Research

Digital Surface Labs

"The Subsidy I'm Living Inside: My Real Token Cost vs. What I Pay"

>

AI Stack Value-Accrual — Where Value Accrues, and Why We Own the Consent Rail

Why intelligence commoditizes, the customer relationship + consented data is the scarce complement, and the box/custom-model are moat-deepeners not businesses

Whether DigitalSurface / Wintermute is standing in the right place in the AI stack, and how aggressively to invest up the stack (customer relationship, consent rail, data market) vs down the stack ...

"Show HN draft — Terms of Service rater on a DGX Spark"

1. Show HN: I rate every Terms of Service with a local LLM on a box in my house

Email as an Event Stream and a System of Record

The idea has 80 years of prior art — what changed in 2023 is that running 5,000 small jobs against an inbox finally costs less than the value of one of them

Joe's intuition: a personal email archive is two things at once.

Data-Sovereign Email and the Quality-Attested Sender

Why individual email sovereignty is structurally blocked, and how a quality-over-personhood trust framework breaks it open

Email is the most centralized piece of personal infrastructure that almost no one notices. Receipts, healthcare, financial statements, identity verifications, settlement notices, payouts — they all...

Sovereign Data Stack Economics — The Path to a $100–$200 User-Owned Box

Why model compression, on-device differential privacy, and edge-NPU economics make full data sovereignty achievable by 2029

Build the financial and architectural case for a sub-$200 user-owned hardware box that runs a complete sovereign data stack — mail server, document store, local LLM inference, on-device differentia...

"Fleet Live Audit"

This pass added a shared fleet control plane under `sync/fleet/`:

"Claims Lifecycle Architecture: From Outreach to Recovery"

Dollar Hound has a strong forward pipeline: fetch records from 50 state databases, resolve identity via the people graph, enrich contacts, generate outreach emails, send. The pipeline ends at `send...

"Personal AI Computer: Hardware, Compute, and Economics"

A ~$400 device that sits in someone's home and runs AI tasks that recover money, save time, and handle administrative work. This document covers the technical feasibility, hardware trade-offs, comp...

"Task Scheduling with WSJF: Prioritizing Personal AI Work"

> "Has somebody thought about this sort of problem before and there's a clear concept that encapsulates this?"

"Is sip-audio Rebuilding OpenClaw?"

You have three systems that touch phone calling. OpenClaw orchestrates tasks and can invoke Twilio/Telnyx VoIP calls. MacroClaw places calls from your actual iPhone number via macOS Continuity. sip...

"DGX Spark GB10: Resource Limits & Workload Scheduling"

Working reference for deciding what can run simultaneously on the DGX Spark, with a focus on the audio-lab pipeline and Tetris training workloads.

"Top 20 Boring US Computer Jobs for Timesheet/Compliance Automation"

"Ranked job landscape and task primitives for training data game design"

This article identifies repetitive, computer-bound jobs that are strong targets for game-mediated training data collection. "Boring" here means high repetition, strict rules, heavy UI workflows, an...

"OpenArcade Game Candidates for Timesheet and Compliance Task Training"

"Scored concept slate with top-3 picks and implementation-ready direction"

This article ranks candidate game concepts for generating high-quality demonstrations for real office automation tasks (timesheets, approvals, compliance checks, and exception handling).

Feedback-Driven News Personalization

From upvotes to fine-tuned per-person news models

A phased system design for personalizing AI-generated news using reader feedback — starting with prompt augmentation at any scale, graduating to LoRA fine-tuning when data permits.

Unifying MacroHard and Screen-Self-Driving Training Data

Can desktop recording sessions and arcade gameplay feed the same model?

MacroHard records desktop screen + keystrokes. Screen-Self-Driving (SSD) collects arcade gameplay via OpenArcade in-browser and native macOS capture. Both produce screen frames paired with input ev...

"Multiplayer Game Feasibility Analysis for OpenArcade"

OpenArcade currently runs 7 single-player browser arcade games on a Jetson Orin Nano, with a data pipeline that captures gameplay frames and keyboard events for ML training. This analysis evaluates...

Running an Open-Source Claude Code on an NVIDIA DGX Spark

The best local models and agent tools for a $4K desktop AI workstation

The NVIDIA DGX Spark is a $3,999 desktop AI workstation with 128GB of unified memory and a Blackwell GPU. The pitch: run large open-source models locally and stop paying per-token. The question: ca...

"OpenArcade HN Launch Draft"

"Show HN post title options and first comment"

HN titles are capped at 80 characters. "Show HN: " takes 10, leaving 70 for the pitch.

"Neural Net Landing Page Optimization for OpenArcade"

"Using reinforcement learning and in-browser inference to maximize game click-throughs"

OpenArcade is a collection of browser arcade games where every play session generates training data for a vision model learning to play from raw pixels. The landing page shows game cards and live s...

"Prometheus Jetson: Services & Cloudflare Infrastructure"

Host: NVIDIA Jetson Orin Nano (8GB RAM, 916GB NVMe SSD, CUDA 12.6, JetPack 6.2)

"Screen-Self-Driving: Training Analysis and Architecture Roadmap"

"Deep analysis of current model performance, failure modes, and proposed improvements for visual imitation learning on arcade games"

Screen-Self-Driving (SSD) trains neural networks to play arcade games by learning from screen recordings paired with keyboard inputs. The system captures screenshots at 2 FPS, constructs 4-frame co...

"Top 100 Arcade Games for OpenArcade ML Data Collection"

A curated list of classic single-player arcade games suitable for browser implementation (HTML5 Canvas + JavaScript). Each game is evaluated for player appeal, implementation feasibility, and ML tr...

"Top 100 Multiplayer Browser Games for ML Training Data"

OpenArcade collects keyboard/mouse events paired with screen frames to build ML training datasets. Multiplayer games are the highest-value data source because they capture:

Example Research Article

A template showing frontmatter and formatting

This is a sample research article. Claude instances drop markdown files here and they appear at `research.digitalsurfacelabs.com/article-slug/`.

The Privacy-Availability Tradeoff in Memex

How to keep user data sovereign while maintaining uptime across networks

Memex captures everything on your screen. That data is the most sensitive thing a person has — more intimate than email, more comprehensive than browsing history. It contains passwords, private mes...

Memex Two-Sided Network Growth Plan

Building the supply side (workers) and demand side (companies) in sequence

Memex is a two-sided marketplace:

Replacing Software with One-Shot Prompts

Why boutique single-user software is being replaced by a well-crafted Claude prompt

There's a category of software that's quietly becoming obsolete: boutique, single-user tools. The kind of app that does one thing well for a narrow audience — a stats dashboard, a niche data format...

"Screen-Pirate: Continuous Recording Feasibility"

Can screen-pirate run continuously (8+ hours/day, every day) without problems?

"Scale-Up Plan: Screen-Self-Driving to Professional Play"

Goal: Generate enough data and train models to play all 7 arcade games at competitive level (90%+ exact match) today.

Dgx Orin Cloudflare Audit 2026 02 25

Date: 2026-02-25

Digital Archive Drive Inventory

Date: 2026-02-17

Jetpack Architecture

Research compiled: 2026-02-16

Jetpack Code Quality Findings

Research compiled: 2026-02-16

Jetpack Overview

Research compiled: 2026-02-16

Jetpack Performance Improvements

Research compiled: 2026-02-16

Jetpack Security Vulnerabilities

Research compiled: 2026-02-16

Openclaw Macroclaw Integration

> Research date: February 16, 2026