The Build Queue opened August 1 with one priority: a CarrierIQ correction-to-next-run record showing a wrong output, diagnosis, interface change, release criterion, and verified subsequent run. August 9 kept it at the top and added: "Do not add another demo first." August 12 sharpened the form to a decision rehearsal. August 19 added authority and intervention power at every review state. August 22's three-claim panel separated what must be proven — user corrected the output, system incorporated it, next run improved — into distinct claims.
Each cycle made the specification more precise and harder to satisfy with a vague mockup. The artifact page at junochen.com looks the same as it did three weeks ago.
Three housekeeping items flagged before the queue opened — a private essay still publicly retrievable, an unsettled third Lab on the homepage, a TinyFish attribution on Retail Velocity — remain open. These require edits, not builds. They're still not done.
Render, Staff Product Designer — Agent Experience: Covers human-agent handoffs and verified next-run loops. CarrierIQ's review states are relevant but stop at the exact sequence the role tests for. No correction-lineage artifact to anchor outreach on.
OpenAI, Product Design Leadership — Growth: No forward-looking visual artifact exists in any of the four domains. The Trust essay provides the framework; the visual deliverable that should lead hasn't been produced. Outreach currently defaults to the support layer.
Any interview panel asking "show how a human correction changes system behavior": CarrierIQ covers the first half. The second half — changed behavior, verified next run — can only be answered verbally through TinyFish experience, which can't appear in written materials.
All AI-native outreach: The portfolio strategy calls for forward-looking artifacts as the lead layer. That layer doesn't exist. Agentic Labs is carrying both roles — proof of production and proof of frontier thinking — when it was designed for only the first.