Labs
Lab reports on agent architecture. Each one states its hypothesis, method and results, says what went wrong, and publishes the repo, the data and the seed.
-
Asking a model to write a probability throws away what it already knows One forward pass, two readouts. Read as a word the model said No to all fifty tickets and ranked at chance. Read as a distribution over that same token it ranked all fifty correctly. Plus the corpus arithmetic that decides whether any calibration number you read this year means anything. 12 min -
A second attempt was worth six points. Deciding when to take one was worth one. Lab-001 changed two things at once and reported a seventeen point gap. This varies them independently across five arms, reproduces both original numbers exactly, and finds that the metric Lab-001 called routing quality was measuring agreement with a keyword map. 9 min -
The poison was stored, retrieved and cited every time. What changed was what it was allowed to cause. An agent reads an incident ticket carrying a false causal claim, remembers it, and meets a matching incident later with a clean ticket. Four configurations, measured on three separate things: whether the claim is stored, whether it is cited, and whether it reaches a forbidden action. 9 min -
Prompt hardening moved three models three different ways. The gate moved all of them to zero. Fifty support tickets, thirty carrying injected instructions, run through three guard configurations on three local models. Compliance and reachability are measured separately, because they are not the same number and only one of them is under your control. 10 min -
The baseline got the right answer and emitted one dispatchable action in thirty runs Lab-002's sixty committed run records, scored twice. Once on the answer and once on the path. The two metrics rank the two systems in opposite directions, and the single assertion that explains it is whether the proposed remediation could actually be executed. 7 min -
The durable agent, and whether ADK 2.0's crash-resilient runtime earns its overhead A static router and a dynamic, human-gated ADK 2.0 Workflow graph run over 30 synthetic incidents on a local model. The multi-agent graph loses on accuracy at 21x the cost, but its durable runtime recovers every crashed run for free. 12 min -
Multi-agent vs router on 100 customer-support queries 100 real support queries from a public dataset, run twice — once through a single-agent workflow router, once through a 3-agent hierarchical multi-agent system. Measured against the same eval rubric. 12 min