@deftai/directive-content 0.55.1 → 0.56.0

package/scripts/capacity_show.py

@@ -0,0 +1,653 @@
+#!/usr/bin/env python3
+"""capacity_show.py -- offline capacity-allocation accounting (#1419 Slice 4).
+
+Surfaced via ``task capacity:show``. Derives per-bucket tallies directly from
+the vBRIEF lifecycle folders (``vbrief/{proposed,pending,active,completed,
+cancelled}/``) -- filesystem-truth, fully offline, no ``gh`` / network calls.
+
+What it reports
+---------------
+For every protected bucket declared in
+``plan.policy.capacityAllocation.buckets`` (see ``scripts/policy.py``):
+
+* **Forward view** -- the in-flight mix: summed kind-aware weight of
+  ``pending/`` + ``active/`` vBRIEFs in the bucket.
+* **Backward view** -- the trailing-window mix: summed weight of ``completed/``
+  vBRIEFs whose ``plan.metadata.completedAt`` falls inside the configured
+  ``window`` (days). The backward view drives the target-vs-actual *deficit*
+  column (acceptance a4).
+* **Outcome (rework) overlay** -- weight of completed-in-window vBRIEFs flagged
+  as rework (``plan.metadata.outcome == "rework"`` or
+  ``plan.metadata.rework == true``).
+* **Cost overlay** -- summed grounded cost actuals (``plan.metadata.cost``) of
+  completed-in-window vBRIEFs; rendered ``none/estimate-only`` when no grounded
+  actuals exist.
+
+Kind-aware counting (acceptance a2)
+-----------------------------------
+* ``kind == "story"`` (or unset) counts its own weight (1).
+* An ``epic`` / ``phase`` that HAS children on disk (a ``plan.references[]``
+  entry of ``type == "x-vbrief/plan"``) counts 0 -- its children are counted
+  directly.
+* An UNDECOMPOSED ``epic`` / ``phase`` (no child references) counts its
+  ``plan.metadata.estimatedChildren`` (or the policy ``defaultEpicEstimate``,
+  framework default 3).
+
+Advisory guards
+---------------
+* **minSampleSize (acceptance a1)** -- when the number of classified
+  completions in the window is below ``minSampleSize``, the engine reports
+  advisory mode and defers to ordering (the deficit numbers are still printed
+  but flagged as not yet load-bearing).
+* **unit:cost guarded fallback (acceptance a5 / OQ2)** -- ``unit == "cost"`` is
+  selectable but self-guards: when grounded cost actuals are insufficient
+  (coverage below :data:`COST_COVERAGE_FLOOR`), the engine falls back to the
+  advisory ``vbrief-count`` unit and warns; the cost overlay renders
+  ``none/estimate-only``. The cost-sync telemetry itself is out of scope /
+  upstream-blocked.
+
+This surface is ADVISORY: it always exits 0. The companion three-state gate
+``scripts/verify_capacity.py`` (``task verify:capacity``) is where an opt-in
+``enforce`` posture can surface a non-zero exit.
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+from dataclasses import dataclass, field
+from datetime import UTC, datetime
+from pathlib import Path
+
+# Make sibling helpers importable both as __main__ and when imported by tests.
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+from _stdio_utf8 import reconfigure_stdio  # noqa: E402
+from policy import (  # noqa: E402, I001
+    CAPACITY_UNIT_COST,
+    DEFAULT_CAPACITY_UNIT,
+    DEFAULT_PENDING_DECISIONS_THRESHOLD,
+    AutonomyRecommendation,
+    CapacityAllocation,
+    pending_decisions_nudge_line,
+    recommend_autonomy_level,
+    resolve_autonomy,
+    resolve_capacity_allocation,
+    summarize_decision_backlog,
+)
+
+reconfigure_stdio()
+
+# ---------------------------------------------------------------------------
+# Constants
+# ---------------------------------------------------------------------------
+
+#: Lifecycle folders that contribute to the FORWARD (in-flight) view.
+FORWARD_FOLDERS: tuple[str, ...] = ("pending", "active")
+
+#: Lifecycle folder that contributes to the BACKWARD (trailing-window) view.
+BACKWARD_FOLDER: str = "completed"
+
+#: Bucket label used when a vBRIEF carries no explicit ``capacityBucket`` and
+#: the policy declares no ``defaultBucket``.
+UNASSIGNED_BUCKET: str = "unassigned"
+
+#: Minimum fraction of classified completions that must carry a positive cost
+#: actual before ``unit:cost`` is treated as grounded. Below this the engine
+#: falls back to advisory ``vbrief-count`` (acceptance a5 / OQ2).
+COST_COVERAGE_FLOOR: float = 0.5
+
+#: Kinds treated as parents whose undecomposed estimate is counted (a2).
+_PARENT_KINDS: frozenset[str] = frozenset({"epic", "phase"})
+
+
+# ---------------------------------------------------------------------------
+# Data model
+# ---------------------------------------------------------------------------
+
+
+@dataclass(frozen=True)
+class VbriefRecord:
+    """One vBRIEF's capacity-relevant facts, derived from disk."""
+
+    bucket: str
+    kind: str
+    weight: float
+    folder: str
+    classified: bool  # carried an explicit capacityBucket
+    in_window: bool  # completed within the trailing window (backward only)
+    completed_at_present: bool  # a non-empty completedAt string is on disk
+    is_rework: bool
+    cost: float | None
+
+
+@dataclass
+class BucketTally:
+    """Per-bucket forward/backward/rework/cost rollup."""
+
+    bucket_id: str
+    target: float
+    forward_weight: float = 0.0
+    backward_weight: float = 0.0
+    rework_weight: float = 0.0
+    cost_actual: float | None = None
+
+
+@dataclass
+class CapacityReport:
+    """Computed capacity report -- the testable core of ``capacity:show``."""
+
+    configured: bool
+    source: str
+    unit_requested: str
+    unit_effective: str
+    cost_fallback: bool
+    cost_fallback_reason: str | None
+    window_days: int
+    min_sample_size: int
+    classified_completions: int
+    unclassified_completions: int
+    advisory_mode: bool
+    advisory_reasons: list[str] = field(default_factory=list)
+    buckets: list[BucketTally] = field(default_factory=list)
+    total_forward: float = 0.0
+    total_backward: float = 0.0
+    policy_error: str | None = None
+    # Pending human-clearance backlog + earned-autonomy dial (#1419 Slice 5).
+    pending_decisions: int = 0
+    pending_decisions_threshold: int = DEFAULT_PENDING_DECISIONS_THRESHOLD
+    pending_by_kind: dict[str, int] = field(default_factory=dict)
+    pending_nudge: str = ""
+    autonomy_enabled: bool = True
+    autonomy: AutonomyRecommendation | None = None
+
+    def bucket_deficit(self, tally: BucketTally) -> float:
+        """Backward target-vs-actual deficit (positive == under target).
+
+        ``deficit = target_fraction * total_backward - backward_weight``. A
+        positive value means the bucket received LESS than its target share of
+        the trailing window's completed work -- i.e. it is being starved.
+        """
+        target_weight = tally.target * self.total_backward
+        return round(target_weight - tally.backward_weight, 4)
+
+
+# ---------------------------------------------------------------------------
+# Derivation helpers
+# ---------------------------------------------------------------------------
+
+
+def _parse_iso(value: object) -> datetime | None:
+    """Parse an ISO-8601 ``...Z`` timestamp to an aware datetime, or None."""
+    if not isinstance(value, str) or not value:
+        return None
+    try:
+        parsed = datetime.fromisoformat(value.replace("Z", "+00:00"))
+    except ValueError:
+        return None
+    if parsed.tzinfo is None:
+        parsed = parsed.replace(tzinfo=UTC)
+    return parsed
+
+
+def _plan_has_children(plan: dict) -> bool:
+    """True when the plan declares an ``x-vbrief/plan`` child reference."""
+    refs = plan.get("references")
+    if not isinstance(refs, list):
+        return False
+    return any(
+        isinstance(ref, dict) and ref.get("type") == "x-vbrief/plan"
+        for ref in refs
+    )
+
+
+def _coerce_cost(value: object) -> float | None:
+    """Return a positive numeric cost actual, or None when absent/invalid."""
+    if isinstance(value, bool):
+        return None
+    if isinstance(value, int | float) and value > 0:
+        return float(value)
+    return None
+
+
+def classify_record(
+    plan: dict,
+    folder: str,
+    allocation: CapacityAllocation,
+    now: datetime,
+) -> VbriefRecord:
+    """Derive a :class:`VbriefRecord` from a single vBRIEF ``plan`` block."""
+    metadata = plan.get("metadata") if isinstance(plan.get("metadata"), dict) else {}
+    kind_raw = metadata.get("kind")
+    kind = kind_raw if isinstance(kind_raw, str) and kind_raw else "story"
+
+    explicit_bucket = metadata.get("capacityBucket")
+    classified = isinstance(explicit_bucket, str) and bool(explicit_bucket.strip())
+    if classified:
+        bucket = explicit_bucket.strip()
+    elif allocation.default_bucket:
+        bucket = allocation.default_bucket
+    else:
+        bucket = UNASSIGNED_BUCKET
+
+    weight = _record_weight(kind, plan, metadata, allocation)
+
+    in_window = False
+    # Mirror capacity_backfill's ``has_completed_at`` semantics: a non-empty
+    # string counts as present even if it does not parse (backfill preserves it
+    # and never re-stamps, so such an item can never enter the window).
+    raw_completed_at = metadata.get("completedAt")
+    completed_at_present = isinstance(raw_completed_at, str) and bool(
+        raw_completed_at.strip()
+    )
+    if folder == BACKWARD_FOLDER:
+        completed_at = _parse_iso(raw_completed_at)
+        if completed_at is not None:
+            age_days = (now - completed_at).total_seconds() / 86400.0
+            in_window = 0 <= age_days <= allocation.window_days
+
+    outcome = metadata.get("outcome")
+    is_rework = (isinstance(outcome, str) and outcome.lower() == "rework") or (
+        metadata.get("rework") is True
+    )
+
+    return VbriefRecord(
+        bucket=bucket,
+        kind=kind,
+        weight=weight,
+        folder=folder,
+        classified=classified,
+        in_window=in_window,
+        completed_at_present=completed_at_present,
+        is_rework=is_rework,
+        cost=_coerce_cost(metadata.get("cost")),
+    )
+
+
+def _record_weight(
+    kind: str, plan: dict, metadata: dict, allocation: CapacityAllocation
+) -> float:
+    """Kind-aware weight for one vBRIEF (acceptance a2)."""
+    if kind in _PARENT_KINDS:
+        if _plan_has_children(plan):
+            # Decomposed parent -- children are counted directly.
+            return 0.0
+        # Undecomposed epic/phase -- count its estimated children.
+        estimated = metadata.get("estimatedChildren")
+        if isinstance(estimated, int) and not isinstance(estimated, bool) and estimated > 0:
+            return float(estimated)
+        return float(allocation.default_epic_estimate)
+    # Stories (and any unknown kind) count their own single weight.
+    return 1.0
+
+
+def iter_vbrief_plans(vbrief_root: Path) -> list[tuple[str, dict]]:
+    """Yield ``(folder, plan)`` for every readable vBRIEF in the lifecycle dirs.
+
+    Unreadable / malformed files are skipped (offline accounting is
+    best-effort over filesystem-truth, not a validator).
+    """
+    import json
+
+    out: list[tuple[str, dict]] = []
+    for folder in (*FORWARD_FOLDERS, BACKWARD_FOLDER):
+        folder_path = vbrief_root / folder
+        if not folder_path.is_dir():
+            continue
+        for child in sorted(folder_path.iterdir()):
+            if not (child.is_file() and child.name.endswith(".vbrief.json")):
+                continue
+            try:
+                data = json.loads(child.read_text(encoding="utf-8"))
+            except (OSError, ValueError):
+                continue
+            plan = data.get("plan") if isinstance(data, dict) else None
+            if isinstance(plan, dict):
+                out.append((folder, plan))
+    return out
+
+
+# ---------------------------------------------------------------------------
+# Report computation
+# ---------------------------------------------------------------------------
+
+
+def compute_report(
+    project_root: Path,
+    *,
+    now: datetime | None = None,
+    allocation: CapacityAllocation | None = None,
+) -> CapacityReport:
+    """Compute the :class:`CapacityReport` for *project_root* (offline)."""
+    now = now or datetime.now(UTC)
+    allocation = allocation or resolve_capacity_allocation(project_root)
+    vbrief_root = project_root / "vbrief"
+
+    records = [
+        classify_record(plan, folder, allocation, now)
+        for folder, plan in iter_vbrief_plans(vbrief_root)
+    ]
+
+    # Seed tallies from the configured buckets (preserving declaration order),
+    # then add any extra buckets discovered on disk (sorted) so unassigned /
+    # mis-tagged work is still visible.
+    tallies: dict[str, BucketTally] = {}
+    for bucket in allocation.buckets:
+        tallies[bucket.bucket_id] = BucketTally(
+            bucket_id=bucket.bucket_id, target=bucket.target
+        )
+    for record in records:
+        if record.bucket not in tallies:
+            tallies[record.bucket] = BucketTally(bucket_id=record.bucket, target=0.0)
+
+    classified_completions = 0
+    # Completed vBRIEFs that carry no explicit capacityBucket AND that a backfill
+    # could actually pull into the window-scoped classified set (#1606). An
+    # unclassified completion with an explicit completedAt OUTSIDE the trailing
+    # window is EXCLUDED: backfill would stamp its bucket but leave completedAt
+    # out of window, so classified_completions never rises and advisory mode
+    # would persist silently -- promising such a migration is misleading. An
+    # absent completedAt is included: backfill stamps the git landing time,
+    # which may land in window. A positive count while sample-short means
+    # `task capacity:backfill` can classify history and cross minSampleSize.
+    unclassified_completions = sum(
+        1
+        for record in records
+        if record.folder == BACKWARD_FOLDER
+        and not record.classified
+        and (record.in_window or not record.completed_at_present)
+    )
+    cost_eligible = 0  # classified completions in window
+    cost_with_actual = 0
+    for record in records:
+        tally = tallies[record.bucket]
+        if record.folder in FORWARD_FOLDERS:
+            tally.forward_weight += record.weight
+        elif record.folder == BACKWARD_FOLDER and record.in_window:
+            tally.backward_weight += record.weight
+            if record.is_rework:
+                tally.rework_weight += record.weight
+            if record.classified:
+                classified_completions += 1
+                cost_eligible += 1
+                if record.cost is not None:
+                    cost_with_actual += 1
+                    tally.cost_actual = (tally.cost_actual or 0.0) + record.cost
+
+    total_forward = sum(t.forward_weight for t in tallies.values())
+    total_backward = sum(t.backward_weight for t in tallies.values())
+    total_rework = sum(t.rework_weight for t in tallies.values())
+
+    # Bucket ordering: configured buckets first (declaration order), then
+    # discovered extras alphabetically for deterministic output.
+    configured_ids = [b.bucket_id for b in allocation.buckets]
+    extras = sorted(bid for bid in tallies if bid not in configured_ids)
+    ordered = [tallies[bid] for bid in (*configured_ids, *extras)]
+
+    unit_requested = allocation.unit if allocation.unit in {
+        DEFAULT_CAPACITY_UNIT,
+        CAPACITY_UNIT_COST,
+    } else DEFAULT_CAPACITY_UNIT
+    unit_effective, cost_fallback, cost_reason = _resolve_effective_unit(
+        unit_requested, cost_eligible, cost_with_actual
+    )
+
+    advisory_reasons: list[str] = []
+    if not allocation.configured:
+        advisory_reasons.append(
+            "capacityAllocation not configured -- showing discovered buckets only"
+        )
+    sample_short = classified_completions < allocation.min_sample_size
+    if sample_short:
+        advisory_reasons.append(
+            f"only {classified_completions} classified completion(s) in window "
+            f"(< minSampleSize={allocation.min_sample_size}) -- deferring to ordering"
+        )
+        # Actionable discoverability hint (#1606): when buckets ARE configured
+        # but completed history is unclassified, point the operator at the
+        # one-time backfill that crosses minSampleSize.
+        if allocation.configured and unclassified_completions > 0:
+            advisory_reasons.append(
+                f"{unclassified_completions} completed vBRIEF(s) are unclassified "
+                "-- run `task capacity:backfill --apply` (one-time) to classify "
+                "history and activate capacity accounting"
+            )
+    if cost_fallback and cost_reason:
+        advisory_reasons.append(cost_reason)
+
+    # Pending human-clearance backlog + earned-autonomy dial (#1419 Slice 5).
+    # The backlog count is derived from the durable audit log; the autonomy
+    # dial is computed from the override rate (primary) + rework rate
+    # (guardrail) over the SAME trailing window and is ADVISORY-ONLY -- the
+    # recommendation is surfaced, never auto-applied.
+    backlog = summarize_decision_backlog(
+        project_root, now=now, window_days=allocation.window_days
+    )
+    rework_rate = total_rework / total_backward if total_backward > 0 else 0.0
+    autonomy_policy = resolve_autonomy(project_root)
+    # Honour the enabled flag: a project that sets autonomy.enabled=false gets
+    # no dial recommendation at all (autonomy stays None), so the render guard
+    # below suppresses the line. Default policy is enabled.
+    autonomy = (
+        recommend_autonomy_level(
+            autonomy_policy.default_level,
+            override_rate=backlog.override_rate,
+            rework_rate=rework_rate,
+            sample_size=backlog.resolved_in_window,
+            p0_reversal=backlog.p0_reversal_in_window,
+            policy=autonomy_policy,
+        )
+        if autonomy_policy.enabled
+        else None
+    )
+    pending_nudge = pending_decisions_nudge_line(backlog.pending_count)
+
+    return CapacityReport(
+        configured=allocation.configured,
+        source=allocation.source,
+        unit_requested=unit_requested,
+        unit_effective=unit_effective,
+        cost_fallback=cost_fallback,
+        cost_fallback_reason=cost_reason,
+        window_days=allocation.window_days,
+        min_sample_size=allocation.min_sample_size,
+        classified_completions=classified_completions,
+        unclassified_completions=unclassified_completions,
+        advisory_mode=sample_short or not allocation.configured,
+        advisory_reasons=advisory_reasons,
+        buckets=ordered,
+        total_forward=total_forward,
+        total_backward=total_backward,
+        policy_error=allocation.error,
+        pending_decisions=backlog.pending_count,
+        pending_by_kind=dict(backlog.by_kind),
+        pending_nudge=pending_nudge,
+        autonomy_enabled=autonomy_policy.enabled,
+        autonomy=autonomy,
+    )
+
+
+def _resolve_effective_unit(
+    unit_requested: str, cost_eligible: int, cost_with_actual: int
+) -> tuple[str, bool, str | None]:
+    """Apply the unit:cost guarded fallback (acceptance a5 / OQ2).
+
+    Returns ``(unit_effective, cost_fallback, reason)``. ``cost`` falls back to
+    advisory ``vbrief-count`` when grounded actuals are insufficient (no
+    eligible completions, or coverage below :data:`COST_COVERAGE_FLOOR`).
+    """
+    if unit_requested != CAPACITY_UNIT_COST:
+        return unit_requested, False, None
+    if cost_eligible == 0:
+        return (
+            DEFAULT_CAPACITY_UNIT,
+            True,
+            "unit:cost requested but no classified completions carry grounded "
+            "cost actuals -- falling back to advisory vbrief-count "
+            "(cost overlay: none/estimate-only)",
+        )
+    coverage = cost_with_actual / cost_eligible
+    if coverage < COST_COVERAGE_FLOOR:
+        return (
+            DEFAULT_CAPACITY_UNIT,
+            True,
+            f"unit:cost requested but only {cost_with_actual}/{cost_eligible} "
+            f"({coverage:.0%}) classified completions carry grounded cost "
+            f"actuals (< {COST_COVERAGE_FLOOR:.0%} floor) -- falling back to "
+            "advisory vbrief-count (cost overlay: none/estimate-only)",
+        )
+    return CAPACITY_UNIT_COST, False, None
+
+
+# ---------------------------------------------------------------------------
+# Rendering
+# ---------------------------------------------------------------------------
+
+
+def _format_cost(value: float | None) -> str:
+    """Render the cost-overlay cell."""
+    if value is None:
+        return "none/estimate-only"
+    return f"{value:.2f}"
+
+
+def _append_backlog_and_autonomy(lines: list[str], report: CapacityReport) -> None:
+    """Append the pending-decisions backlog + advisory autonomy block (#1419 S5)."""
+    lines.append(
+        f"  Pending human decisions: {report.pending_decisions} "
+        f"(threshold {report.pending_decisions_threshold})"
+    )
+    if report.pending_by_kind:
+        kinds = ", ".join(
+            f"{kind}={count}"
+            for kind, count in sorted(report.pending_by_kind.items())
+        )
+        lines.append(f"    by kind: {kinds}")
+    if report.pending_nudge:
+        lines.append(f"  {report.pending_nudge}")
+    if report.autonomy_enabled and report.autonomy is not None:
+        rec = report.autonomy
+        lines.append(
+            f"  Autonomy dial (advisory-only): {rec.current_level} -> "
+            f"{rec.recommended_level} [{rec.action}]"
+        )
+        lines.append(f"    {rec.rationale}")
+
+
+def render_report(report: CapacityReport) -> str:
+    """Render the :class:`CapacityReport` as a human-readable text block."""
+    lines: list[str] = []
+    lines.append("Capacity allocation (advisory, offline / filesystem-truth)")
+    lines.append(
+        f"  unit: {report.unit_effective}"
+        + (
+            f" (requested {report.unit_requested}; cost fallback active)"
+            if report.cost_fallback
+            else ""
+        )
+    )
+    lines.append(
+        f"  window: trailing {report.window_days}d | "
+        f"classified completions: {report.classified_completions} "
+        f"(minSampleSize {report.min_sample_size}) | source: {report.source}"
+    )
+    # Surface the schema error when a malformed capacityAllocation block fell
+    # back to defaults (source == 'default-on-error') so the operator sees the
+    # actual reason, not just a generic "not configured" advisory line.
+    if report.policy_error:
+        lines.append(f"  CONFIG ERROR: {report.policy_error}")
+
+    if report.advisory_mode:
+        lines.append("  MODE: ADVISORY -- deferring to selection ordering.")
+    for reason in report.advisory_reasons:
+        lines.append(f"    - {reason}")
+
+    _append_backlog_and_autonomy(lines, report)
+
+    if not report.buckets:
+        lines.append("  (no buckets configured and no classified work on disk)")
+        return "\n".join(lines)
+
+    header = (
+        f"  {'bucket':<16} {'target':>7} {'fwd':>7} {'back':>7} "
+        f"{'deficit':>8} {'rework':>7} {'cost':>18}"
+    )
+    lines.append(header)
+    lines.append("  " + "-" * (len(header) - 2))
+    for tally in report.buckets:
+        deficit = report.bucket_deficit(tally)
+        lines.append(
+            f"  {tally.bucket_id:<16} "
+            f"{tally.target * 100:>6.1f}% "
+            f"{tally.forward_weight:>7.1f} "
+            f"{tally.backward_weight:>7.1f} "
+            f"{deficit:>+8.2f} "
+            f"{tally.rework_weight:>7.1f} "
+            f"{_format_cost(tally.cost_actual):>18}"
+        )
+    lines.append(
+        f"  {'TOTAL':<16} {'':>7} {report.total_forward:>7.1f} "
+        f"{report.total_backward:>7.1f}"
+    )
+    if report.cost_fallback:
+        lines.append(
+            "  Note: cost overlay shows none/estimate-only -- no grounded cost "
+            "telemetry (out of scope / upstream-blocked, OQ2)."
+        )
+    return "\n".join(lines)
+
+
+# ---------------------------------------------------------------------------
+# CLI
+# ---------------------------------------------------------------------------
+
+
+def evaluate(
+    project_root: Path, *, now: datetime | None = None
+) -> tuple[int, CapacityReport | None, str]:
+    """Pure entry point: returns ``(exit_code, report, rendered_text)``.
+
+    ``capacity:show`` is a display surface, so the exit code is always 0 on a
+    valid project root; only an invalid ``--project-root`` yields exit 2.
+    """
+    if not project_root.is_dir():
+        return 2, None, (
+            f"capacity_show: --project-root is not a directory: {project_root}\n"
+            "  Recovery: pass an existing project root."
+        )
+    report = compute_report(project_root, now=now)
+    return 0, report, render_report(report)
+
+
+def _build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="capacity_show.py",
+        description=(
+            "Offline capacity-allocation accounting (#1419 Slice 4). Derives "
+            "per-bucket target-vs-actual mix from the vBRIEF lifecycle folders "
+            "with outcome (rework) and cost overlays. Advisory -- always exits 0 "
+            "on a valid project root."
+        ),
+    )
+    parser.add_argument(
+        "--project-root",
+        default=".",
+        help="Project root path (default: current working directory).",
+    )
+    return parser
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = _build_parser()
+    args = parser.parse_args(argv)
+    project_root = Path(args.project_root).resolve()
+    code, _report, message = evaluate(project_root)
+    if code == 0:
+        print(message)
+    else:
+        print(message, file=sys.stderr)
+    return code
+
+
+if __name__ == "__main__":
+    sys.exit(main())