Research papers
Open-access research from NXTG.ai on AI-agent governance, evaluation, and coordination. The evidence archive and three of the papers each have a citable Zenodo DOI. The origin report has no DOI of its own; it cites the DOIs of its three datasets. Where a work has a dataset, it is public on HuggingFace.
ASIF: An Origin Report on a Self-Measuring Multi-Agent System at Portfolio Scale
Waliuddin, Asif · 2026-07-06
Industrial experience report on ASIF, the system I run my own company on, with receipts published, and its first pre-registered result: a clean routing tie.
The CRUCIBLE Protocol: Auditing Measurement Integrity in AI-Assisted Software Development
Waliuddin, Asif · 2026-08-06
CRUCIBLE is a governance protocol for auditing whether an AI-assisted project's measurement system provides evidence not controlled by the same producer. It is a nine-gate protocol (eight audit gates plus a forward extension) that separates test quality from measurement integrity and routes audits independently of the code's author.
The Repository as Coordination Layer: Federated AI Agent Governance via Version-Controlled Shared Context
Waliuddin, Asif · 2026-08-06
The first documented production-scale evaluation of git-as-multi-agent-coordination at portfolio governance scope: two autonomous AI agents on two physically separate machines governed 21 software projects for 21 weeks (152 days) through a single shared version-controlled repository, with no message broker, shared database, or orchestration service.
Enabling Bureaucracy: Engineering Patterns for Governing a Fleet of Autonomous AI Agents at Portfolio Scale
Waliuddin, Asif · 2026-08-06
A documented production deployment in which two autonomous Claude Code agents govern 18 software projects across two machines through a 7-layer autonomy enablement stack, a 23-principle constitution across 3 override tiers, and a file-based directive protocol that survives process death and machine reboots using only git remotes and cron synchronization. Contributes four engineering patterns: portfolio-scale federated governance, the 7-layer stack, the federated directive protocol, and an inverted Constitutional AI methodology extracting principles from a 54-decision precedent corpus.
Evidence archive
TRACE Index v0: evidence archive (probe scripts, redacted raw outputs, checksums)
NextGen AI - Intelligent Systems (NXTG.AI) · 2026-10-21 · Dataset, CC BY 4.0
“Evidence archive for TRACE Index v0. For six agent-runtime defects already reported on vendors' public issue trackers, NXTG tried to reproduce each one on hardware it owns, under one method pre-registered before any probe ran, and publishes what happened, success or not. Results when the run window closed (2026-10-17 12:00 UTC): 1 reproduced, 2 not reproduced, 3 not finished.”
From the Zenodo record