QuantDinger AI Trading OS

Least privilege, user confirmation, data-flow transparency, sensitive-data handling, and external-effect controls have strong static support: ports bind to loopback by default; production containers are described as non-root, read-only, and capability-dropped; agent tokens are scoped, rate-limited, hashed, and audited; and agent trading defaults to paper mode. Live agent trading requires token scope, paper_only=false, a server switch, and operator limits and allowlists. The README explicitly warns that real orders can be submitted and maps flows among clients, APIs, databases, Redis, workers, and broker adapters. Credentials and MFA secrets are encrypted, environment files use mode 600, and operators are warned against public infrastructure exposure and example passwords. Dependency security receives 2 because CI shows lock validation, constrained tool versions, image SBOMs, and provenance, but the supplied material contains neither vulnerability-scanning configuration/results nor the dependency manifests themselves. Rollback receives 2 because PostgreSQL and durable-Redis backups, durable commands, and reconciliation are mentioned, but live-order reversal, deployment rollback, and disaster-recovery procedures are not shown. Source attribution receives 2: the README identifies Open Byte Inc., support contacts, and public channels, while the license appendix mentions quantdinger.com; however, publisher identity is unverified and the evidence lacks a definitive copyright-owner declaration or contributor governance.

The architecture, execution flows, version conventions, and CI configuration are internally consistent. CI statically demonstrates syntax checks, linting, migrations, sharded tests, release gates, Compose validation, version-drift checks, and encoding checks, supporting full self-consistency credit. Dependency availability receives 2: Python 3.12, PostgreSQL 18, Redis 8, lock-file installation, prebuilt images, and multi-architecture publishing are documented, but major capabilities depend on external AI providers, exchanges, and brokers, with no supplied offline fallback or availability matrix. Failure messages receive 2 because configuration checks, health endpoints, observability, alerts, and an explicit CI error for empty test shards are shown, while concrete user-facing runtime error examples, an error taxonomy, and detailed recovery behavior are absent.

The audience and scenarios clearly cover independent traders, Python strategy authors, and small teams across research, backtesting, paper trading, live execution, and monitoring. Capability boundaries distinguish HTTP handling, long-running trading loops, finite Celery work, schedulers, and third-party responsibilities, while explicitly disclaiming investment advice. Trigger precision is strong because live agent trading has four independent gates and work is deliberately routed among durable commands, finite jobs, and long-lived runtimes. Environment fit earns full credit for prebuilt and source installations, Linux/macOS/Windows instructions, Docker Compose, local and production overlays, optional monitoring, resource-constrained deployment guidance, rootless/NFS notes, and multi-architecture images.

Information architecture is unusually concrete, with repository layout, execution-path mappings, and a table identifying where each category of change belongs. Installation notes cover quick and source installs, mandatory secrets, administrator migration behavior, production validation, and troubleshooting references. Naming and version stability are supported by VERSION, semantic release tags, Strategy API V2, /api/agent/v1, consistency checks, and an API compatibility workflow. Examples and FAQ receive 2: many commands and guide links are provided, but the supplied material does not contain a full FAQ or an end-to-end agent-call example. Known limitations earn full credit because the sources state live-trading risk, jurisdictional responsibility, third-party exclusions, main-only security support, separately hosted frontend source, and optional monitoring. The complete Apache-2.0 license matches the metadata. Versioning and changelog practices are supported by tags, generated release notes, and version checks. Maintenance responsibility receives 2 because product attribution, a support address, security response targets, and a main-only support policy exist, but publisher identity remains unknown and no maintainer roster, governance model, or long-term support commitment is shown.

Outputs appear directly usable: inspectable strategy code, backtests, experiments, reports, execution state, audit logs, metrics, dashboards, and alerts are exposed through web, mobile, API, Agent Gateway, and MCP surfaces. Marginal value earns full credit because the repository integrates AI research, strategy development, backtesting, paper/live execution, and monitoring into one self-hosted workflow while preserving operator control and a shared service layer. Cost-benefit receives 2: the core stack can remain smaller and observability is optional, while prebuilt images reduce setup effort; however, PostgreSQL, two Redis instances, several workers, and optional Prometheus/Grafana/Alertmanager create substantial operational overhead, and no resource benchmarks, cost estimates, or measured performance benefits are supplied.

Major claims are mapped to named directories, configurations, documents, OpenAPI artifacts, tests, and CI steps; execution and change-location tables make the claimed design especially traceable. Cross-source corroboration receives 2 because SECURITY.md and the workflows support some README claims about security, versions, and checks, but the supplied evidence excludes the core trading, authorization, and encryption implementations and contains no independent or executed results. Fact/inference separation is strong: defaults, production recommendations, external responsibilities, and disclaimers are distinguished clearly, and this assessment does not treat configured CI as proof that a run succeeded.