Skip to Main Content

Aperture — See What Your Agent Actually Saw. AI agents often fail not because information is unavailable, but because critical evidence gets lost when execution histories are summarized or truncated. Inspired by a real engineering challenge documented by Raindrop, we built Aperture, an evidence-aware AI agent harness that investigates failures in multi-agent systems. Using the AEGIS dataset, Aperture starts with compact execution histories, selectively retrieves missing evidence, and verifies that every fault attribution references information actually delivered to the investigator. In one experiment, short context missed a faulty agent (F1: 0.667). Aperture recovered the missing evidence and correctly identified both faulty agents (F1: 1.0). Unlike ordinary retrieval systems, Aperture maintains an exact record of what the model received and enforces citation provenance through a deterministic gate outside editable retrieval policies. Our goal is to make agent investigations more reliable, transparent, and auditable. The prototype demonstrates evidence recovery and source verification, not general accuracy or cost improvements.

Team

not_a_goddess

More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon

MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

MongoMation project preview

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

LabRat project preview

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Recursive Harness Scientist project preview

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

SuperBid project preview

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…

WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…

WorldSTATE project preview