Back to Prototypes

Prototype 03

Conceptual 38% mapped

Interpretive Loop Diagrams

Visual frameworks for making the interpretive layers of AI systems explicit. Maps conversation quality metrics, satisfaction indicators, and the iterative refinement cycle between human intent and model behavior.

OODA Cycles & Iterative Reasoning

Observe, Orient, Decide, Act — at multiple levels of abstraction

SPINE

ooda_loop.py

OODA executor: Observe, Orient, Decide, Act with API tool usage. Context stacks (6 layers), MemoryHooks integration.

8do-mcp

Ralph Loop

Verdict-gated task orchestration. Evidence-based verification (lint, typecheck, test, build) with ACCEPT/REVISE/REJECT gates. Circuit breaker safety.

8me-mcp

Loop-until-done

Ralph Wiggum autonomous loop — predecessor to 8do-mcp. Tier-based deliverables (0-5.5) showing iterative progression.

Five-Point Protocol

Framework

Reproducible execution protocol: Clarify, Scope, Plan, Execute, Verify. Transforms intent into structured execution — the core "how we work" pattern for agentic workflows.

Quality Metrics & Satisfaction Indicators

Measuring output quality at every stage

evaluation-mcp (Evalla)

6 tools. Rubric-based scoring (0-100), batch scoring, ranking (MinMax/ZScore/Percentile), verdicts with reasoning. Confidence decay modeling.

content-analyzer-mcp

9 tools. Content quality scoring, aesthetic classification, semantic outlier detection (Isolation Forest), Gemini multimodal vibe checks.

showcase-mcp

check_public_readiness — 5-category validation including humility audit.

Context Engineering & Layered Interpretation

How context flows through structured layers to shape model behavior

SPINE Context Stacks

6-layer structured context: global, character, command, constraints, context, input. Enrichment sections for progressive refinement.

context-glue-mcp

Budget-managed context assembly, scenario composition with metadata, multi-scenario merging with dedup, platform-specific vocabulary (Claude/GPT/Gemini/Grok).

observation-workbench

Recursive reasoning: Observation, Route Resolver, Proposed Avenues, Execute Templates, Record Output + Verdict. 248-template library.

Visual Frameworks & Diagramming

Making system behavior visible and navigable

Tachylite

~5,700 line visual workflow canvas. Node-based editor, browser experience store, wiki-style linking, presentation mode.

tachylite-mcp

27 tools. Headless workflow CRUD: create, add_node, connect, auto_layout (grid/hierarchical/force), import Mermaid, describe in natural language.

browser-mcp

Click memory system — records interaction patterns, exports to Tachylite workflows.

intelligence-engine

Visual knowledge graph exploration via web UI. AST to semantic representation to visual code relationship browser.

Dashboard Infrastructure

Four production systems providing visual layers — not yet federated into a single view

intelligence-engine

Web UI :8420

Sigma.js interactive knowledge graph, 6-tab dashboard (timeline, phases, health, compare, quality, AI memory), 33 REST endpoints. 53,000+ entities across 272 projects.

portfolio-ops-hub

Web UI :8000

vis.js project relationship graph (183 projects, 207 relationships), 6-tab ops dashboard (overview, projects, knowledge, graph, intelligence, memory). Real-time scan/rebuild.

spine-dashboard

FastAPI + React

Orchestration monitoring: run history, agent activity, task queue, token/cost metrics, trace viewer, log stream. Federated UI communicating with SPINE via HTTP API.

Tachylite

Canvas Editor

~5,700 line visual workflow canvas. Node-edge editor with auto-layout (grid/hierarchical/force), Mermaid import, 27 MCP tools. Zero dependencies.

Prompt-as-Object Reasoning

Treating prompts as objects to be reasoned about, not just instructions to execute

Project-Prompt-as-a-Tool

Dynamic prompt-to-tool conversion via MCP. Sequential chaining, Jinja2 templates, runtime config. Prompts become composable, chainable objects.

content-mcp

PromptContract pattern — prompts as traceable, evaluable specification objects with lifecycle tracking.

Interpretive Loop Architecture

Interpretive Loop Architecture diagram showing Human Intent flowing through SPINE Context Stacks into three parallel loops (OODA, Ralph, Five-Point Protocol), converging at a Quality Gate with ACCEPT, REVISE, and REJECT outputs

Capability Coverage

Claim Implementation Status
Visual frameworks tachylite-mcp (27 tools), browser-mcp click memory Production
Interpretive layers explicit SPINE context stacks (6 layers), context-glue-mcp Production
Conversation quality metrics Evalla rubrics, content-analyzer-mcp scoring Production
Satisfaction indicators Evalla verdicts (accept/revise/reject) Production
Iterative refinement cycle SPINE OODA, 8do-mcp Ralph Loop, Five-Point Protocol (framework) Production
Human intent to model behavior context-glue-mcp scenarios, SPINE context stacks Partial
Prompts as objects Prompt-as-a-Tool, content-mcp PromptContracts Production
Unified visual dashboard IE graph (Sigma.js) + portfolio-ops-hub (vis.js) + spine-dashboard (React) + Tachylite canvas — not yet federated Partial

Remaining Gaps — Path to 100%

Federated dashboard entry point

Single UI composing IE graph, spine-dashboard widgets, portfolio-ops-hub overview, and Tachylite canvas

Candidates: React shell with iframe/module federation

Live conversation metrics

Instrument SPINE OODA to emit quality/satisfaction signals per cycle

Candidates: Evalla hooks in OODA + spine-dashboard polling

Refinement visualization

Before/after of each OODA iteration as a diff diagram

Candidates: tachylite-mcp workflow generation

Context layer inspector

Visual breakdown of each of SPINE's 6 context layers

Candidates: context-glue-mcp + tachylite-mcp

The building blocks exist and are individually production-ready: 4 dashboard UIs, OODA executors, evaluation rubrics, context stacks, and workflow canvases. The remaining gap is federation — composing these into a single entry point. 14+ projects compose this prototype.

Prototype 01

Research Agent Suite →

Prototype 02

Creative Engine →

Experiments

Lab →