Skip to content

Instantly share code, notes, and snippets.

@williamp44
williamp44 / repl-compute-layer-insight-public.md
Created March 13, 2026 15:34
The REPL Compute Layer: Why AI Should Send Code, Not Data

The REPL Compute Layer: Why AI Should Send Code, Not Data

The Problem

Most AI tool integrations follow the same pattern: fetch raw data, paste it into the prompt, ask the model to analyze it. A CSV with 500 rows. A database dump. An API response with nested JSON. The LLM reads every byte, burns tokens parsing structure it cannot see efficiently, and produces a summary that a three-line script could have generated.

This approach is expensive, slow, and lossy:

  • Token cost scales with data volume. 50 Reddit posts at ~500 tokens each = 25,000 input tokens just to score and rank them. At Opus pricing, that is $0.42 per query.
  • Context windows are finite. A 200K context window sounds large until you load a real dataset, a system prompt, conversation history, and tool definitions. The data competes with reasoning space.

Agent Teams Guide: Sequential Ralph Loop vs Native Agent Teams

Date: 2026-02-11


Two Ways to Execute PRDs

1. Ralph Loop (ralph.sh) — Sequential, one task at a time

2. Native Agent Teams (Claude Code feature) — Parallel, multiple agents

@williamp44
williamp44 / PRD_TRADE_ANALYZER.md
Last active February 15, 2026 05:01
Ralph Loop: Example PRD

PRD: Trade Analyzer

Introduction

Analyze trading snapshots from CSV to calculate P&L, win rate, and expected value metrics (Standard EV, Kelly Criterion, Sharpe Ratio) aggregated by ISO week. Pure functional core for calculations with console output.

Goals

  • Load trade snapshots from CSV with type validation
  • Calculate realized P&L by pairing ENTRY/EXIT events
@williamp44
williamp44 / test.md
Created January 28, 2026 15:31
test to make public

test to make public