This research does not have a single intended audience. Some visitors are researchers looking for prior work. Some are practitioners trying to solve a reliability problem. Some are leaders evaluating what rigorous AI reliability research looks like. Each group needs a different starting point.

Choose the profile that fits you best. The reading path will point you to what matters most for your context — and away from what you can safely skip for now.

I am a

For researchers and academics

You are working on adjacent problems — reliability, verification, human-AI interaction, extended cognition, or long-horizon AI collaboration — and you want to understand how this research frames those problems and what it has found. Start with the theoretical grounding, then move to the empirical findings.

01 — Start
Research Overview
The central question, six research themes, and five major findings. Gives you the theoretical landscape before you read individual papers.
→ research/
02 — Foundation
Papers 01 + 02
The reliability architecture and memory governance papers. Together they define the theoretical framework on which everything else builds.
→ papers/
03 — Empirical
Paper 03: Failure Taxonomy
Six named failure modes derived from empirical observation. The first systematic taxonomy of structural AI collaboration failures.
→ papers/
04 — Recent
Papers 07 + 08
Distributed cognition as the right frame for human-AI systems, and the verification layer as the structural gap in current reliability approaches.
→ papers/
05 — Active
Project RITAM
The active reliability architecture research. Open investigation questions and current status — where collaboration might intersect.
→ ritam/
06 — Priority record
Early Observations
What the research documented before those concepts became mainstream — the intellectual priority record.
→ early-observations/
07 — Applied layer
Applied AI Guides
Twelve field-specific guides showing how this research translates into practice — for practitioners in adjacent fields working with AI.
→ Applied AI Guides

For practitioners building or running AI workflows

You are working with AI systems in production — building collaborative pipelines, managing long-horizon AI projects, or dealing with reliability failures you cannot quite name or diagnose. Start with the failure library, then take the assessment to locate your specific gaps.

01 — Diagnose
Failure Library
Six documented failure modes with descriptions, signals, and what they cost. Start here to identify whether a problem you have seen has a name and a structural cause.
→ failure-library/
02 — Assess
Reliability Assessment
Twenty questions across five reliability domains. Identifies which failure modes your workflow is most exposed to and which papers address them.
→ assessment/
03 — Understand
Papers 01 + 02
The structural conditions for reliable AI collaboration — context integrity and memory governance. These are the most directly applicable papers for practitioners.
→ papers/
04 — Apply
Project NIYOM
The execution layer project — how the framework's architectural principles translate into implementable workflow structures.
→ niyom/
05 — Translate
Terminology Translator
Maps the industry vocabulary you already use — context engineering, RAG, grounding, AI memory — to the Living Framework concepts that describe the same phenomena more precisely.
→ terminology/
06 — Self-directed
Applied AI Guides
Twelve guides written for practitioners in your field — managers, lawyers, teachers, HR professionals, recruiters, administrators. Grounded in the same research, built for immediate use.
→ Applied AI Guides
07 — If needed
Work directly with Rishi
If the assessment identifies significant exposure across multiple failure modes, a direct engagement may help you prioritise and address them.
→ advisory/

For leaders and decision-makers

You need to understand the reliability risks in your organisation's AI usage at a level that lets you make informed decisions — about investment, oversight, or where to focus attention. Start with the research overview, then evaluate specific exposure with the assessment.

01 — Orient
About the Research
What this research is, how it has developed, and what makes it different from general AI commentary or enterprise AI product marketing.
→ about/
02 — Understand
Research Overview
The central question and five major findings. Readable without prior technical knowledge — designed to give a clear view of what the research has established.
→ research/
03 — Identify risks
Failure Library
The six structural failure modes and what they cost in practice. Concrete enough to apply to your organisation's AI workflow without technical detail.
→ failure-library/
04 — Locate gaps
Reliability Assessment
Twenty questions across five domains. Even if your team takes this, the domain breakdown reveals where your AI collaboration architecture has structural gaps.
→ assessment/
05 — See direction
Research Roadmap
Where the research is going and what questions remain open. Useful for understanding the frontier and what will likely emerge from this work.
→ roadmap/
06 — Self-directed
Applied AI Guides
Twelve field-specific guides for self-directed application. If direct engagement is not the right fit yet, these give your team an immediate, research-grounded starting point.
→ Applied AI Guides
07 — If appropriate
Advisory Engagement
If the assessment reveals significant structural gaps, a direct engagement can focus the research specifically on your organisation's context.
→ advisory/
Good to know

No account required

All papers are published on Zenodo with open access. The assessment runs entirely in your browser — no data is collected. Everything on this site is freely accessible.

The framework is evolving

This is called a "Living Framework" because it develops as the research does. Papers 01–08 are published. RITAM and NIYOM are active. New findings get incorporated. The roadmap shows what is currently under investigation.

Independent research

There is no institution, no funding body, no corporate sponsor. The research is not affiliated with any employer or company. Findings reflect what the evidence supports, not what any stakeholder would prefer.

Starting from a specific failure?

If you have a concrete failure in mind — context drift, numerical errors, trust breakdowns — go to the Failure Library first and work backwards from there to the relevant papers and assessment domains.

Everything on this site