{
  "version": "0.2.0-preview.3",
  "date": "2026-10-20",
  "scope": "Reply wording checks and communication preference, not universal comprehension or host-wide interception",
  "baseline": "Full real-model build journey remains the historical preview.1 evidence in validation.json",
  "environment": {
    "os": "macOS 15.5",
    "architecture": "arm64",
    "node": "24.12.0",
    "codexCLI": "0.154.0",
    "dsh": "0.2.1-rc.2"
  },
  "checked": [
    "Known jargon, unexplained technical terms, excessive technical density and internal bookkeeping rejected in actual drafts",
    "Plain default; explicit user-background statement limited to the familiar topic, with a plain-language reset",
    "Quoted, hypothetical, question and negated statements cannot establish the tested technical preferences",
    "General programming background does not automatically permit database or operations terminology",
    "Saved preferences survive project reads, separate service processes and product-direction changes",
    "Review receipt bound to the exact draft and communication preference version",
    "Actual native Codex synthetic progress/final replies had passing reviews before emission; rejected drafts were revised",
    "Final code independently rechecked against reproduced defects, including visible formatting variants",
    "153 local Node checks and 6 development-control checks passed",
    "66 clean-distribution Node checks plus DSH dedicated-entry integration passed",
    "DSH clean profile installation uses the shared wording-check implementation"
  ],
  "retainedLimitations": [
    "The assistant must use the review workflow; not every host output channel is forcibly intercepted",
    "Known wording heuristics plus assistant semantic self-review can miss issues; not independent factual certification",
    "Native unattended Codex preference saves required approval and were denied; no approval settings were relaxed to make those calls succeed",
    "Service-level preference persistence was tested, but the denied native saves are not claimed as successful",
    "User-background records are assistant-attributed, not authenticated user consent or action permission",
    "No fresh full native DSH model/UI journey in this iteration",
    "Synthetic dialogue checks do not establish real user comprehension or satisfaction",
    "Model test costs were not reliably available; unknown is not zero"
  ],
  "privacy": {
    "userProjectsIncluded": false,
    "credentialsIncluded": false,
    "rawSessionLogsIncluded": false,
    "checkerMakesAdditionalModelCalls": false
  }
}