FIELD NOTE 02 · SYSTEMS
From operational systems to a research testbed
The systems described here motivate the experiment. They do not substitute for experimental data, a technical audit, or independent validation.
1. System context
NOAERTH is an independent software and venture initiative. Its reported internal architecture separates coordination, execution, and retention functions. Within that architecture, Core is used for task management and policy, Forge for work execution and tests, and Storage for archival and recovery. These names are working roles, not standardized research instruments.
Verification boundary. This site does not claim that these services have been independently audited, that every autonomous component is live, or that prior agent-generated completion reports are research evidence.
2. Why this environment is interesting
An operating environment with different models, provider limits, fallback policies, and approval gates offers conditions that can be observed and manipulated. But production activity is a poor experimental setting without controls: workloads differ, model availability changes, and partial task histories can be lost during recovery. The experimental harness must therefore run in an isolated environment with fixed task inputs and durable records.
3. Proposed research roles
| Role | Contribution to a controlled study |
|---|---|
| Core | Task manifests, assignment policy, approvals and state transitions. |
| Forge | Sandboxed experimental runs, evaluation tools, artifact capture. |
| Storage | Versioned manifests, retention, redacted logs, provenance snapshots. |
Research processes should not run unrestricted against production repos, live applications, private data, or third-party services. Credential scope, data retention, resource limits, and rollback need to be designed before the first real trial.
4. Evidence hierarchy
The reporting system should distinguish proposed work, executed activity, agent-reported completion, independently verified completion, and replicated results. “Blocked” and “inconclusive” are valid states. A successful webpage deployment or a passing unit test is not evidence that a research hypothesis has been confirmed.
5. What can be examined now
The research specification describes the intended comparison groups and measures. A small descriptive scorer and its synthetic test fixtures show how aggregate outputs may be calculated once independently evaluated data exists. No production traces, confidential code, validated experimental dataset, or measured performance advantage is published.
The next artifact needed is a dated, reproducible pilot run and a clearly written account of failures. Until that exists, the appropriate public claim is methodological work in progress.