Submitted builds · Ring the Changes
What the review found
These versions are preserved for comparison, including submissions the reviewer would keep private as tools.
| Version | Observed findings | Reviewer recommendation |
|---|---|---|
GPT-6.1 Sol Anonymous C · High | Fading guidance and the trace of actual strikes held up in synthetic completion tests. The tap button initially sits below view in a phone viewport. | Preferred; publishable browser experiment. |
Claude Opus 5.5 Anonymous B · High | Rich method study and lead feedback. Extra pulls disappear from final scores; the hidden ringing stage stays visible on setup and results. | Keep private until scoring and screen state faults are corrected. |
Gemini 4 Argon Anonymous A · High | Correct method rows, but reversed dodge labels. Missed strikes acquire an idealised visitor trace; place controls replace actual timing, and live settings leave stale targets. | Keep private as a prototype with teaching and performance feedback faults. |
What the checks establish
Chrome desktop and emulated phone, source checks and synthetic inputs. No actual listening, physical ringing, test on a real phone or evidence of human learning. A passing rules verifier missed several UI and scoring faults.
Review and hosting are separate
The review assessed frozen submissions. Opus runtime was preserved at its quota stop; Sol supplied missing documentation and a verifier. Sol also replaced a forbidden dialog form, secured links and made one self-link explicit for hosting. The reported scoring, teaching and layout faults remain.
Hosting and packaging: GPT-6.1 Sol at Extra High. Original submissions are retained separately; each linked ledger records original and final file hashes and exact integration additions.