Audit bundle for Kingy workflow 1.5.1, NOT an activated buying recipe. All three Kingy runs failed frozen support and citation thresholds. The active download remains workflow 1.4. No fresh 1.5.1 installation is claimed.

All 240 original attempts, native conversations, model logs, candidate, common scorer, source corpus, pre-output references and settings are preserved. Corpus authored by a separate agent with no prior study history, then sealed before output access. This is agent independence, not human validation or independent physical testing. The corpus is public and consumed now; reproduction cannot be described as unseen qualification.

Native public GUI control transport was amended after A1 and before the first native held-out question; candidate/corpus/prompts/scorer/settings remained unchanged. Control setup errors are private retained evidence, not competitor reliability results. Each question has one original response. Semantic uncertainty remains in the frozen bounds. No general quality, privacy, installation, energy, cloud cost or hardware-upgrade advantage is established.

Frozen scripts retain their measurement-workspace paths and are audit materials, not a turnkey native GUI runner. Model weights, runtime binaries, credentials, and unrelated clipboard content are excluded. Candidate code: MIT; synthetic corpus: CC0-1.0; model: Apache-2.0.
