Platform Overview · INST-001
UluOps
A closed-loop system for AI validation and analysis. Four subsystems forming infrastructure for trustworthy machine judgment — from agent definitions to runtime execution to recursive self-improvement.
Zone 1 · System Architecture
Registry
Agent definitions, workflows, pipelines. The institutional memory of the platform.
Ops Tracker
Validation runs, issue lifecycle, taxonomy analytics. Forensic evidence of quality.
Appreciation
Recursive quality compounding. The system improves itself through structured feedback.
SDK
Runtime execution, Claude provider integration, developer-facing tooling.
CLOSED LOOP
Zone 2 · Agent Types
Six cognitive operations
Every agent in the ecosystem is one of six types. Each type represents a fundamentally different cognitive operation — a different question the agent asks about the artifact it's examining.
Explorer
Discovery
Searches for things that exist but haven't been found. Navigates possibility space without a predetermined target.
Asks: "What's out there?"
Executor
Action
Takes validated decisions and executes them. Transforms judgments into state changes in the world.
Asks: "What should happen now?"
Generator
Creation
Produces new artifacts — code, definitions, configurations. Creates things that didn't exist before.
Asks: "What should I build?"
Analyst
Interpretation
Examines artifacts through a specific cognitive lens. Reveals hidden structure, load distribution, and systemic properties.
Asks: "What does this mean?"
Validator
Verification
Tests artifacts against criteria — correctness, security, quality, compliance. Produces scored judgments with evidence.
Asks: "Is this right?"
Forecaster
Prediction
Traces causal chains into states that don't yet exist. Reasons about what artifacts will make true about the world.
Asks: "What will this become?"
Zone 3 · Agent Taxonomy
Four layers of cognition
Agents are organized into cognitive layers based on what they reason about. Each layer operates at increasing levels of abstraction — from concrete artifact properties to epistemological commitments. Click any agent to inspect.
Zone 4 · Failure Taxonomy
Four domains of failure
Every issue discovered by every agent is classified into one of four failure domains. This shared language makes findings comparable across agents, layers, and even across problem domains.
STR
Structural
Problems in how things are built — architecture, dependencies, load distribution, coupling.
→ Accidental levers
→ Load concentration
→ Coupling to internals
SEM
Semantic
Problems in meaning — naming, intent mismatch, misleading abstractions, violated contracts.
→ Intent drift
→ Naming contradictions
→ Abstraction leaks
PRA
Pragmatic
Problems in real-world use — usability, operational risk, deployment concerns, edge cases.
→ Operational blind spots
→ Deployment fragility
→ Edge case gaps
EPI
Epistemic
Problems in knowledge — hidden assumptions, untested beliefs, confidence without evidence.
→ Buried assumptions
→ Untested invariants
→ Cargo cult patterns