src/timeline_model.py

from __future__ import annotations

from typing import Any

from .progress import progress_from_events

PHASE_ORDER = [
    "start",
    "model",
    "agent",
    "hardware",
    "deploy",
    "live_validation",
    "recovery",
    "archive",
    "done",
]

PHASE_LABELS = {
    "start": "Start",
    "model": "Model",
    "agent": "Agent",
    "hardware": "GPU",
    "deploy": "Deploy",
    "live_validation": "Live test",
    "recovery": "Recovery",
    "archive": "Archive",
    "done": "Done",
}

PHASE_STEPS = {
    "start": {"token_context", "bootstrap", "dependencies", "auth", "workspace", "node", "pi_install"},
    "model": {"model_analysis", "model_prescan"},
    "agent": {"pi_config", "pi_run", "traces", "pi_model_resolution", "pi_verification"},
    "hardware": {"hardware_strategy", "create_space_hardware", "create_space"},
    "deploy": {"metadata_sanitize", "requirements_sanitize", "upload_files", "space_logs", "space_runtime", "live_wait"},
    "live_validation": {"endpoint_discovery", "api_validation", "generation_smoke", "inference_gate"},
    "recovery": {
        "failure_detected",
        "failure_diagnosis",
        "pi_diagnosis",
        "repair_decision",
        "wait_for_logs",
        "factory_rebuild",
        "repair",
        "repair_diagnosis",
        "repair_brief",
        "repair_plan",
        "repair_patch",
        "repair_upload",
        "repair_validation",
    },
    "archive": {"report_write", "artifact_manifest", "anonymous_eval", "eval_publish", "eval_archive"},
    "done": {"done", "failure", "technical_blocker", "technical_blocker_boot_only", "manual_hardware_required"},
}

STEP_TO_PHASE = {step: phase for phase, steps in PHASE_STEPS.items() for step in steps}

SUCCESS_STATUSES = {"success", "done", "passed", "completed", "full_inference_success"}
PARTIAL_STATUSES = {"full_inference_candidate_health_passed", "health_only", "partial", "partial_validation", "completed_with_warnings"}
FAILED_STATUSES = {"failed", "failure", "error"}
RUNNING_STATUSES = {"started", "running", "waiting", "pending", "scheduled"}
CANCELLED_STATUSES = {"cancelled", "canceled"}
MANUAL_STATUSES = {"manual_hardware_required", "manual_action_required", "generated_needs_manual_hardware", "waiting_manual_hardware"}
BLOCKED_STATUSES = {"technical_blocker", "technical_blocker_boot_only", "blocked"}
AUTH_STATUSES = {"auth_refresh_required", "oauth_expired", "auth_expired", "repair_validation_inconclusive_auth", "inconclusive_auth_expired"}

TERMINAL_SOURCE_KEYS = ("status", "final_status", "effective_status", "effective_verdict", "display_status", "verdict", "result_status", "validation_status")


def _lower(value: Any) -> str:
    return str(value or "").strip().lower()


def _terminal_success(bundle: dict[str, Any], verdict: str | None = None) -> bool:
    """Return true when durable artifacts prove a terminal full-inference success.

    Historical run hydration may be requested with a truncated or missing event
    list.  A final success artifact must still replay the whole product journey
    as complete instead of leaving early UI phases pending.
    """
    normalized = _normalize_verdict_token(verdict) if verdict else ""
    if normalized == "full_inference_success":
        return True
    for src in (
        bundle.get("final_status_reconciliation") or {},
        bundle.get("summary") or {},
        bundle.get("summary_file") or {},
        bundle.get("state") or {},
        bundle.get("live_status") or {},
    ):
        if not isinstance(src, dict):
            continue
        if _normalize_verdict_token(src.get("status") or src.get("final_status") or src.get("display_status")) == "full_inference_success":
            return True
    smoke = _smoke(bundle)
    signals = _signals(bundle)
    if _lower(smoke.get("status")) == "success" or smoke.get("ok") is True or signals.get("generation_smoke_passed") is True:
        # Smoke success is treated as terminal-success proof only when a durable
        # final/summary source also says the run finished successfully.
        for src in (bundle.get("final_status_reconciliation") or {}, bundle.get("summary") or {}, bundle.get("summary_file") or {}):
            if isinstance(src, dict) and _normalize_verdict_token(src.get("status")) in {"full_inference_success", "success"}:
                return True
    return False


def _events(bundle: dict[str, Any]) -> list[dict[str, Any]]:
    return [event for event in (bundle.get("events") or []) if isinstance(event, dict)]


def _event_steps(bundle: dict[str, Any]) -> set[str]:
    return {_lower(event.get("step")) for event in _events(bundle) if event.get("step")}


def _events_for_phase(bundle: dict[str, Any], phase: str) -> list[dict[str, Any]]:
    steps = PHASE_STEPS.get(phase, set())
    return [event for event in _events(bundle) if _lower(event.get("step")) in steps]


def _latest_event_for_phase(bundle: dict[str, Any], phase: str) -> dict[str, Any] | None:
    events = _events_for_phase(bundle, phase)
    return events[-1] if events else None


def _latest_events_by_step(bundle: dict[str, Any], phase: str) -> dict[str, dict[str, Any]]:
    latest: dict[str, dict[str, Any]] = {}
    for event in _events_for_phase(bundle, phase):
        step = _lower(event.get("step"))
        if step:
            latest[step] = event
    return latest


def _has_downstream_proof(bundle: dict[str, Any], phase: str) -> bool:
    steps = _event_steps(bundle)
    phase_index = PHASE_ORDER.index(phase)
    return any(steps.intersection(PHASE_STEPS[p]) for p in PHASE_ORDER[phase_index + 1 :])


def _has_successful_phase_event(bundle: dict[str, Any], phase: str) -> bool:
    return any(_lower(e.get("status")) in SUCCESS_STATUSES for e in _events_for_phase(bundle, phase))


def _has_status(bundle: dict[str, Any], statuses: set[str]) -> bool:
    state = bundle.get("state") or {}
    summary = bundle.get("summary") or {}
    gate = bundle.get("inference_gate") or {}
    smoke = bundle.get("generation_smoke") or {}
    values = {
        _lower(state.get("status")),
        _lower(state.get("gate_status")),
        _lower(summary.get("status")),
        _lower(gate.get("status")),
        _lower(smoke.get("status")),
    }
    values.update(_lower(event.get("status")) for event in _events(bundle))
    return bool(values.intersection(statuses))


def _gate(bundle: dict[str, Any]) -> dict[str, Any]:
    return bundle.get("inference_gate") or {}


def _signals(bundle: dict[str, Any]) -> dict[str, Any]:
    gate = _gate(bundle)
    return gate.get("implementation_signals") if isinstance(gate.get("implementation_signals"), dict) else {}


def _smoke(bundle: dict[str, Any]) -> dict[str, Any]:
    smoke = bundle.get("generation_smoke") or {}
    if smoke:
        return smoke
    gate_smoke = (_gate(bundle).get("generation_smoke") or {})
    return gate_smoke if isinstance(gate_smoke, dict) else {}


def _runtime(bundle: dict[str, Any]) -> dict[str, Any]:
    runtime = bundle.get("space_runtime") or {}
    return runtime if isinstance(runtime, dict) else {}


def _live_status(bundle: dict[str, Any]) -> dict[str, Any]:
    live = bundle.get("live_status") or {}
    return live if isinstance(live, dict) else {}


def _latest_event_by_step(bundle: dict[str, Any], step_name: str) -> dict[str, Any]:
    target = _lower(step_name)
    for event in reversed(_events(bundle)):
        if _lower(event.get("step")) == target:
            return event
    return {}



def _hardware_attempt_rows(bundle: dict[str, Any]) -> list[dict[str, Any]]:
    """Return normalized hardware-at-creation attempts.

    v198.26.11 uses these only to explain the Live Test story.  They must not
    influence the run verdict.
    """
    sources = (bundle.get("hardware_attempts") or {}, bundle.get("hardware_strategy") or {}, bundle.get("state") or {})
    for source in sources:
        attempts = source.get("attempts") if isinstance(source, dict) else None
        if isinstance(attempts, list):
            return [a for a in attempts if isinstance(a, dict)]
    return []


def _attempt_hardware(attempt: dict[str, Any]) -> str:
    return str(
        attempt.get("hardware")
        or attempt.get("requested_hardware")
        or attempt.get("target_hardware")
        or attempt.get("sku")
        or attempt.get("flavor")
        or ""
    ).strip()


def _attempt_status(attempt: dict[str, Any]) -> str:
    return _lower(attempt.get("status") or attempt.get("result") or attempt.get("outcome"))


def _selected_hardware(bundle: dict[str, Any]) -> str:
    for source in (bundle.get("hardware_attempts") or {}, bundle.get("hardware_strategy") or {}, bundle.get("summary") or {}, bundle.get("state") or {}):
        if isinstance(source, dict):
            value = str(source.get("selected_hardware") or source.get("hardware") or "").strip()
            if value:
                return value
    return ""


def _zero_gpu_fallback_story(bundle: dict[str, Any]) -> tuple[bool, str, str]:
    attempts = _hardware_attempt_rows(bundle)
    selected = _selected_hardware(bundle)
    zero_failed = False
    fallback_hardware = selected
    fallback_success = False
    for attempt in attempts:
        hw = _attempt_hardware(attempt)
        status = _attempt_status(attempt)
        hw_lower = _lower(hw)
        is_zero = "zero" in hw_lower or hw_lower.startswith("zero-")
        if is_zero and (status in FAILED_STATUSES or status in {"refused", "unavailable", "denied", "quota_exceeded", "quota"} or "fail" in status):
            zero_failed = True
        if not is_zero and (status in SUCCESS_STATUSES or status in {"selected", "created", "ok"}):
            fallback_success = True
            fallback_hardware = hw or selected
    if zero_failed and (fallback_success or (selected and "zero" not in _lower(selected))):
        return True, fallback_hardware or selected or "fallback hardware", "ZeroGPU unavailable → using " + (fallback_hardware or selected or "fallback hardware")
    return False, fallback_hardware or selected, ""


def _latest_step_status(bundle: dict[str, Any], steps: set[str]) -> str:
    for event in reversed(_events(bundle)):
        if _lower(event.get("step")) in steps:
            return _lower(event.get("status"))
    return ""


def _latest_step_message(bundle: dict[str, Any], steps: set[str]) -> str:
    for event in reversed(_events(bundle)):
        if _lower(event.get("step")) in steps:
            return str(event.get("message") or "").strip()
    return ""


BUILD_ERROR_STAGE_TOKENS = {"BUILD_ERROR", "BUILD_FAILED", "CONFIG_ERROR", "NO_APP_FILE"}
RUNTIME_ERROR_STAGE_TOKENS = {"RUNTIME_ERROR", "APP_START_ERROR", "STARTUP_ERROR"}
BUILDING_STAGE_TOKENS = {"BUILDING", "BUILD_QUEUED", "QUEUED", "PENDING_BUILD"}
STARTING_STAGE_TOKENS = {"STARTING", "RUNTIME_STARTING", "APP_STARTING", "LOADING", "RESTARTING"}
RUNNING_STAGE_TOKENS = {"RUNNING", "RUNNING_BUILDING", "RUNNING_APP_STARTING"}


def _runtime_stage_value(runtime: dict[str, Any]) -> str:
    return str(runtime.get("stage") or runtime.get("runtime_stage") or runtime.get("status") or runtime.get("phase") or "").strip()


def _runtime_stage_upper(runtime: dict[str, Any]) -> str:
    return _runtime_stage_value(runtime).upper().replace("-", "_").replace(" ", "_")


def _stage_has(upper_stage: str, tokens: set[str]) -> bool:
    if not upper_stage:
        return False
    return any(token in upper_stage for token in tokens)


def _runtime_error_message(runtime: dict[str, Any]) -> str:
    return str(runtime.get("error") or runtime.get("message") or runtime.get("last_error") or "").strip()


def _runtime_stage_facts(bundle: dict[str, Any], *, live: dict[str, Any], runtime: dict[str, Any], smoke: dict[str, Any], gate: dict[str, Any]) -> dict[str, Any]:
    signals = _signals(bundle)
    runtime_upload_epoch = bundle.get("runtime_upload_epoch") or {}
    upload_status = _latest_step_status(bundle, {"upload_files"})
    runtime_uploaded = bool(
        runtime_upload_epoch.get("last_upload_completed_at")
        or runtime_upload_epoch.get("upload_sequence")
        or upload_status in SUCCESS_STATUSES
    )
    stage_value = _runtime_stage_value(runtime)
    upper_stage = _runtime_stage_upper(runtime)
    smoke_status = _lower(smoke.get("status") or _latest_step_status(bundle, {"generation_smoke"}))
    api_status = _latest_step_status(bundle, {"endpoint_discovery", "api_validation"})
    endpoints = gate.get("endpoints") or gate.get("api_endpoints") or gate.get("named_endpoints") or []
    selected_endpoint = str(smoke.get("api_name") or gate.get("api_name") or gate.get("selected_endpoint") or gate.get("selected_api_name") or "").strip()
    endpoint_known = bool(selected_endpoint or (isinstance(endpoints, list) and endpoints) or api_status in SUCCESS_STATUSES or smoke_status == "success" or signals.get("generation_smoke_passed") is True)
    health_passed = signals.get("health_passed") is True or api_status in SUCCESS_STATUSES
    health_failed = signals.get("health_passed") is False or api_status in FAILED_STATUSES
    smoke_passed = smoke_status == "success" or smoke.get("ok") is True or signals.get("generation_smoke_passed") is True
    smoke_failed = smoke_status in FAILED_STATUSES or smoke_status in {"timeout", "exception"}
    build_error = _stage_has(upper_stage, BUILD_ERROR_STAGE_TOKENS)
    runtime_error = _stage_has(upper_stage, RUNTIME_ERROR_STAGE_TOKENS)
    building = _stage_has(upper_stage, BUILDING_STAGE_TOKENS) or ("BUILD" in upper_stage and not build_error)
    starting = _stage_has(upper_stage, STARTING_STAGE_TOKENS)
    running = _stage_has(upper_stage, RUNNING_STAGE_TOKENS) or "RUNNING" in upper_stage or endpoint_known or health_passed or smoke_passed
    return {
        "runtime_uploaded": runtime_uploaded,
        "stage_value": stage_value,
        "upper_stage": upper_stage,
        "build_error": build_error,
        "runtime_error": runtime_error,
        "building": building,
        "starting": starting,
        "running": running,
        "endpoint_known": endpoint_known,
        "selected_endpoint": selected_endpoint,
        "api_status": api_status,
        "health_passed": health_passed,
        "health_failed": health_failed,
        "smoke_status": smoke_status,
        "smoke_passed": smoke_passed,
        "smoke_failed": smoke_failed,
        "error_message": _runtime_error_message(runtime),
    }


def build_runtime_stage_history(bundle: dict[str, Any], *, live: dict[str, Any] | None = None, runtime: dict[str, Any] | None = None, smoke: dict[str, Any] | None = None, gate: dict[str, Any] | None = None) -> dict[str, Any]:
    """Granular Space runtime state for the Live Test panel.

    v198.26.13 keeps verdict logic untouched and only translates durable Space
    facts into a user-facing state machine: build queued/building/build error,
    runtime starting/running/runtime error, API ready, and generation verified.
    """
    live = live if isinstance(live, dict) else _live_status(bundle)
    runtime = runtime if isinstance(runtime, dict) else _runtime(bundle)
    smoke = smoke if isinstance(smoke, dict) else _smoke(bundle)
    gate = gate if isinstance(gate, dict) else _gate(bundle)
    facts = _runtime_stage_facts(bundle, live=live, runtime=runtime, smoke=smoke, gate=gate)
    stages: list[dict[str, Any]] = []

    def add(stage: str, label: str, status: str, message: str, *, diagnostic: bool = False) -> None:
        stages.append({
            "stage": stage,
            "id": stage,
            "label": label,
            "status": status,
            "message": message,
            "detail": message,
            "diagnostic": diagnostic,
        })

    if facts["build_error"]:
        add("space_build", "Space build", "failed", facts["error_message"] or "Build error detected on Hugging Face", diagnostic=True)
        add("runtime_startup", "Runtime startup", "blocked", "Runtime startup cannot continue until the build error is diagnosed.", diagnostic=True)
        current_stage = "build_error"
        current_message = "Build error detected. Switching to diagnostic mode."
        status = "failed"
    elif facts["runtime_error"]:
        add("space_build", "Space build", "success", "Space build completed")
        add("runtime_startup", "Runtime startup", "failed", facts["error_message"] or "Runtime error detected while starting the app", diagnostic=True)
        current_stage = "runtime_error"
        current_message = "Runtime error detected. Switching to diagnostic mode."
        status = "failed"
    elif facts["smoke_passed"]:
        add("space_build", "Space build", "success", "Space build completed")
        add("runtime_startup", "Runtime startup", "success", "Space is running")
        add("api_readiness", "Gradio API", "ready", f"API ready{': ' + facts['selected_endpoint'] if facts['selected_endpoint'] else ''}")
        add("generation_smoke", "Generation smoke", "verified", "Generation verified")
        current_stage = "generation_verified"
        current_message = "Generation verified."
        status = "success"
    elif facts["smoke_failed"]:
        add("space_build", "Space build", "success", "Space build completed" if facts["running"] else "Build stage passed")
        add("runtime_startup", "Runtime startup", "success" if facts["running"] else "running", "Space is running" if facts["running"] else "Runtime is starting")
        if facts["endpoint_known"]:
            add("api_readiness", "Gradio API", "ready", f"API ready{': ' + facts['selected_endpoint'] if facts['selected_endpoint'] else ''}")
        else:
            add("api_readiness", "Gradio API", "running", "Waiting for Gradio API")
        add("generation_smoke", "Generation smoke", "failed", smoke.get("error") or smoke.get("message") or "Generation smoke failed", diagnostic=True)
        current_stage = "generation_smoke_failed"
        current_message = "Runtime is reachable, but generation was not verified."
        status = "failed"
    elif facts["endpoint_known"] or facts["api_status"] in SUCCESS_STATUSES:
        add("space_build", "Space build", "success", "Space build completed")
        add("runtime_startup", "Runtime startup", "success", "Space is running")
        add("api_readiness", "Gradio API", "ready", f"API ready{': ' + facts['selected_endpoint'] if facts['selected_endpoint'] else ''}")
        add("generation_smoke", "Generation smoke", "pending", "Waiting for generation smoke")
        current_stage = "api_ready"
        current_message = "Gradio API is ready. Waiting for generation smoke."
        status = "in_progress"
    elif facts["health_failed"]:
        add("space_build", "Space build", "success" if facts["running"] else "running", "Space build completed" if facts["running"] else "Waiting for build completion")
        add("runtime_startup", "Runtime startup", "failed", "Runtime became reachable but health/API did not pass", diagnostic=True)
        current_stage = "runtime_error"
        current_message = "Runtime health check failed. Switching to diagnostic mode if it does not recover."
        status = "failed"
    elif facts["running"]:
        add("space_build", "Space build", "success", "Space build completed")
        add("runtime_startup", "Runtime startup", "running", "Space is running; waiting for Gradio API")
        add("api_readiness", "Gradio API", "running", "Waiting for Gradio API and health check")
        current_stage = "runtime_starting"
        current_message = "Space is running. Waiting for Gradio API."
        status = "in_progress"
    elif facts["building"]:
        add("space_build", "Space build", "building", "Space is building on Hugging Face")
        add("runtime_startup", "Runtime startup", "pending", "Runtime startup begins after build completes")
        current_stage = "building"
        current_message = "Space is still building on Hugging Face."
        status = "in_progress"
    elif facts["runtime_uploaded"]:
        add("space_build", "Space build", "queued", "Runtime uploaded; waiting for Hugging Face build to start")
        add("runtime_startup", "Runtime startup", "pending", "Waiting for Space build")
        current_stage = "build_queued"
        current_message = "Runtime uploaded. Waiting for Hugging Face build."
        status = "in_progress"
    else:
        add("space_build", "Space build", "pending", "Waiting for runtime upload")
        add("runtime_startup", "Runtime startup", "pending", "Waiting for Space build")
        current_stage = "waiting_for_upload"
        current_message = "Waiting for runtime upload before build can start."
        status = "pending"

    diagnostic_mode = any(bool(item.get("diagnostic")) for item in stages)
    return {
        "schema_version": "runtime_stage_history.v198_26_13",
        "status": status,
        "current_stage": current_stage,
        "current_message": current_message,
        "terminal": status in {"success", "failed"},
        "diagnostic_mode": diagnostic_mode,
        "runtime_stage": facts["stage_value"],
        "stages": stages,
    }


def build_live_test_story(bundle: dict[str, Any], *, live: dict[str, Any] | None = None, runtime: dict[str, Any] | None = None, smoke: dict[str, Any] | None = None, gate: dict[str, Any] | None = None) -> dict[str, Any]:
    """Curated Live Test story for Active Run.

    This is deliberately not a raw event dump. It compresses retry/fallback
    mechanics and now includes a granular runtime stage machine so users know
    whether they are waiting for HF build, runtime startup, API readiness, or
    a diagnostic-worthy error.
    """
    live = live if isinstance(live, dict) else _live_status(bundle)
    runtime = runtime if isinstance(runtime, dict) else _runtime(bundle)
    smoke = smoke if isinstance(smoke, dict) else _smoke(bundle)
    gate = gate if isinstance(gate, dict) else _gate(bundle)
    summary = bundle.get("summary") or {}
    state = bundle.get("state") or {}
    steps: list[dict[str, Any]] = []

    def add(step_id: str, label: str, status: str, detail: Any = "", ts: Any = "") -> None:
        steps.append({
            "id": step_id,
            "stage": step_id,
            "label": label,
            "status": status or "pending",
            "detail": str(detail or "")[:500],
            "ts": ts or "",
        })

    fallback_used, fallback_hw, fallback_detail = _zero_gpu_fallback_story(bundle)
    selected_hw = fallback_hw or _selected_hardware(bundle) or str(summary.get("selected_hardware") or state.get("selected_hardware") or "").strip()
    create_status = _latest_step_status(bundle, {"create_space", "create_space_hardware"})
    upload_status = _latest_step_status(bundle, {"upload_files"})
    create_message = _latest_step_message(bundle, {"create_space", "create_space_hardware"})
    runtime_upload_epoch = bundle.get("runtime_upload_epoch") or {}
    runtime_uploaded = bool(
        runtime_upload_epoch.get("last_upload_completed_at")
        or runtime_upload_epoch.get("upload_sequence")
        or upload_status in SUCCESS_STATUSES
    )

    zero_attempt_failed_without_fallback = (not fallback_used) and any(
        ("zero" in _lower(_attempt_hardware(attempt))) and _attempt_status(attempt) in FAILED_STATUSES
        for attempt in _hardware_attempt_rows(bundle)
    )
    if fallback_used:
        add("hardware", "Hardware provisioning", "fallback", fallback_detail)
    elif zero_attempt_failed_without_fallback or create_status in FAILED_STATUSES:
        add("hardware", "Hardware provisioning", "failed", create_message or "Hardware provisioning failed")
    elif selected_hw:
        add("hardware", "Hardware provisioning", "selected", f"{selected_hw} selected")
    else:
        add("hardware", "Hardware provisioning", "running" if create_status in RUNNING_STATUSES else "pending", "Selecting hardware")

    if create_status in FAILED_STATUSES and not fallback_used and not runtime_uploaded:
        add("space", "Space setup", "failed", create_message or "Space creation failed")
    elif create_status in SUCCESS_STATUSES or runtime_uploaded or summary.get("target_space") or state.get("target_space"):
        suffix = f" on {selected_hw}" if selected_hw else ""
        add("space", "Space setup", "success", f"Space created{suffix}")
    elif create_status in RUNNING_STATUSES:
        add("space", "Space setup", "running", "Creating Space")
    else:
        add("space", "Space setup", "pending", "Waiting for Space creation")

    if runtime_uploaded:
        add("runtime_upload", "Runtime deployment", "uploaded", "Runtime uploaded")
    elif upload_status in RUNNING_STATUSES:
        add("runtime_upload", "Runtime deployment", "running", "Uploading runtime")
    elif upload_status in FAILED_STATUSES:
        add("runtime_upload", "Runtime deployment", "failed", _latest_step_message(bundle, {"upload_files"}) or "Runtime upload failed")
    else:
        add("runtime_upload", "Runtime deployment", "pending", "Waiting for runtime upload")

    runtime_stage_history = build_runtime_stage_history(bundle, live=live, runtime=runtime, smoke=smoke, gate=gate)
    for row in runtime_stage_history.get("stages") or []:
        if not isinstance(row, dict):
            continue
        add(str(row.get("id") or row.get("stage") or "runtime_stage"), str(row.get("label") or "Runtime stage"), str(row.get("status") or "pending"), row.get("message") or row.get("detail") or "")

    story_status = runtime_stage_history.get("status") or "in_progress"
    if story_status == "success":
        # v198.26.25: a terminal generation smoke success should hydrate the
        # whole Live Test story as complete on historical reload, even when the
        # cached run selection shell lacked early create/upload events.
        for row in steps:
            row_id = str(row.get("id") or "")
            row_status = _lower(row.get("status"))
            if row_id == "hardware" and row_status in {"", "pending", "running", "selected"}:
                row["status"] = "selected"
                if not str(row.get("detail") or "").strip() or "selecting" in str(row.get("detail") or "").lower():
                    row["detail"] = f"{selected_hw} selected" if selected_hw else "Hardware selected"
            elif row_id == "space" and row_status in {"", "pending", "running"}:
                row["status"] = "success"
                if not str(row.get("detail") or "").strip() or "waiting" in str(row.get("detail") or "").lower():
                    row["detail"] = "Space created"
            elif row_id == "runtime_upload" and row_status in {"", "pending", "running"}:
                row["status"] = "uploaded"
                if not str(row.get("detail") or "").strip() or "waiting" in str(row.get("detail") or "").lower():
                    row["detail"] = "Runtime uploaded"

    headline = (
        "Generation verified" if story_status == "success"
        else "Live test needs attention" if story_status == "failed"
        else runtime_stage_history.get("current_message") or "Waiting for Space runtime"
    )
    return {
        "schema_version": "live_test_story.v198_26_13",
        "status": story_status,
        "headline": headline,
        "steps": steps,
        "fallback_used": fallback_used,
        "selected_hardware": selected_hw,
        "runtime_stage_history": runtime_stage_history,
    }


def _runtime_history(bundle: dict[str, Any], *, live: dict[str, Any], runtime: dict[str, Any], smoke: dict[str, Any], gate: dict[str, Any]) -> list[dict[str, Any]]:
    """Return the curated Live Test story steps for timeline details.

    v198.26.11 replaces duplicate raw create/upload events with a process story:
    hardware fallback, Space setup, runtime deployment, HF build/runtime, API, smoke.
    """
    return build_live_test_story(bundle, live=live, runtime=runtime, smoke=smoke, gate=gate).get("steps") or []

def build_live_validation_model(bundle: dict[str, Any]) -> dict[str, Any]:
    """Compact live validation telemetry for the UI.

    The timeline phase remains the coarse product step. This model gives the
    Live Validation card concrete runtime facts while the Space is building,
    starting, exposing Gradio endpoints, or running generation smoke tests.
    """
    runtime = _runtime(bundle)
    smoke = _smoke(bundle)
    gate = _gate(bundle)
    signals = _signals(bundle)
    api_event = _latest_event_by_step(bundle, "api_validation")
    smoke_event = _latest_event_by_step(bundle, "generation_smoke")
    live_status = _live_status(bundle)
    live_stage = str(live_status.get("stage") or "").strip()
    live_message = str(live_status.get("message") or "").strip()
    runtime_stage = str(runtime.get("stage") or runtime.get("status") or "").strip()
    runtime_error = str(runtime.get("error") or "").strip()
    runtime_stage_history = build_runtime_stage_history(bundle, live=live_status, runtime=runtime, smoke=smoke, gate=gate)
    health_passed = signals.get("health_passed") is True or _lower(api_event.get("status")) in SUCCESS_STATUSES
    health_failed = signals.get("health_passed") is False or _lower(api_event.get("status")) in FAILED_STATUSES
    smoke_status = _lower(smoke.get("status") or smoke_event.get("status"))
    endpoints = gate.get("endpoints") or gate.get("api_endpoints") or gate.get("named_endpoints") or []
    endpoint_count = len(endpoints) if isinstance(endpoints, list) else 0
    selected_endpoint = str(
        smoke.get("api_name")
        or gate.get("api_name")
        or gate.get("selected_endpoint")
        or gate.get("selected_api_name")
        or ""
    ).strip()
    endpoint_known = bool(endpoint_count or selected_endpoint or smoke_status == "success" or signals.get("generation_smoke_passed") is True)

    severity = "info"
    stage = "waiting_for_space"
    label = "Waiting for Space runtime"
    next_action = "Waiting for the generated Space to build and start."

    upper_stage = runtime_stage.upper()
    granular_stage = str(runtime_stage_history.get("current_stage") or "")
    if granular_stage == "build_error":
        severity = "error"
        stage = "space_build_error"
        label = "Space build error"
        next_action = "Build error detected. Switching to diagnostic mode."
    elif granular_stage == "runtime_error":
        severity = "error"
        stage = "space_runtime_error"
        label = "Space runtime error"
        next_action = "Runtime error detected. Switching to diagnostic mode."
    elif smoke_status == "success" or signals.get("generation_smoke_passed") is True:
        severity = "success"
        stage = "generation_verified"
        label = "Generation verified"
        next_action = "The generated Space passed the automatic generation smoke test."
    elif smoke and smoke_status in FAILED_STATUSES.union({"timeout", "exception"}):
        severity = "warning"
        stage = "generation_smoke_failed"
        label = "Generation not verified"
        if smoke.get("auto_retry", {}).get("retried"):
            next_action = "The factory retried once with schema-corrected values. Use Prefill Space Test to retry manually if needed."
        elif smoke.get("retryable_with_schema_payload") or smoke.get("auto_retry_supported"):
            next_action = "Use Prefill Space Test after the run finishes to retry with schema-adjusted args."
        else:
            next_action = "Use Prefill Space Test after the run finishes to retry with adjusted args."
    elif live_stage == "generation_smoke_retry":
        severity = "info"
        stage = "generation_smoke_retry"
        label = "Retrying generation smoke"
        next_action = "Retrying once with schema-corrected Gradio choice values."
    elif live_stage == "generation_smoke":
        severity = "info"
        stage = "generation_smoke_running"
        label = "Generation smoke running"
        next_action = "Waiting for the generated endpoint response."
    elif live_stage == "endpoint_discovery":
        severity = "info"
        stage = "endpoint_discovery"
        label = "Discovering Gradio endpoints"
        next_action = "Inspecting the Space API schema before validation."
    elif _latest_event_by_step(bundle, "endpoint_discovery"):
        event = _latest_event_by_step(bundle, "endpoint_discovery")
        severity = "success" if _lower(event.get("status")) in SUCCESS_STATUSES else "info"
        stage = "endpoint_discovery"
        label = "Gradio endpoint discovery"
        next_action = str(event.get("message") or "Endpoint discovery completed.")
    elif health_passed:
        severity = "info"
        stage = "health_passed"
        label = "Health passed; waiting for generation smoke"
        next_action = "Waiting for Gradio endpoints and automatic generation validation."
    elif health_failed:
        severity = "warning"
        stage = "health_not_confirmed"
        label = "Health not confirmed"
        next_action = "Waiting for the Space API to become reachable or fail conclusively."
    elif granular_stage == "building":
        stage = "space_building"
        label = "Space is building"
        next_action = "Hugging Face is still building the Space runtime."
    elif granular_stage == "build_queued":
        stage = "space_build_queued"
        label = "Space build queued"
        next_action = "Runtime files are uploaded; waiting for Hugging Face build to start."
    elif granular_stage == "runtime_starting":
        stage = "space_starting"
        label = "Runtime is starting"
        next_action = "Space is running; waiting for Gradio API and health check."
    elif granular_stage == "api_ready":
        stage = "api_ready"
        label = "Gradio API ready"
        next_action = "Waiting for the automatic generation smoke test."
    elif endpoint_count:
        stage = "gradio_schema_detected"
        label = "Gradio endpoints detected"
        next_action = "Waiting for automatic generation smoke validation."

    return {
        "stage": stage,
        "label": label,
        "severity": severity,
        "runtime_stage": runtime_stage,
        "runtime_error": runtime_error,
        "health": "passed" if health_passed else "failed" if health_failed else "pending",
        "gradio_schema": "detected" if endpoint_known else "pending",
        "endpoint_count": endpoint_count,
        "selected_endpoint": selected_endpoint,
        "generation_smoke": "passed" if smoke_status == "success" else "failed" if smoke and smoke_status in FAILED_STATUSES.union({"timeout", "exception"}) else "pending",
        "live_status_stage": live_stage,
        "live_status_updated_at": live_status.get("updated_at") or "",
        "message": smoke.get("error") or smoke.get("message") or live_message or api_event.get("message") or runtime_error or "",
        "next_action": next_action,
        "runtime_history": _runtime_history(bundle, live=live_status, runtime=runtime, smoke=smoke, gate=gate),
        "runtime_stage_history": runtime_stage_history,
        "live_test_story": build_live_test_story(bundle, live=live_status, runtime=runtime, smoke=smoke, gate=gate),
    }


def _eval_publish(bundle: dict[str, Any]) -> dict[str, Any]:
    status = bundle.get("eval_publish_status") or bundle.get("eval_publish") or {}
    return status if isinstance(status, dict) else {}


def _build_error_observation(bundle: dict[str, Any]) -> dict[str, Any]:
    observation = bundle.get("build_error_observation") or {}
    return observation if isinstance(observation, dict) else {}


def _final_blocker_label(bundle: dict[str, Any]) -> str:
    observation = _build_error_observation(bundle)
    reason = _lower(observation.get("reason"))
    tail = str(observation.get("tail") or observation.get("first_error") or "")
    if reason or "pyenv install 3.1" in tail or "BUILD FAILED" in tail:
        if "pyenv install 3.1" in tail or "python3.1" in tail.replace(" ", ""):
            return "Build failed: invalid Python version requested pyenv install 3.1"
        return "Build failed after repair"
    blockage = bundle.get("blockage") or {}
    if isinstance(blockage, dict):
        status = str(blockage.get("status") or "").replace("_", " ").strip()
        decision = blockage.get("decision") if isinstance(blockage.get("decision"), dict) else {}
        reason_text = str(decision.get("reason") or blockage.get("reason") or "").strip()
        if reason_text:
            return reason_text[:220]
        if status:
            return status.title()
    return ""


def _is_record_not_ready(status: dict[str, Any]) -> bool:
    return _lower(status.get("reason")) == "record_not_ready"


def _has_terminal_event(bundle: dict[str, Any]) -> bool:
    return any(_lower(event.get("step")) in {"done", "failure", "technical_blocker", "technical_blocker_boot_only", "manual_hardware_required"} for event in _events(bundle))


def _latest_recovery_event(bundle: dict[str, Any]) -> dict[str, Any] | None:
    events = _events_for_phase(bundle, "recovery")
    return events[-1] if events else None


def _phase_status_from_events(bundle: dict[str, Any], phase: str) -> str:
    latest_by_step = _latest_events_by_step(bundle, phase)
    if not latest_by_step:
        return "pending"
    statuses = {_lower(event.get("status")) for event in latest_by_step.values()}
    if statuses.intersection(FAILED_STATUSES):
        return "failed"
    if statuses.intersection({"warning"}) or statuses.intersection(PARTIAL_STATUSES):
        return "warning"
    if statuses.intersection(RUNNING_STATUSES):
        # While a phase is actively running, the running signal must win over
        # earlier successful setup events in the same phase. Example: pi_config
        # can be success while pi_run is started/running; the Agent phase should
        # remain active and pulsing until a downstream proof or terminal event
        # later marks it complete.
        return "running"
    if statuses.intersection(SUCCESS_STATUSES):
        return "complete"
    return "complete"


def _normalize_verdict_token(value: Any) -> str:
    normalized = _lower(value)
    if not normalized:
        return ""
    if normalized in AUTH_STATUSES:
        return "auth_refresh_required"
    if normalized in {"succeeded", "success", "done", "completed", "passed", "repair_success"}:
        return "full_inference_success"
    if normalized in SUCCESS_STATUSES:
        return "full_inference_success" if normalized == "full_inference_success" else "success"
    if normalized in PARTIAL_STATUSES or normalized in {"partial", "health_only", "full_inference_candidate_health_passed", "completed_with_warnings"}:
        return "partial_validation"
    if normalized in MANUAL_STATUSES:
        return "manual_action_required"
    if normalized in BLOCKED_STATUSES:
        return "technical_blocker_boot_only" if normalized == "technical_blocker_boot_only" else "technical_blocker"
    if normalized in FAILED_STATUSES:
        return "failed"
    if normalized in CANCELLED_STATUSES or normalized == "stopped":
        return "cancelled"
    if normalized == "stale":
        return "stale"
    if normalized in {"running", "pending", "queued", "started", "waiting", "scheduled"}:
        return "running"
    return normalized


def _first_terminal_verdict_from_sources(bundle: dict[str, Any]) -> str:
    """Return authoritative terminal verdict for canonical timeline convergence.

    v198.26.7: final reconciliation and summary must win over stale state,
    live_status, or repair artifacts.  This keeps Active Run and Timeline aligned
    with Run Explorer once the backend has written the final success/partial/fail.
    """
    sources: list[dict[str, Any]] = []
    for src in (
        bundle.get("final_status_reconciliation") or {},
        bundle.get("summary") or {},
        bundle.get("summary_file") or {},
        bundle.get("effective_run_status") or {},
        (bundle.get("summary") or {}).get("effective_run_status") if isinstance((bundle.get("summary") or {}).get("effective_run_status"), dict) else {},
        bundle.get("eval_record") or {},
        bundle.get("state") or {},
        bundle.get("live_status") or {},
    ):
        if isinstance(src, dict):
            sources.append(src)
    for source in sources:
        for key in TERMINAL_SOURCE_KEYS:
            verdict = _normalize_verdict_token(source.get(key))
            if verdict and verdict != "running":
                return verdict
        if source.get("ok") is True or source.get("generation_smoke_passed") is True:
            return "full_inference_success"
    return ""


def _verdict(bundle: dict[str, Any]) -> str:
    terminal = _first_terminal_verdict_from_sources(bundle)
    if terminal:
        return terminal
    gate = _gate(bundle)
    state = bundle.get("state") or {}
    live = bundle.get("live_status") or {}
    repair = bundle.get("repair_outcome") or {}
    for value in (state.get("status"), live.get("status") if _lower(live.get("stage")) == "done" else "", gate.get("status"), state.get("gate_status"), repair.get("post_repair_validation"), repair.get("failure_type")):
        normalized = _normalize_verdict_token(value)
        if normalized:
            return normalized
    return "running" if _events(bundle) else "unknown"


def _final_visual_status(verdict: str) -> str:
    if verdict in {"full_inference_success", "success", "passed"}:
        return "success"
    if verdict in {"partial_validation", "partial", "health_only", "completed_with_warnings"}:
        return "warn"
    if verdict in {"failed", "failure", "technical_blocker", "technical_blocker_boot_only"}:
        return "error"
    if verdict == "auth_refresh_required":
        return "warn"
    if verdict in {"manual_action_required", "cancelled"}:
        return "stopped"
    return "running"


def _status_label(verdict: str) -> tuple[str, str]:
    if verdict == "full_inference_success":
        return "Full inference success", "Space boots, generation passed, and latency was measured."
    if verdict == "partial_validation":
        return "Completed with partial validation", "The run finished, but full generation was not verified."
    if verdict == "technical_blocker_boot_only":
        return "Technical blocker", "Health may pass, but no generation endpoint exists for full inference."
    if verdict == "technical_blocker":
        return "Technical blocker", "The run found a technical blocker."
    if verdict == "auth_refresh_required":
        return "Auth refresh required", "HF OAuth expired or is too close to expiry; sign in again before retrying validation."
    if verdict == "manual_action_required":
        return "Manual action required", "The run needs user action before validation can continue."
    if verdict == "failed":
        return "Failed", "The run did not complete successfully."
    if verdict == "cancelled":
        return "Cancelled", "The run was cancelled."
    if verdict in {"running", "pending", "unknown"}:
        return "Running", ""
    return verdict.replace("_", " ").title(), ""


def _hardware_warning(bundle: dict[str, Any]) -> dict[str, Any] | None:
    events = _events_for_phase(bundle, "hardware")
    preferred_failed = any(_lower(e.get("step")) == "create_space_hardware" and _lower(e.get("status")) in FAILED_STATUSES for e in events)
    fallback_ok = any(_lower(e.get("step")) in {"create_space", "create_space_hardware"} and _lower(e.get("status")) in SUCCESS_STATUSES and _lower((e.get("data") or {}).get("hardware")) for e in events)
    if preferred_failed and fallback_ok:
        return {"code": "hardware_fallback_used", "label": "Fallback GPU used", "detail": "Preferred hardware was unavailable; the factory continued on fallback hardware."}
    return None


def _pi_model_warning(bundle: dict[str, Any]) -> dict[str, Any] | None:
    for event in _events(bundle):
        if _lower(event.get("step")) == "pi_model_resolution" and _lower(event.get("status")) == "warning":
            data = event.get("data") or {}
            requested = data.get("requested_model") or data.get("configured_model") or "requested model"
            effective = data.get("effective_model") or "observed model"
            return {"code": "pi_model_changed", "label": "Pi assistant model changed", "detail": f"{requested} → {effective}"}
    return None


def _logs_warning(bundle: dict[str, Any]) -> dict[str, Any] | None:
    logs = bundle.get("space_logs_index") or bundle.get("space_logs") or {}
    if not isinstance(logs, dict):
        return None
    quality = _lower(logs.get("log_quality") or logs.get("quality"))
    if quality and quality not in {"full", "complete"}:
        return {"code": "logs_partial", "label": "Space logs partial", "detail": f"Log quality: {quality}."}
    return None


def _archive_warning(bundle: dict[str, Any], *, terminal: bool) -> dict[str, Any] | None:
    status = _eval_publish(bundle)
    if not status or status.get("published") is True:
        return None
    # record_not_ready is expected during the terminal write/reconcile window.
    # It is a neutral finalization state, not an action required for users.
    if _is_record_not_ready(status):
        return None
    if status.get("attempted"):
        return {"code": "eval_archive_not_published", "label": "Eval archive not published", "detail": str(status.get("reason") or "Backend archive copy did not complete.")}
    return None


def _build_phase_details(bundle: dict[str, Any], phase: str) -> list[dict[str, Any]]:
    details: list[dict[str, Any]] = []
    if phase == "deploy":
        blocker = _final_blocker_label(bundle)
        if blocker and _build_error_observation(bundle):
            details.append({"label": blocker, "status": "failed"})
        else:
            latest = _latest_event_for_phase(bundle, phase)
            if latest:
                details.append({"label": str(latest.get("message") or latest.get("step") or PHASE_LABELS[phase]), "status": _lower(latest.get("status")) or "info"})
    elif phase == "live_validation":
        live = build_live_validation_model(bundle)
        runtime_stage = live.get("runtime_stage") or ""
        runtime_history = live.get("runtime_history") if isinstance(live.get("runtime_history"), list) else []
        for item in runtime_history:
            if isinstance(item, dict) and item.get("label"):
                details.append({"label": str(item.get("label")), "status": str(item.get("status") or "info"), "detail": str(item.get("detail") or "")})
        if runtime_stage and not runtime_history:
            details.append({"label": f"Space runtime: {runtime_stage}", "status": "info"})
        if live.get("health") == "passed":
            details.append({"label": "Health passed", "status": "complete"})
        elif live.get("health") == "failed":
            details.append({"label": "Health not confirmed", "status": "warning"})
        smoke = _smoke(bundle)
        generation_passed = live.get("generation_smoke") == "passed"
        if live.get("gradio_schema") == "detected":
            count = live.get("endpoint_count") or 0
            selected = str(live.get("selected_endpoint") or smoke.get("api_name") or "").strip()
            if selected:
                details.append({"label": f"Endpoint selected: {selected}", "status": "complete"})
            elif count:
                details.append({"label": f"Gradio endpoints detected: {count}", "status": "complete" if generation_passed else "info"})
        elif not generation_passed:
            details.append({"label": "Waiting for Gradio endpoints", "status": "info"})
        if generation_passed:
            pass
        elif live.get("generation_smoke") == "failed":
            reason = smoke.get("failure_type") or smoke.get("error") or live.get("message") or "Generation was not verified"
            details.append({"label": f"Generation not verified: {reason}", "status": "warning"})
            auto_retry = smoke.get("auto_retry") if isinstance(smoke.get("auto_retry"), dict) else {}
            if auto_retry.get("retried"):
                details.append({"label": "Smoke test retried once with schema-corrected choices", "status": "info"})
            elif smoke.get("retryable_with_schema_payload") or smoke.get("auto_retry_supported"):
                details.append({"label": "Schema-adjusted retry available in Space Test", "status": "info"})
        else:
            details.append({"label": "Generation smoke pending", "status": "info"})
        blocker = _final_blocker_label(bundle)
        if blocker:
            details.append({"label": blocker, "status": "failed"})
        recommendation = smoke.get("recommended_zero_gpu_duration_seconds") or (_gate(bundle).get("zero_gpu_duration_recommendation") or {}).get("recommended_zero_gpu_duration_seconds")
        if recommendation:
            source_hw = str(_smoke(bundle).get("hardware") or _gate(bundle).get("hardware") or (bundle.get("summary") or {}).get("selected_hardware") or "").strip()
            suffix = f" · measured on {source_hw}" if source_hw else ""
            details.append({"label": f"ZeroGPU estimate: {recommendation}s{suffix}", "status": "complete"})
    elif phase == "archive":
        eval_status = _eval_publish(bundle)
        if eval_status:
            if eval_status.get("published") is True:
                details.append({"label": "Eval archive published", "status": "complete"})
                if eval_status.get("archive_relative_path"):
                    details.append({"label": str(eval_status.get("archive_relative_path")), "status": "info"})
            elif eval_status.get("attempted"):
                if _is_record_not_ready(eval_status):
                    details.append({"label": "Eval archive finalizing", "status": "info"})
                else:
                    details.append({"label": f"Eval archive not published: {eval_status.get('reason') or 'unknown'}", "status": "warning"})
            else:
                details.append({"label": "Eval archive pending", "status": "pending"})
    elif phase == "recovery":
        steps = _event_steps(bundle)
        if not steps.intersection(PHASE_STEPS["recovery"]):
            details.append({"label": "Recovery not needed", "status": "skipped"})
        else:
            failed = any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in FAILED_STATUSES for e in _events(bundle))
            if failed:
                details.append({"label": "Patch attempted; final blocker remains", "status": "failed"})
                blocker = _final_blocker_label(bundle)
                if blocker:
                    details.append({"label": blocker, "status": "failed"})
            elif any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in SUCCESS_STATUSES for e in _events(bundle)):
                details.append({"label": "Repair revalidated", "status": "complete"})
            else:
                details.append({"label": "Recovery in progress", "status": "running"})
    else:
        latest = _latest_event_for_phase(bundle, phase)
        if latest:
            details.append({"label": str(latest.get("message") or latest.get("step") or PHASE_LABELS[phase]), "status": _lower(latest.get("status")) or "info"})
    return details


def _phase_summary(bundle: dict[str, Any], phase: str, status: str) -> str:
    if phase == "agent":
        # A Pi model mismatch is an audit note rendered in Run notes, not a
        # phase-level issue. Keep the Agent phase focused on Pi execution and
        # trace collection so successful terminal runs do not show a misleading
        # NEEDS ATTENTION state.
        pass
    if phase == "hardware":
        warning = _hardware_warning(bundle)
        if warning:
            return warning["label"]
    if phase == "deploy":
        blocker = _final_blocker_label(bundle)
        if blocker and _build_error_observation(bundle):
            return blocker
    if phase == "live_validation":
        blocker = _final_blocker_label(bundle)
        if blocker and _final_visual_status(_verdict(bundle)) == "error":
            return "Live validation incomplete"
        signals = _signals(bundle)
        smoke = _smoke(bundle)
        if signals.get("generation_smoke_passed") is True or _lower(smoke.get("status")) == "success":
            latency = smoke.get("latency_seconds") or smoke.get("observed_latency_seconds")
            return f"Generation passed in {latency}s" if latency is not None else "Generation passed"
        if signals.get("health_passed") is True:
            return "Health passed; generation not verified"
        live = build_live_validation_model(bundle)
        return str(live.get("label") or ("Live validation pending" if status in {"pending", "running"} else "Live validation incomplete"))
    if phase == "recovery":
        steps = _event_steps(bundle)
        if not steps.intersection(PHASE_STEPS["recovery"]):
            return "Not needed"
        if any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in FAILED_STATUSES for e in _events(bundle)):
            return "Patch attempted; final blocker remains"
        if any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in SUCCESS_STATUSES for e in _events(bundle)):
            return "Repair revalidated"
        latest = _latest_recovery_event(bundle) or {}
        latest_step = _lower(latest.get("step"))
        if latest_step in {"repair_validation"}:
            return "Revalidating repair"
        if latest_step in {"repair", "repair_plan", "repair_patch", "repair_upload"}:
            return "Repairing"
        return "Diagnosing repair"
    if phase == "archive":
        eval_status = _eval_publish(bundle)
        if eval_status.get("published") is True:
            return "Eval archive published"
        if eval_status.get("attempted") and not _is_record_not_ready(eval_status):
            return "Eval archive attempted"
        if eval_status.get("attempted") and _is_record_not_ready(eval_status):
            return "Finalizing archive" if status in {"pending", "running"} else "Final artifacts written"
        return "Archive pending" if status in {"pending", "running"} else "Final artifacts written"
    if phase == "done":
        label, _ = _status_label(_verdict(bundle))
        return label
    latest = _latest_event_for_phase(bundle, phase)
    if latest:
        return str(latest.get("message") or PHASE_LABELS[phase])
    if status == "skipped":
        return "Not needed"
    return "Pending" if status == "pending" else PHASE_LABELS[phase]


def _phase_status(bundle: dict[str, Any], phase: str, verdict: str) -> str:
    steps = _event_steps(bundle)
    terminal = _final_visual_status(verdict) in {"success", "warn", "error", "stopped"}
    terminal_success = _terminal_success(bundle, verdict)

    if terminal_success and phase not in {"done", "recovery"}:
        # v198.26.25: historical/full reloads can arrive with only final
        # artifacts or recent events. A full inference success proves that all
        # pre-live phases completed, so do not leave early dots gray.
        if phase == "hardware" and _hardware_warning(bundle):
            return "warning"
        return "complete"

    if phase == "done":
        if verdict in {"full_inference_success", "success", "passed"}:
            return "complete"
        if verdict in {"partial_validation", "partial", "completed_with_warnings", "health_only"}:
            return "warning"
        if verdict in {"failed", "technical_blocker", "technical_blocker_boot_only"}:
            return "failed"
        if verdict in {"manual_action_required", "cancelled"}:
            return "stopped"
        return "pending"

    if phase == "recovery":
        if terminal_success and not steps.intersection(PHASE_STEPS["recovery"]):
            return "skipped"
        if not steps.intersection(PHASE_STEPS["recovery"]):
            return "skipped" if terminal else "pending"
        if _final_visual_status(verdict) == "success":
            return "complete"
        if any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in FAILED_STATUSES for e in _events(bundle)):
            return "failed"
        if any(_lower(e.get("step")) == "repair_validation" and _lower(e.get("status")) in SUCCESS_STATUSES for e in _events(bundle)):
            return "complete"
        latest_recovery = _latest_recovery_event(bundle) or {}
        latest_status = _lower(latest_recovery.get("status"))
        if latest_status in FAILED_STATUSES:
            return "failed"
        return "warning" if terminal else "running"

    if phase == "agent":
        latest_by_step = _latest_events_by_step(bundle, phase)
        pi_run = latest_by_step.get("pi_run") or {}
        # v198.26.9: terminal success with downstream proof closes Agent.  An
        # old pi_run=running signal must not keep the visible dot pulsing after
        # deploy/live/archive/final success has already happened.
        if terminal and _final_visual_status(verdict) == "success" and _has_downstream_proof(bundle, phase):
            return "complete"
        if _lower(pi_run.get("status")) in RUNNING_STATUSES and not terminal:
            return "running"
        if terminal:
            material_events = [e for e in _events_for_phase(bundle, phase) if _lower(e.get("step")) != "pi_model_resolution"]
            material_statuses = {_lower(e.get("status")) for e in material_events}
            if material_statuses.intersection(FAILED_STATUSES) and _final_visual_status(verdict) != "success":
                return "failed"
            if material_statuses.intersection(RUNNING_STATUSES) and not _has_downstream_proof(bundle, phase):
                return "running"
            if material_statuses.intersection(SUCCESS_STATUSES) or _has_downstream_proof(bundle, phase):
                return "complete"

    if phase == "deploy":
        blocker = _final_blocker_label(bundle)
        if blocker and _build_error_observation(bundle):
            return "failed"
    if phase == "live_validation":
        blocker = _final_blocker_label(bundle)
        if blocker and _final_visual_status(_verdict(bundle)) == "error":
            return "failed"
        signals = _signals(bundle)
        smoke = _smoke(bundle)
        if _final_visual_status(verdict) == "success":
            return "complete"
        if signals.get("generation_smoke_passed") is True or _lower(smoke.get("status")) == "success":
            return "complete"
        if signals.get("health_passed") is True or smoke:
            return "warning"
        return _phase_status_from_events(bundle, phase)

    if phase == "archive":
        eval_status = _eval_publish(bundle)
        if eval_status.get("published") is True:
            return "complete"
        if eval_status.get("attempted"):
            if _is_record_not_ready(eval_status):
                return "pending" if not terminal else "complete"
            return "warning"
        if terminal:
            return "complete"
        return "pending"

    if phase == "deploy" and _build_error_observation(bundle):
        return "failed"

    # Pi model changes are surfaced as run notes, not phase warnings.
    if phase == "hardware" and _hardware_warning(bundle):
        return "warning"

    event_status = _phase_status_from_events(bundle, phase)
    if event_status != "pending":
        if event_status == "running" and terminal and (_has_downstream_proof(bundle, phase) or _has_successful_phase_event(bundle, phase)):
            return "complete"
        return event_status

    # On terminal runs, phases before the first missing/optional later phase can
    # be considered complete when downstream proofs exist.
    if terminal and _has_downstream_proof(bundle, phase):
        return "complete"
    return "pending"


def build_run_timeline_model(bundle: dict[str, Any]) -> dict[str, Any]:
    """Return the canonical product progress/timeline projection for a run.

    This model is deliberately derived from persisted run artifacts and events,
    not from transient DOM state. It separates process completion from the final
    result verdict so a completed run can be success, partial, failed, blocked,
    or manual-action-required without the progress bar lying.
    """
    bundle = bundle or {}
    state = bundle.get("state") or {}
    progress = progress_from_events(_events(bundle), state=state)
    verdict = _verdict(bundle)
    visual_status = _final_visual_status(verdict)
    label, subtitle = _status_label(verdict)
    terminal = visual_status in {"success", "warn", "error", "stopped"}
    percent = 100 if terminal else int(progress.get("progress") or 0)

    warnings = [w for w in (_pi_model_warning(bundle), _hardware_warning(bundle), _logs_warning(bundle), _archive_warning(bundle, terminal=terminal)) if w]
    phases: list[dict[str, Any]] = []
    for phase in PHASE_ORDER:
        status = _phase_status(bundle, phase, verdict)
        phases.append(
            {
                "id": phase,
                "label": PHASE_LABELS[phase],
                "status": status,
                "summary": _phase_summary(bundle, phase, status),
                "details": _build_phase_details(bundle, phase),
            }
        )

    return {
        "schema_version": "run_timeline_model.v1",
        "progress": {
            "percent": max(0, min(100, percent)),
            "visual_status": visual_status,
            "label": label,
            "subtitle": subtitle,
            "process_status": progress.get("status") or "unknown",
            "verdict": verdict,
            "terminal": terminal,
        },
        "phases": phases,
        "warnings": warnings,
        "live_validation": build_live_validation_model(bundle),
    }