{
  "version": "0.1",
  "review_date": "2026-06-18",
  "public_boundary": "The atlas publishes scrubbed pattern labels only. Raw source excerpts, source titles, UUIDs, emails, people, client names, private paths, screenshots, and exact routing rules are withheld.",
  "failure_modes": [
    {
      "id": "privacy_leakage_risk",
      "label": "Privacy leakage risk",
      "definition": "The agent treats disclosure as something to clean up after writing, rather than a constraint that shapes the artifact from the first step.",
      "archive_signal": "Public-facing work improved when synthetic fixtures, withheld fields, and public/private/mixed classification were declared up front.",
      "mitigation": "Require a boundary statement before file creation and fail the run if private identifiers are used as proof.",
      "benchmark_cases": [
        "private-boundary",
        "public-safe-repo",
        "client-communication-rewrite",
        "billing-reconciliation"
      ]
    },
    {
      "id": "lost_settled_decision",
      "label": "Lost settled decision",
      "definition": "The agent reintroduces an architecture, claim, or product direction the user already rejected.",
      "archive_signal": "Corrected assumptions stayed alive only when they were promoted out of chat and into a durable decision artifact.",
      "mitigation": "Maintain a settled-decisions file and make future architecture or copy generation read it first.",
      "benchmark_cases": [
        "settled-decisions-preservation",
        "architecture-adjudication",
        "anti-template-design-governance"
      ]
    },
    {
      "id": "ungrounded_synthesis",
      "label": "Ungrounded synthesis",
      "definition": "The answer is fluent and plausible, but the evidence lanes behind it are unclear.",
      "archive_signal": "The strongest syntheses separated source inventory, local review, outside research, assumptions, and final recommendation.",
      "mitigation": "Force the output to label verified facts, inferences, missing evidence, and first executable action.",
      "benchmark_cases": [
        "evidence-humility",
        "evidence-lane-synthesis",
        "status-synthesis"
      ]
    },
    {
      "id": "tool_overreach",
      "label": "Tool overreach",
      "definition": "The agent treats tool availability, API capability, or execution permission as stronger than the evidence supports.",
      "archive_signal": "Capability claims became fragile whenever the answer blurred model behavior, harness behavior, and operational integration.",
      "mitigation": "Separate model, harness, tool, permission, cost, and workflow fit before recommending a runtime.",
      "benchmark_cases": [
        "tool-comparison",
        "runtime-router",
        "managed-workspace-architecture"
      ]
    },
    {
      "id": "generic_design_drift",
      "label": "Generic design drift",
      "definition": "The agent produces a polished artifact that looks correct in a vacuum but loses the author's taste and the evidence surface.",
      "archive_signal": "Public pages improved only after the design direction was made more academic, specific, and less pitch-like.",
      "mitigation": "Add an anti-template design check and post-build audit for copy, density, typography, and evidence visibility.",
      "benchmark_cases": [
        "anti-template-design-governance",
        "artifact-critique"
      ]
    },
    {
      "id": "missing_ledger",
      "label": "Missing ledger",
      "definition": "The agent narrates progress instead of maintaining restartable state for what was found, skipped, moved, verified, or blocked.",
      "archive_signal": "Long-running connector and filesystem jobs became harder to resume when the live ledger was absent or reconstructed late.",
      "mitigation": "Require a state ledger with counts, confidence, review queue, rollback plan, and remaining work.",
      "benchmark_cases": [
        "connector-state-ledger",
        "agent-work-ledger",
        "autonomous-build-kit"
      ]
    },
    {
      "id": "runtime_mismatch",
      "label": "Runtime mismatch",
      "definition": "The agent chooses one execution surface for jobs with different supervision, write access, duration, and blast radius.",
      "archive_signal": "Read-only monitoring, scoped batch writing, broad-write automation, drafting, and parallel research behaved like different operational classes.",
      "mitigation": "Route each task by supervision level, sandbox, approval mode, memory boundary, cost, and integration risk.",
      "benchmark_cases": [
        "runtime-router",
        "production-agent-roster",
        "tool-comparison"
      ]
    },
    {
      "id": "fake_certainty",
      "label": "Fake certainty",
      "definition": "The answer hides sparse evidence, unavailable sources, or unresolved ambiguity under confident prose.",
      "archive_signal": "The most useful responses named missing sources and treated scarcity as a result instead of filling the gap.",
      "mitigation": "Reward explicit uncertainty, source gaps, and refusal to pad weak evidence.",
      "benchmark_cases": [
        "evidence-humility",
        "status-synthesis",
        "billing-reconciliation"
      ]
    }
  ]
}