# PhyloCode Research Evidence

> PhyloCode studies how repository state and history forecast measured software outcomes across projects, longer horizons, and calibrated activity ranges.

Last updated: 2026-07-23.

## Evidence summary

The public evidence currently supports three connected views of software evolution:

1. **Cross-repository lifecycle and growth rankings.** Dormancy, reactivation, and file-growth evidence was positive in all 21 repository-outcome evaluations on unseen repositories.
2. **Durable signal at longer horizons.** When the six-month target window moved to months 19–24, forecast lift retained 83% for dormancy, 99% for reactivation, and 109% for growth.
3. **Calibrated activity-volume distributions.** Forecasts now estimate a distribution over integration activity during the next six months, adding a plausible range and explicit uncertainty to event rankings.

## Activity-volume forecast evidence

The calibrated activity-volume model outperformed three distinct forecast comparisons:

- persistence from the current state;
- repository-aware historical averages;
- a matched model using recent activity alone.

The advantage held for both the full activity distribution and activity magnitude when activity was nonzero.

Evidence consistency:

- **27 of 28** evaluated repositories favored the activity-distribution model over the repository-aware baseline;
- **5 of 5** random seeds favored the model;
- full-distribution calibration error was **0.039**;
- nonzero-activity calibration error was **0.034**;
- both calibration errors were below the **0.05** evaluation threshold.

## Current research instruments

### Activity-volume distributions

Estimate a calibrated distribution for integration activity over the next six months. The output is a range of plausible activity volumes with uncertainty, rather than a single point estimate.

### Dormancy and reactivation rankings

Rank active components by their likelihood of going quiet and dormant components by their likelihood of returning during the next six months.

### Growth likelihood

Rank where active components are more likely to gain files during the next six months.

## Interpretation

These results concern measured operational repository outcomes: integration activity, dormancy, reactivation, and file-count growth. The evidence is best described as cross-repository forecasting of software dynamics.

The longer-horizon result shows that the measured signal remains useful well beyond the immediate next window. The distributional result extends the evidence from binary event rankings to calibrated ranges of future activity.

## Terminology

- **Component:** A coherent unit within a repository used as the forecasting resolution.
- **Dormancy:** An active component becomes inactive during the next six months.
- **Reactivation:** An inactive component resumes activity during the next six months.
- **Growth:** An active component gains files during the next six months.
- **Integration activity:** Repository events integrated into a component during the forecast window.
- **Calibration error:** A measure of how closely forecast probabilities match observed frequencies; lower is better.
- **Forecast lift:** Improvement over the corresponding baseline forecast.

## Canonical links

- Website: https://phylocode.com/
- Research: https://phylocode.com/research
- Tools: https://phylocode.com/tools
- Work with us: https://phylocode.com/work-with-us
- Contact: hello@phylocode.com