[Discussion 1/3] Archive native design evidence

evidence/design/review-explore-design-review.json

@@ -0,0 +1,112 @@
+{
+  "$schema": "fleet.design-review.v1",
+  "version": 1,
+  "project": "codevetter",
+  "target": "Native Review and Explore",
+  "surfaceMode": "read",
+  "mode": "preserve",
+  "register": "product",
+  "context": {
+    "product": "PRODUCT.md",
+    "design": "DESIGN.md"
+  },
+  "direction": {
+    "references": [
+      "https://px0.ai/#why"
+    ],
+    "probes": [],
+    "selected": "existing-design",
+    "approval": "not-required",
+    "before": "evidence/design/native-acceptance-2026-09-01/review-findings.png",
+    "library": {
+      "strategy": "upstream-first",
+      "primary": "existing-project-system",
+      "sources": [],
+      "runtime": "existing",
+      "customReplacement": {
+        "used": false,
+        "authorization": "not-required",
+        "reason": ""
+      }
+    },
+    "contract": {
+      "purpose": "CodeVetter helps developers verify agent-authored changes through reproducible execution evidence and exact source navigation.",
+      "purposeSource": "PRODUCT.md; owner PRD in GitHub issue #285",
+      "canonicalPurpose": "CodeVetter determines whether a coding agent completed a software task correctly using reproducible execution evidence.",
+      "purposeAlignment": "match",
+      "driftNote": "",
+      "audience": "Developers and technical owners reviewing agent changes in unfamiliar repositories",
+      "job": "Paste a GitHub URL, inspect exact source or diff, follow architecture and evidence locations",
+      "thesis": "Preserve the Evidence Bench and put pinned source at the center of a compact native workbench",
+      "system": "Existing true-black/light native palette, bronze focus, 12-point monospace source, hairline pane separators, compact native controls",
+      "signature": "Persistent commit and blob identity connect every source window to its evidence",
+      "risk": "A dedicated AppKit source plane trades editor features for bounded viewport rendering; keep it read-only and qualify accessibility separately"
+    }
+  },
+  "evidence": {
+    "screenshots": [
+      {
+        "width": 390,
+        "path": "artifacts/navigator-review/landing-390.png"
+      },
+      {
+        "width": 768,
+        "path": "artifacts/navigator-review/landing-768.png"
+      },
+      {
+        "width": 1440,
+        "path": "artifacts/navigator-review/explore-1440.png"
+      }
+    ],
+    "projectCheck": {
+      "command": "pnpm navigator:test; XcodeBuildMCP swift-package test; XcodeBuildMCP macos build (Debug)",
+      "status": "pass"
+    },
+    "critique": {
+      "score": 33,
+      "maximum": 40
+    },
+    "audit": {
+      "score": 17,
+      "maximum": 20
+    },
+    "unresolved": {
+      "p0": 0,
+      "p1": 0
+    },
+    "detector": {
+      "posture": "advisory",
+      "findings": [
+        "Native window qualification uses 980, 1280 and 1440 points. 390 and 768 captures test the isolated import surface, not a supported macOS window size.",
+        "Impeccable is not installed; native layout critique, contrast inspection, offscreen captures and build/tests were used.",
+        "Foreground keyboard and VoiceOver interaction remains unqualified under repository idle-screen rules."
+      ]
+    },
+    "comprehension": {
+      "status": "not-applicable",
+      "reviewer": "",
+      "answers": {
+        "product": "",
+        "audience": "",
+        "value": "",
+        "mechanism": "",
+        "proof": "",
+        "nextAction": ""
+      },
+      "purposeScore": {
+        "product": 0,
+        "audience": 0,
+        "value": 0,
+        "mechanism": 0,
+        "proof": 0,
+        "nextAction": 0,
+        "total": 0
+      },
+      "mismatches": []
+    }
+  },
+  "ownerFeedback": {
+    "decision": "agent-selected",
+    "note": "The agent selected the direction; no owner approval was required."
+  }
+}

evidence/design/review-scope-2026-09-15/design-review.json

@@ -0,0 +1,115 @@
+{
+  "$schema": "fleet.design-review.v1",
+  "version": 1,
+  "project": "codevetter",
+  "target": "Review branch/base selection and scope-bound setup",
+  "surfaceMode": "operate",
+  "mode": "preserve",
+  "register": "product",
+  "context": {
+    "product": "PRODUCT.md",
+    "design": "DESIGN.md"
+  },
+  "direction": {
+    "references": [
+      "PRODUCT.md",
+      "DESIGN.md"
+    ],
+    "probes": [],
+    "selected": "existing-native-review",
+    "approval": "not-required",
+    "before": "evidence/design/native-acceptance-2026-09-01/review-intent.png",
+    "library": {
+      "strategy": "upstream-first",
+      "primary": "existing-project-system",
+      "sources": [],
+      "runtime": "existing",
+      "customReplacement": {
+        "used": false,
+        "authorization": "not-required",
+        "reason": ""
+      }
+    },
+    "contract": {
+      "purpose": "CodeVetter helps developers verify agent changes using reproducible execution evidence.",
+      "purposeSource": "PRODUCT.md; owner-requested Review workflow repair in issue #285",
+      "canonicalPurpose": "CodeVetter determines whether a coding agent completed a software task correctly using reproducible execution evidence.",
+      "purposeAlignment": "match",
+      "driftNote": "Site Health has no purposeContract; its portfolio boundary agrees. Usage remains bounded side work.",
+      "audience": "The operator managing coding-agent usage",
+      "job": "Choose a branch and base, inspect the exact diff, then start an explicit scope-bound review",
+      "thesis": "Keep the existing source workbench and expose comparison controls directly above its diff",
+      "system": "Existing native pickers, compact chrome, source layout and restrained semantic colors",
+      "signature": "Displayed merge-base/head becomes the locked verification input",
+      "risk": "Do not silently substitute a stale range or execute against an unrelated checkout"
+    }
+  },
+  "evidence": {
+    "screenshots": [
+      {
+        "width": 390,
+        "path": "evidence/design/review-scope-2026-09-15/scope-controls-390.png"
+      },
+      {
+        "width": 768,
+        "path": "evidence/design/review-scope-2026-09-15/scope-controls-768.png"
+      },
+      {
+        "width": 1440,
+        "path": "evidence/design/review-scope-2026-09-15/scope-controls-1440.png"
+      }
+    ],
+    "projectCheck": {
+      "command": "XcodeBuildMCP swift-package test (109 passing); cargo test branch_picker; cargo clippy --all-targets",
+      "status": "pass"
+    },
+    "critique": {
+      "score": 34,
+      "maximum": 40
+    },
+    "audit": {
+      "score": 17,
+      "maximum": 20
+    },
+    "unresolved": {
+      "p0": 0,
+      "p1": 0
+    },
+    "detector": {
+      "posture": "advisory",
+      "findings": [
+        "Manual native fallback review, not Impeccable output; see evidence/design/review-scope-2026-09-15/review.md.",
+        "Native full-window checks use 980/1280/1440 points in both appearances; 390/768 isolate the comparison control.",
+        "The before image is the historical verification-setup visual reference, not a pixel baseline of the pre-repair navigator.",
+        "No foreground interaction or end-to-end agent execution in this invocation. Uncommitted changes are browse-only for the existing committed-source verification engine.",
+        "First broader suite hit known Usage render variance: 51.655 ms vs unchanged 50 ms; final results recorded separately."
+      ]
+    },
+    "comprehension": {
+      "status": "not-applicable",
+      "reviewer": "",
+      "answers": {
+        "product": "",
+        "audience": "",
+        "value": "",
+        "mechanism": "",
+        "proof": "",
+        "nextAction": ""
+      },
+      "purposeScore": {
+        "product": 0,
+        "audience": 0,
+        "value": 0,
+        "mechanism": 0,
+        "proof": 0,
+        "nextAction": 0,
+        "total": 0
+      },
+      "mismatches": []
+    }
+  },
+  "ownerFeedback": {
+    "decision": "keep",
+    "note": "Preserve existing design. Owner requested the specific dropdown and review-entry repair: fix thing please. No visual overhaul."
+  }
+}

evidence/design/review-scope-2026-09-15/review.md

@@ -0,0 +1,70 @@
+# Review comparison repair — 2026-09-15
+
+Tracking: [#285](https://github.com/Codevetter/codevetter/issues/285), tasks 9–11.
+Owner request: branch/base dropdowns after selecting a repository and an obvious
+way to ask for a review. Implemented locally; no commit, push, installation or release.
+
+## Design decision
+
+Preserve lane: reuse the current native shell, source layout, compact pickers,
+typography and colors. The owner supplied the specific interaction to repair;
+this is not a new visual system. Historical verification-setup styling reference:
+../native-acceptance-2026-09-01/review-intent.png (not a pixel baseline of the navigator).
+
+The comparison bar now sits above the diff. Local and remote-tracking branches
+are discovered read-only, bounded to 2,000 refs. Show diff pins merge-base/head
+without checkout or fetch. Review change opens the existing verification setup
+with that exact range locked; instructions and reviewer remain editable.
+Changing a comparison disables eligibility for an old plan. Refresh retains
+the applied branch/base. Repository changes clear scope and prior proof.
+
+Local changes are explicitly worktree-vs-HEAD inspection. The existing verifier
+requires committed source and a clean checkout at the selected head, and does not
+silently verify another branch. Remote PR/commit context stays pinned; imported
+source without a base is not falsely presented as a change comparison.
+The Cancel action is shown only for an active review, not background metadata reads.
+
+## Verification
+
+- Final `pnpm test:native:background`: 109 Swift tests, 15 navigator Rust tests,
+  all five isolated performance gates, native Debug build passed. Two explicit
+  live-network navigator tests were skipped.
+- Focused regression covers branch catalog, exact base/head handoff, pending
+  selection, refresh, missing refs, repository switch and old-plan eligibility.
+- Rust fixture verifies divergent-branch merge-base, detached HEAD, unrelated
+  history rejection, invalid refs, and unchanged original HEAD/index.
+- Final branch test and navigator Clippy with warnings denied passed after the
+  multiple-merge-base rejection refinement; navigator library rebuilt.
+- Documentation and diff whitespace checks passed. Complexity gate reported
+  no applicable files (it does not establish Swift/Rust complexity coverage).
+
+The first broader run failed the existing Usage hosted-render gate at
+51.655 ms p95 versus the unchanged 50 ms budget. Final isolated Usage p95 was
+30.054 ms; all final gates passed. Preserve this variance under #288 rather than
+discarding the failed observation. No Usage performance code or threshold changed
+as part of the Review repair.
+
+## Native visual review
+
+Screenshots are synthetic repository fixtures, not a real agent review or
+verification verdict. Full-window captures cover 980, 1280 and 1440 points in
+dark and light appearance. The 390/768 captures isolate the responsive comparison
+bar; they are not claims of whole-app phone-sized support. The native minimum
+window remains 980×640. Setup controls remain in the existing scrollable form.
+
+The full-suite offscreen pass showed missing cached text in some screenshots;
+the isolated render test was rerun and its complete renders were inspected and
+retained here. This is a render-harness observation, not proven installed-app behavior.
+Plan may appear disabled during fixture background initialization; eligibility
+is separately tested, and no agent or executable review was started by this test.
+
+Manual native fallback critique: **34/40** (continuity 9, hierarchy 9, scope clarity 9,
+compact layout 7). Audit: **17/20** (read-only scope 4, identity/handoff 4, state
+honesty 4, native label/layout inspection 3, interaction qualification 2).
+No observed P0/P1 in the scoped checks. These are manual judgments, not
+Impeccable output or full accessibility certification. Impeccable is unavailable.
+The preserve-lane design receipt validator passed.
+
+Foreground dropdown clicks, keyboard-to-pixel timing, and end-to-end provider
+execution remain unqualified on this invocation; no fresh idle-screen approval
+was inferred. No additional VoiceOver work. Existing Usage changes were preserved.