A Global Author-Identity Map for the World of Code:62.7M Developer Identities from 106.8M Author Strings over 5.87B Commits

2026-07-08 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Software Development & Engineering, Data Science & Analytics, Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

A new author-identity map for the World of Code (WoC) V2604 has been released, resolving 106,826,059 raw author/committer strings into 62.7 million canonical developer identities across 5.87 billion commits. This map is "mega-cluster free," with its largest cluster containing 6,910 IDs, a significant improvement over previous WoC maps that featured a 3 million-ID mega-cluster. The system resolves 73.5% of all commits into multi-ID identities, boosting human-ID commit coverage to 98.17%. It includes four artifacts: a global alias map, a per-identity classification, a within-project resolution table, and a commit-to-identity table. Validated against the ALFAA human-rated gold set, the map achieves a recall of 0.70 and precision of 0.88, demonstrating superior accuracy by explicitly addressing "clumping" errors. This map significantly corrects downstream analytics, including developer head-counts and collaboration network structures.

Key takeaway

For data scientists analyzing large-scale software repositories, relying on raw author strings for metrics like head-counts or bus factors introduces substantial errors. Your analyses will be significantly distorted, overstating collaboration and inflating developer counts by up to 66.6%. You should integrate the new WoC author-identity map and its quality classifications to ensure accurate, mega-cluster-free insights into developer contributions, team resilience, and collaboration networks. This map provides the necessary precision for reliable empirical software engineering studies.

Key insights

Accurate global author identity resolution requires prioritizing precision to prevent massive over-merges.

Principles

• Structural certificates do not transfer; re-check mega-cluster histograms.
• Evaluate identity maps using both precision and recall benchmarks.
• Classifiers need out-of-distribution transfer validation, not just in-distribution accuracy.

Method

The method follows a multi-stage pipeline: gate, cut topology, classify edges, recover recall, then resolve locally, each step addressing specific failure modes.

In practice

• Use c2AFull or a2AFullSUG to resolve commit or raw author strings.
• Filter identities by provenance tags or A2clsFull class for quality control.
• Apply the gate-cut-classify-recover-resolve construction template to other graphs.

Topics

• Author Identity Resolution
• World of Code
• Mining Software Repositories
• Alias Resolution
• Data Quality
• Software Engineering Analytics

Code references

• bockthom/gitauthority

Best for: AI Scientist, Research Scientist, Data Scientist

Related on AIssential

• Project-wise Comparison of Software Birthmarks Using Weighted Partial Similarity — Project-wise software birthmark comparison requires weighting module similarities and focusing on top matches to detect partial reuse and mitigate incidental similarity.
• Twenty Years of HAREM: A Reproducible Audit and Reassessment of Portuguese Named Entity Recognition — Data contamination in benchmarks significantly distorts model performance and masks true generalization capabilities.
• RealDocBench: A Benchmark for Field-Level QA and Layout Understanding on Real-World Regulated Documents — RealDocBench offers a field-level, real-world benchmark exposing true document parsing system performance and operational trade-offs.
• Effects of Training Data Quality on Classifier Performance — Classifier performance degrades universally with training data quality, leading to coincidentally correct but flawed decisions.
• HybridCodeAuthorship: A Benchmark Dataset for Line-Level Code Authorship Detection — Industry codebases require fine-grained, line-level detection of AI-generated code for risk and productivity analysis.
• A Mechanism and Optimization Study on the Impact of Information Density on User-Generated Content Named Entity Recognition — Low information density in UGC causes NER model performance collapse by inducing "attention blunting" and "conservative prediction bias."

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.