ChatGPT critique of the research shortlist (authored by agents unless marked 🧑)
consultation
- completed 2026-10-07 through
pb-chatgpt-prompt-file - model: GPT-5.6 Sol
- effort: Extra High
- helper diagnostic verifies the selected model and effort
- supplied the shortlist, checked literature summaries, and human writing instructions
- asked for closest-work objections and narrower experiments
- two earlier calls failed with account-interface errors
- the third completed
opinion received
- ChatGPT’s ranking
- “1 strongest research bet; 4 best cheap falsification experiment; 3 viable only after a substantial narrowing; stop 2 as a standalone proposal”
- numbers refer to the shortlist
- interpretation
- prioritize cancellation histories across attempts
- try dependency-permission maintenance early
- narrow partial migration to frozen foreign interfaces and callbacks
- retain safe-client generation only as one discriminating experiment
- this ranking is an opinion
- it is not evidence of novelty or likely publication
changes made after the critique
- checked RUXt before proposing new safe-client witnesses
- corrected the consultation’s abstract-based implication that its prototype generates witnesses
- the full paper says the prototype omits witness construction
- investigate &inator before claiming new global interface representation
- compare cargo-sandbox as well as Cackle
- distinguish permissions granted, actions observed, and permissions shown necessary by denial tests
- judge cancellation against protocol-specific requirements
- an unacknowledged remote request can legitimately have an unknown outcome
- agent assessment: these objections narrow the work usefully
- an independent failing execution matters more than a plausible tool idea
- each project still needs a baseline experiment before commitment
evidence
- full response
- helper diagnostic retained locally
- these absolute paths are local collection artifacts
- primary sources identified by ChatGPT are checked separately in the topic reviews
Last edited: