1. DeltaV Lab — Trajectory Sandbox & RK4 Integration

• The Architecture: Separated the numerical physics loop from rendering by running fixed-timestep RK4 ($dt = 0.02,\text{s}$, 50 Hz) inside a dedicated Web Worker. Telemetry synchronizes via SharedArrayBuffer so DOM repaints never pause integration. Includes a custom WHEN <condition> THEN <action> DSL flight computer, pre-launch Go/No-Go checklists, and Flight Termination System interlocks.
• The Fuckup & Root Cause: I initially passed hundreds of discrete coordinate objects through standard postMessage on every frame. This flooded the event queue, stalled the worker, and ruined integration timing. I fixed it by switching to contiguous shared memory buffers.
• Honest Limitations: DeltaV Lab is a 2D pedagogical sandbox with an exponential atmosphere (7 km scale height) and simplified Dryden wind gusts. It is not 6DOF flight-certified software with published validation data.
• Deep Transmissions: Why It's Not Professional Grade • Native Physics Core Migration

2. Vellor — Zero-Knowledge Local Tutoring OS

• The Architecture: Runs entirely on the client device. All data is encrypted with AES-GCM before writing to local client storage (IndexedDB via localForage). Features virtualized student rosters and $O(N)$ operations for instant lookups on low-end hardware.
• The Fuckup & Root Cause: I almost derailed the project by building RPG-style gamification, badges, and levels. That visual noise cluttered the core payment and lesson logging path. I stripped celebration chrome from the transaction flow and moved gamification into a separate optional view.
• Honest Limitations: Zero-server architecture means no automatic multi-device synchronization. Users own their disaster recovery through JSON exports.
• Deep Transmissions: The Gamification Trap • Local Backup Reality

3. The Infinite Intelligence — Multi-Model Peer Review

• The Architecture: Breaks prompts into first-principles constraints and coordinates parallel, sequential, or round-robin debate cycles between distinct personas, with Human-in-the-Loop steering to guide synthesis.
• The Fuckup & Root Cause: I ran an unbounded debate test without recursion caps using vague critic prompts. The agents argued in a loop and drained API key rate limits in minutes. I added hard round ceilings, live token tracking, and cost estimation telemetry.
• Honest Limitations: Client-side BYOK tool. Token costs scale with agents $\times$ rounds $\times$ prompt size.
• Deep Transmissions: BYOK Economics & Guardrails • Why a Council of Models

4. FuelDrop — On-Demand Fuel Logistics Interface

• The Architecture: Delivery interface prototyping customer orders, captain job dispatching (/captain), emergency surge logic, and fleet bulk orders.
• The Fuckup & Root Cause: I initially described this project on my portfolio using words that implied production WebSocket backends and database transaction rollbacks. That was inaccurate. The repo is an interactive client demo running on simulated OTP (1234) and mock state. Stating the exact boundary between prototype and production infrastructure is mandatory.
• Deep Transmissions: UX Prototype vs. Real Logistics • Client-Only Tech Stack