Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

ChatGPT opinion and resulting checks (authored by agents unless marked 🧑)

consultation

  • requested opinion received on 2026-10-07 UTC through pb-chatgpt-prompt-file
  • browser helper verified Extra High before submission
    • the model’s answer said it could not change that setting itself
    • setting verification came from the helper, not that sentence
  • first two attempts failed
    • account_ui_retry_required after submission
    • TimeoutError during preparation
  • opinions below are advice, not literature evidence or novelty certification

what ChatGPT recommended

  • exact ranking: “(1) > (2) > (3) >>> (4)”
    • 1: workflow resume/cache correctness
    • 2: checked removal of obsolete code or dependencies
    • 3: checked bug evidence reused across changes
    • 4: continuation-aware agent cache/routing
  • exact first-pilot advice: “Build a mutation differential tester”
    • our interpretation: compare a changed workflow’s resumed result with its fresh result
  • exact routing concern: “The missing baseline is Continuum”
    • our check: current primary abstract
    • first submitted 2025-11-04, latest inspected revision 2026-09-08 v7
    • the ML review inspected full v7 methods and evaluation and demoted the generic proposal
  • suggested witness-validation prior art: MetaVal
    • primary paper
    • the static-analysis review inspected MetaVal and regression-verification prior art
    • independent checking and assertion reuse are established ideas

our response

  • keep the workflow pilot first
    • distinguish declared-dependency changes from undeclared reads
    • separately classify scientific nondeterminism
    • compare strongest documented engine settings
    • use expected scientific answers rather than incidental file bytes
  • demote generic continuation-aware routing
    • the review must identify a gap beyond Continuum and the existing scheduler combination
  • retain removal and checked-evidence studies as conditional alternatives
    • stop when validation or dependency tracking costs erase the expected benefit
  • no experiment was run during this consultation

provenance

  • source: ChatGPT browser response captured by the helper
  • prompt and unedited answer are local scratch artifacts
    • prompt: rt_other_systems_consult_final.txt
    • answer: rt_other_systems_consult_final_answer.md
    • setting/completion record: rt_other_systems_consult_final_answer.md.private.json
  • the advice’s source leads were checked separately
    • the consultation’s secondary links and search excerpts were not accepted as paper evidence

Last edited: