Game Discovery Is an Evidence Problem
How game marketplaces can expose source-qualified game discovery evidence to personal agents without publishing private player data or ranking models.
Read the paperSoftware systems / Product reliability / Applied AI
Whitepapers
Long-form arguments, evidence and practical operating models for systems people need to trust.
Published whitepapers
Executable UGC Is a Software Supply Chain
Why executable game mods need immutable release identity, explicit capabilities, staged updates and runtime revocation.
Read the paperLanguage Models Are Changing the Gameplay Interface
What current games and research show about bounded language-model interaction, likely next steps and the design work around them.
Read the paperThe Install Button Hides a Distributed System
A client-state model for measuring reliability from published build to verified launchable state.
Read the paperFrom Competitive Play to Engineering Judgement
How competitive play can provide repeated practice in feedback and coordination, while professional value still depends on engineering evidence.
Read the paperPolished Code Can Hide a Broken Model
Why implementation fluency can conceal an incomplete system contract, and how review can expose the assumptions that matter.
Read the paperAgent Reliability Needs Repeated Evidence
Why average success rates are incomplete for probabilistic workflows, and how repeated evidence can define a defensible operating boundary.
Read the paperMCP Tool Readiness
A layered model for operational evidence, consumer policy and safe routing when agents select MCP tools dynamically.
Read the paperAI Makes Familiar Cyber Attacks Cheaper to Run
How faster retries, adaptation and parallel operations change defensive priorities and the controls that retain their value.
Read the paperWhen Agents Delegate, Authority Must Shrink
Why autonomous agents need bounded, stateful authority that cannot expand through delegation, concurrency or cross-domain execution.
Read the paperWhen AI Evaluations Act on the Real World
Why agent evaluations need independent controls for execution, network access, credentials, external effects, grading and incident response.
Read the paperThe AI Bill of Materials Is an Operational Record
Why AI inventories need deployment identity, evidence classes and operational links before they can support security, change and incident response.
Read the paperThe Semantic Layer Was AI Infrastructure All Along
Why mature semantic models and analytics investments shorten the path to reliable enterprise AI consumption.
Read the paperWhen Memory Becomes Production State
Persistent agent memory creates obligations for provenance, scope, freshness, correction, expiry and audit.
Read the paperWhen You Need an MCP Server and When You Really Don't
A decision framework for choosing between instructions, Agent Skills, existing tools and a new MCP server.
Read the paperWhen AI Decides What Matters
AI assistants govern attention before a person makes a decision. This paper examines silent omissions, public evidence gaps and a practical way to test these systems.
Read the paperMaking Autonomous Work Reviewable
Why versioned Markdown and static HTML provide a practical human review boundary as software agents gain more operational autonomy.
Read the paperWhat Return to Office Removes
Remote work widened the lives compatible with engineering. This paper examines return-to-office mandates, lived experience engineering and the consequences of narrowing participation.
Read the paperWhen the Endpoint Is Up but the Product Is Failing
A reliability model for AI products whose behaviour can fail while their infrastructure remains available and inside its latency target.
Read the paperObservability Is a Decision System
Telemetry creates value when evidence reaches a person or system that can make a better decision, not when another dashboard is created.
Read the paperReliability Customers Can Use
Service health becomes a product surface when customers can use it to retry, fail over, pause work, communicate impact or wait.
Read the paperThe Observability Tax
Telemetry volume can grow faster than decision value, increasing storage, query, engineering and cognitive costs.
Read the paperManaged Services Move Reliability Work
Cloud abstraction removes some component operation while introducing dependency, control-plane, recovery, support and exit responsibilities.
Read the paperIncident Command Is an Organisational Interface
Recovery depends on how evidence, authority, ownership and communication move between a failing system and the organisation responding to it.
Read the paperAll papers are available here as static HTML and downloadable Markdown.