Selected Work Display
Relevant source files
The following files were used as context for generating this wiki page:
Selected Work Display refers to the structured presentation of technical projects, generative artifacts, and architectural evidence within the portfolio. The system organizes work into high-level categories—Flagship, Agent Systems, Evaluation, and Tooling—while maintaining a strict "evidence standard" for every claim. This display serves to validate sixteen months of solo building, providing "receipts" for the GestaltView v3 runtime and its associated ecosystem.
The system bifurcates presentation into two primary views: the Selected Work page, which focuses on production systems and methodologies, and the Dynamic Inner World Showcase, which acts as a public gallery for interactive HTML artifacts.
Sources: work.html:36-39, showcase.html:56-61, index.html:105-108
Evidence Standards and Project Lifecycle
Every project or artifact displayed carries a mandatory "evidence label." This classification system ensures that users can distinguish between shipped production code and conceptual roadmaps.
Lifecycle Statuses
| Label | Definition | Context |
|---|---|---|
| Verified / Live | Shipped artifact, test, or observation | Live runtimes, shipped tools |
| Staged | In the build process; documented but not fully deployed | Demos awaiting allowlists, sprint items |
| Planned | Intent or roadmap; not yet built | Future methodology, announced councils |
Sources: work.html:38-39, index.html:125-127, showcase/registry.json:149-158
Project Verification Flow
The portfolio follows a specific loop for project validation: preserving the source, retrieving context, proposing relationship hypotheses, inviting correction, and connecting to action.
The logic ensures that digital intelligence systems do not turn "guesses into facts."
Sources: work.html:39-40, index.html:32-34
Flagship System: GestaltView v3
GestaltView v3 serves as the primary continuity infrastructure. It uses a hybrid retrieval model to preserve fragmented human context and translate it into actionable steps.
Technical Stack
- Client: React/TypeScript
- Server: Vercel serverless API
- Data Layer: Supabase/Postgres with
pgvectorretrieval (768-dim embeddings) - Orchestration: Multi-model routing where cost-aware logic determines whether cheap or frontier models process tasks.
Sources: work.html:46-52
Scale Metrics (August 2026)
- Database: 217 Postgres tables, 299 RLS policies, 75 DB functions.
- Codebase: 722 TypeScript files across separate runtime and corpus repositories.
Sources: work.html:56-58, index.html:101-102
Agent Systems and Interaction
The portfolio highlights Billy, the collaborator agent, which operates as the central node for GestaltView's interaction across multiple surfaces.
Surface Distribution
Billy is implemented across three primary platforms with a fourth in build:
- Web: via Vercel.
- Discord: Commands including
/ask,/status, and/help. - Reddit: Devvit app and polling bot.
- Slack: Staged and awaiting deployment.
Sources: work.html:73-77
Governance and Memory
Agent behavior is governed by Ten Constitutional Invariants: five user-facing commitments and five digital-intelligence commitments. These invariants are machine-readable and shape prompt routing and persistence. Memory is managed via match_knowledge_fragments in Supabase to provide continuity without "reintroduction tax."
Sources: work.html:79-84
Artifact Showcase and Registry
The Dynamic Inner World Showcase is driven by a canonical index file (registry.json) that manages metadata for 11+ standalone interactive HTML artifacts.
Showcase Architecture
The registry.json file serves as the source of truth for the showcase page. It defines the relationships between entries, their "kind" (exhibit, demo, reference, package, etc.), and their runtime status.
The Generative Render Engine
The Render Engine demo implements a contract involving 15 artifact types and 8 synthesis styles. It assembles prompts using a "system block" (style and format directives) and a "user block" (title, source, and PLK mode).
| Component | Description |
|---|---|
| Artifact Types | Markdown, PDF-ready HTML, JSON Blueprints, Session Recaps, Mermaid, etc. |
| Synthesis Styles | Faithful, Convergent, Revolutionary, Technical, Founder Voice. |
| PLK Handling | Light touch, Score only, Full resonance pass, or Off. |
| Provenance | SHA-256 hash + model provenance sealed in a server envelope. |
Sources: showcase/render-engine.html:43-110, showcase/render-engine.html:150-180
Evaluation and Product Tooling
Methodology for testing AI systems is displayed via the Portfolio Throughline Test and the Agent Trainer Sprint.
Portfolio Throughline Test
This eval harness checks system continuity across tasks using:
- Five Local Tasks: Checking for relationship preservation.
- Scoring: A 0–2 rubric across five criteria.
- Failure Taxonomy: Preserved, Collapsed, Distorted, Invented, Overloaded, or Correctly Uncertain.
Sources: work.html:93-98
Agent Trainer Sprint Board
The sprint board displays the revenue pipeline and active bug tracking for the Agent Trainer package. The critical path for revenue involves a four-step node flow.
Note: Payment through Download stages are currently labeled as "blocked" until the pipeline is verified.
The Selected Work Display facilitates a transparent, inspectable view of high-complexity AI systems by combining live runtime access with static design specs and interactive demos. It serves to establish technical authority through visible code metrics and rigorous evaluation standards.
Sources: index.html:125-127, work.html:36-40