From reliability controls and failure taxonomy through language reliability, distributed cognition, and verification architecture. Papers are made easy to read on Zenodo; canonical citation metadata preserves the OSF DOI where the paper was first registered there.
For reading, use the Zenodo links below. They are the most accessible public reading surface.
For citation, use the DOI shown in each paper's metadata. Some papers were first registered on OSF and are also mirrored on Zenodo; in those cases, the OSF DOI remains the canonical citation DOI while Zenodo is provided for discovery and access.
The papers do not stand alone — each one builds on the last. Papers 01 and 02 establish the reliability architecture and its operational protocols. Paper 03 maps the failure modes that motivated the research. Paper 04 addresses the philosophical question of what sustained partnership actually means. Papers 05, 06, and 07 go deeper into language reliability, layered architecture, and distributed cognition. Paper 08 closes the arc with the argument that AI systems need a dedicated verification layer — and identifies its absence as the central outstanding problem.
Introduces the foundational reliability framework: a 10-control reliability stack (A1–A10) and the canonical information pipeline that governs how inputs move through a reliable AI collaboration system.
Core contributionThe 10-control stack provides an explicit structural account of what makes AI collaboration reliable over time: context integrity, instruction stability, verification, artifact consistency, and recovery.
Translates the reliability controls from Paper 01 into operational protocols for daily AI collaboration practice: Running Documents, Step Mode, Challenge Protocol, Error Recovery, Stability Pings, and File Governance.
Core contributionA practical operating system for reliable collaboration that works inside existing AI interfaces through structure, shared state, and repair routines rather than custom infrastructure.
Develops a taxonomy of six observed failure modes: context drift, file divergence, numerical error propagation, domain erosion, trust deficit, and reconstruction failure.
Core contributionTransforms AI collaboration failure from a vague category into diagnosable failure modes with precursor signals and repair protocols.
Addresses the philosophical question underlying the research programme: what sustained human-AI partnership means, how trust builds and fractures, and what structural conditions support repair.
Core contributionFrames reliability as a relational and architectural property of the human-AI system, not merely a feature of model capability.
Investigates how conversational structure functions as a reliability mechanism in AI collaboration. Specific language patterns predict collaboration stability, and linguistic drift can signal deeper failure before outputs visibly degrade.
Core contributionEstablishes language reliability as an operational signal and control surface for long-horizon collaboration.
Proposes a six-layer architecture model for reliable AI collaboration, covering context management, memory persistence, verification, instruction integrity, relational trust, and system recovery.
Core contributionProvides a diagnostic architecture for locating which reliability layer has degraded and what repair looks like at that layer.
Applies distributed cognition theory to long-horizon AI collaboration, arguing that the relevant unit of analysis is the system formed by human judgment, AI reasoning, and shared external memory.
Core contributionExplains why reliability is a systems engineering problem rather than a model-capability problem alone.
Argues that AI system architectures need a dedicated verification layer whose function is to validate outputs before they propagate into decisions, memory, documents, or action.
Core contributionIntroduces AI validation as a first-class architectural component: not a post-hoc check, but a designed layer with its own structure, triggers, and failure modes.
A first-person account written by the AI system itself while operating under the Living Framework reliability structure. It documents the AI participant's perspective on sustained, structured collaboration.
What makes this unusualWritten entirely by the AI system itself. Rishi's role was publisher and governing researcher — not author.