INPUTS. OUTPUTS. LIMITS.
What we have actually tested
Fixed cases, first outputs and individual acceptance checks. Access attempts stay visible without becoming quality scores.
Current recorded coverage
8 distinct feature cases executed across 4 of 42 listed tools: 2 passed, 6 failed, 0 partial. Another 8 tools have access observations without submitted product-task outputs.
86 fixed cases are defined; 78 remain unexecuted. Public utilities and the signed-in Pomelli campaign feature have separate scopes. Real-SKU comparisons and connected store actions remain pending. Each case is counted once using its latest actual output; reused uAgentKit records do not create new executions.
Read the testing method · Editorial responsibility · Download all fixed plans
Executed features
2026-10-04 · 2 initial submissions · Codex browser automation · Public utility
Two initial public-utility submissions with a synthetic desk-mat fact sheet. No catalog workspace, ecommerce agent workflow, bulk generation or store publishing was tested.
Both first outputs missed the fixed requirements: the normal case changed material wording and omitted bullets; the boundary case omitted evidence-gap and no-publication explanations.
Download the complete run
2026-10-05 · 2 initial submissions · Codex browser automation · Public utility · Existing uAgentKit result reused
First free two-step patentability screens for a synthetic removable sensor bottle and an explicitly absent novel feature.
0/2 complete frozen cases passed; each passes 5/6 listed conditions. The 59/100 primary screen lacks native original-document links; the 0/100 boundary does not explicitly acknowledge the supplied missing feature before assessing it.
Download the complete run
2026-10-05 · 2 initial submissions · Codex browser automation · Signed-in UI
Synthetic ecommerce campaign in the existing signed-in Pomelli workspace. Existing RoboSkin business profile is retained; this measures explicit product-brief handling under that configuration, not brand onboarding or a native marketplace integration.
The first images retained unrelated brand imagery; the boundary brief kept unsupported claims visibly unverified.
Download the complete run
2026-10-05 · 2 initial submissions · Codex browser automation · Public utility
Two known-item native search checks. These additional utility cases do not execute or replace the prepared broad prior-art screening cases, and do not evaluate an AI agent or legal clearance.
Known-number lookup passed its three conditions and exposes the original publication. The exact number-plus-date query failed both cutoff conditions: it used a known-number redirect and still opened the 2003 document. This observation does not establish that all Google Patents date filters fail, or provide prior-art or infringement clearance.
Download the complete run