
Offer dropped in the bag.
Kane confirmed a product defect with 98% confidence. Its first generated test had misread the cart; the reviewed replacement read the displayed discount line and total.
K9 tested one real shopper decision: does a 20% first-order offer survive the cart for $189 Halo Wireless Headphones? The failed run exposed a $0 discount. The same Kane-generated test passed after the repair.
The same challenger journey moved from a confirmed defect to the expected discount and total. These images are unchanged screenshots from the sealed Kane packs.

Kane confirmed a product defect with 98% confidence. Its first generated test had misread the cart; the reviewed replacement read the displayed discount line and total.

The reviewed challenger test passed before any demo order. A later suite replay passed both challenger and control journeys and sealed one validated pack.
The coding agent used Kane’s output to correct the test, fix the campaign copy, then repair the challenger default and rerun the test.
The first test calculated the advertised discount instead of checking the cart. Kane’s screenshot showed the real $0 discount.
Kane generated reviewed replacement t-4 to read the cart line and total before any order.
The agent clarified the automatic offer and changed the challenger default. Kane then passed the same test.
The assurance loop ingested the merchant brief, designed approved tests, executed them in Chrome, and measured coverage. K9 can use Kane as its live verification step.
Reviewed _test.md files and replay recordings let another machine run the same journey.
Challenger and control were run together. One validated evidence pack holds both results.
View original pack ↗NDJSON results and pack integrity decide the job result. GitHub secrets and a required branch check still need setup.
View workflow ↗Start the storefront, then replay the approved suite. Kane sign-in is needed for the browser step.
npm run dev
npm run kane:ciCoverage is 3 of 4 criteria for the focused discount use case and 13.4% across the broader six-use-case brief. The remaining criterion was deferred; this page does not claim full project coverage or an active GitHub merge rule. Full packs contain Kane account metadata and are available in the public GitHub repository.