work. Models are fitted and adjusted using your historical data.
For example, the Kalman filter learns the level of normal variation in your metrics, while BOCPD adjusts its model as new data comes in.
Forecasts also learn from your delivery history.
Gitrevio uses your team's past cycle times to estimate p50, p75, and p90 completion dates. As new delivery data comes in, the forecasts can be updated.
The tools behind it
Causal claims on this page are produced by named functions in the tool catalog. Each one publishes its assumptions, its identification strategy and the point at which it declines to give an answer.