Toolchain
The tools I built to run this.
INSERT INTO slopodar (category) VALUES ('Aphorism that sounds good but doesn''t survive scrutiny');
I built a headless API for running bouts programmatically and a set of Go CLI tools for my own workflow. Each tool exists because a specific need arose during the investigation — cost estimation before running experiments, identity verification before trusting results, analysis before drawing conclusions. They're internal tooling, not polished for external use yet. The codebase is open.
The Toolchain
Internal tooling. Public codebase.
pitforge
Agent EngineeringI needed to scaffold agents faster than the UI allowed. Pitforge lints system prompts, runs local streaming bouts, and generates ablation variants.
$ pitforge evolve agent.yaml --strategy ablateView source →pitbench
Cost & LatencyRunning experiments costs money. Pitbench calculates token costs, margins, and latency for multi-turn conversations before spending a credit.
$ pitbench estimate --model sonnet --turns 6View source →pitnet
Agent IdentityPitnet verifies agent identity hashes. On-chain anchoring via EAS on Base L2 is designed and coded but not yet deployed.
$ pitnet verify <attestation-uid>View source →pitlab
Research AnalysisPitlab runs win-rate analysis, first-mover bias detection, engagement curves, and reaction distribution.
$ pitlab survival --data export.jsonView source →Want to run bouts programmatically?
The API code, CLI tools, and everything else are in the public repo.