RStack is being developed as both a product and a research-backed AI-SDLC operating model. This page links the research artifacts that explain what RStack implements today, what it claims, what remains a hypothesis, and how future productivity claims should be measured.
Research thesis
RStack improves the practical use of AI coding agents by wrapping them in a governed lifecycle:
The careful version of the thesis is:
RStack does not claim that AI coding alone guarantees productivity. RStack claims that AI-assisted delivery becomes more reliable and measurable when work is constrained by lifecycle stages, approvals, typed handoffs, evidence, budget envelopes, and Business Hub observability.
Repo research artifacts
Claims discipline
Every RStack paper or docs claim should have at least one of:
- measured evidence from RStack runs, tests, PRs, or CI,
- source evidence from standards, reports, or prior art,
- implementation evidence from code or docs,
- a clear hypothesis label and experiment plan.
Do not claim quantified productivity gains until RStack has measured comparison runs. The current research-safe claim is that RStack creates the lifecycle and evidence structure needed to measure and improve AI-assisted delivery.
Research-backed roadmap
The research program is tracked in GitHub issues:
Suggested paper measurement model
RStack should measure productivity through multiple lenses:
- delivery flow: time to plan, build, validate, and release-readiness,
- quality: builder/validator pass rates, retries, risks, tests run,
- governance: approvals, blocked gates, policy overrides,
- traceability: requirement-to-task-to-evidence completeness,
- cost: estimated vs actual usage when host/provider data is available,
- operator burden: number of human interventions per run.
See research/methodology.md for the full measurement plan. 