AI Launch Evidence Scorecard

The AI Launch Evidence Scorecard is a free launch diagnostic from ResearchAudio. It presents seven checks and is described as taking two minutes, with no login and answers staying in the browser. The scorecard is intended to give a clear read on whether an AI announcement contains enough evidence to support a real decision. It updates as each check is completed and starts with proof that can be verified.

The seven evidence gates cover access, falsifiability, baseline, operating cost, failure cases, reproducibility, and evidence ownership. For access, it asks whether a working product, API, repository, or documented access path exists today. For falsifiability, it asks whether the launch names a number, condition, or outcome that another team can test. It also checks whether comparisons use the same task, tools, budget, and evaluation conditions, and whether price, latency, compute, token use, or another real constraint is disclosed. Failure cases are documented when the announcement states where the system breaks, degrades, or stays untested. Reproducibility requires methods, prompts, data, code, logs, or test artifacts. Evidence ownership and date require a named team, version, and evaluation date.

The scorecard classifies results from 0–2 as Marketing claim, 3–4 as Promising, under-proven, 5–6 as Decision-ready, and 7 as Evidence-complete. It states that a low score is not a verdict but a list of unanswered questions. ResearchAudio describes its broader Evidence Lab as tracing the claim, the method, the cost, and the missing evidence so engineers and product teams can decide without reading twenty launch threads.