arXiv
Before the Pull Request: Mining Multi-Agent Coordination
Dipankar Sarkar
Abstract
Studies how concurrent coding agents claim, divide and collide over shared work before a pull request exists, using grite, a git-native append-only coordination log. Duplicate work falls from 78% to 0% and failure modes become minable from the log.
Autonomous coding agents now open millions of pull requests, yet large-scale studies find their PRs are produced faster but accepted less often - a coordination and trust gap that pull-request-level telemetry cannot explain. We argue the missing signal lives before the PR, in how concurrent agents claim, divide, and collide over shared work. We study this process through grite, our open-source coordination substrate that needs no central server and stores its records inside git itself, so its append-only, signed event log captures the coordination process directly. We show that (i) this shared substrate reduces duplicate and conflicting work at bounded overhead - the share of work that merely re-does a teammate’s task falls from 78% to 0% while useful throughput more than triples; (ii) every agent’s copy of the log converges to the same state with no write silently dropped, where a file-based tracker loses concurrent writes; and (iii) the log is a mineable artefact from which concrete failure modes - conflicting edits, lock starvation, redundant rediscovery, race-to-close - are automatically recoverable with provenance, several invisible in pull-request history. We release the dataset, harness, and mining toolkit.
arXiv comments: 9 pages, 2 tables. LNCS format. Code, dataset, and mining toolkit: https://github.com/neul-labs/grite
Frequently Asked Questions
What does the paper study?
The coordination that happens before a pull request: how concurrent coding agents claim, divide and collide over shared work. Pull-request-level telemetry cannot see this process.
What is grite?
grite is an open-source coordination substrate (MIT licence) that needs no central server and stores an append-only, signed event log inside git itself. In the paper it is both the coordination mechanism and the source of the mined data.
What were the results?
With the shared substrate, the share of work that merely re-does a teammate's task falls from 78% to 0% while useful throughput more than triples; every agent's copy of the log converges with no write silently dropped, where a file-based tracker loses concurrent writes; and failure modes such as conflicting edits, lock starvation, redundant rediscovery and race-to-close are recoverable from the log with provenance.
Where are the dataset and tools?
The dataset, harness and mining toolkit are released with grite at github.com/neul-labs/grite.