@antoneeo/mkt-agentic-sdlc-skill 0.14.0 → 0.15.0

package/skills/mkt-agentic-sdlc/scripts/mkt_check.py

@@ -1,718 +1,752 @@
-#!/usr/bin/env python3
-"""Domain entry point with shared project memory.
-
-Ship this file, sdlc_core.py and knowledge.py together. The core alone omits
-knowledge integrity and this entry point\'s domain-specific behavior.
-"""
-
-import argparse
-import re
-import sys
-from pathlib import Path
-
-sys.path.insert(0, str(Path(__file__).resolve().parent))
-
-try:
-    import sdlc_core
-except ImportError as exc:  # pragma: no cover - exercised by the copied-file test
-    sys.stderr.write(
-        "[ERROR] mkt_check.py cannot find sdlc_core.py next to it: " + str(exc) + "\n"
-        "        The validator needs entry point, sdlc_core.py and knowledge.py.\n"
-        "        Copy all three together; core alone omits memory and overlay checks.\n")
-    sys.exit(1)
-
-# This distribution IS the marketing lens. The default docs root is its own; the
-# core still resolves `--docs-dir` and the env seam on top of it, so a project that
-# has migrated to the family's single `ai_docs/` tree is reachable from here too.
-DOMAIN = "marketing"
-import knowledge
-MKT_DOCS_DIR = "mkt_docs"
-sdlc_core.set_entry_point(DOMAIN, provides=("marketing",), script="mkt_check.py")
-sdlc_core.set_docs_dir(MKT_DOCS_DIR)
-
-sdlc_core.set_profile(
-    skill_name="mkt-agentic-sdlc",
-    unit_noun="engagement",
-    support_files=("templates.md", "frameworks.md", "research.md", "elicitation.md",
-                   "review.md", "routing.md", "guides.md", "vision.md", "hybrid.md", "dispatch.md",
-                   "ENFORCEMENT.md", "memory.md"),
-    capabilities=(
-        # spine
-        "triage", "write_triggers", "workstream_registry", "vision_gate",
-        "design_review_gate", "guide_router", "worktree_hygiene",
-        # marketing overlay
-        "subagent_dispatch", "question_discipline",
-    ),
-    remind_line=(
-        "first decide if this skill governs the work; if yes, LOAD it once -- an "
-        "unread protocol cannot be applied. Declare the engagement level and the "
-        "guide-router verdict before acting. Every claim about the market, the "
-        "audience or a competitor carries its evidence -- an unsourced assertion "
-        "is a finding, not a premise. Replay governing text -- a brief, a "
-        "positioning line, a template -- against the case that motivated it."
-    ),
-    # The strategy is reviewed before any tactic is executed: nine phases, but the
-    # ordering the gate cares about is the same one every domain owes.
-    design_gate_between=("### 6. Strategy", "### 8. Action"),
-)
-
-# The subcommands this overlay implements itself. Anything else on the command
-# line is handed straight to the shared core, so a spine command this file never
-# heard of (as `migrate` once was) cannot be silently dropped by a stale list here.
-OVERLAY_COMMANDS = ("check", "validate", "ledger", "budget", "funnel", "trace", "index")
-
-CANONICAL_DIRS = ("vision", "strategy", "tactics", "deliverables")
-VALID_STATUS = ("CURRENT", "SUPERSEDED", "DRAFT", "DEPRECATED")
-LEDGER_CLASSES = ("FACT", "BENCHMARK", "ASSUMPTION")
-CONFIDENCE_VALUES = ("HIGH", "MED", "MEDIUM", "LOW")
-EV_REF = re.compile(r"\[(EV-\d+)\]")
-# A source that points INSIDE this engagement instead of at the world: a document of
-# OUR OWN under a canonical dir, or the "see X" form. Only FACT can carry it today,
-# because FACT is the one class the URL rule does not reach.
-INTERNAL_SOURCE_RE = re.compile(
-    r"(?:\b(?:mkt_docs|ai_docs|research|strategy|tactics|deliverables|vision)/\S*\.md\b"
-    r"|\b(?:see|vedi|cfr\.?)\s)", re.IGNORECASE)
-# The two origins research.md sanctions for a FACT. A cell that opens by naming the
-# client is exempt: their own primary data may well be a Markdown file they handed
-# over, and telling its author to "reclassify as BENCHMARK and add a URL" would be
-# advice to misclassify.
-CLIENT_ORIGIN_RE = re.compile(
-    r"^\s*(?:user|client|owner|utente|cliente|titolare|proprietario)\b", re.IGNORECASE)
-OBJECTIVE_HEADING = re.compile(r"^###\s+(O\d+)\b", re.MULTILINE)
-OBJECTIVE_TOKEN = re.compile(r"\bO\d+\b")
-EXAMPLE_MARKER = "(example)"
-
-BUDGET_TOLERANCE = 0.01   # ±1% on allocation sum
-FUNNEL_TOLERANCE = 0.05   # ±5% on recomputed funnel cells
-
-
-class Report:
-    def __init__(self):
-        self.errors = []
-        self.warnings = []
-
-    def error(self, msg):
-        self.errors.append(msg)
-
-    def warn(self, msg):
-        self.warnings.append(msg)
-
-    def merge(self, other):
-        self.errors.extend(other.errors)
-        self.warnings.extend(other.warnings)
-
-    def print(self, title):
-        for w in self.warnings:
-            print(f"[WARN]  {title}: {w}")
-        for e in self.errors:
-            print(f"[ERROR] {title}: {e}")
-
-    def exit_code(self, strict=False):
-        if self.errors:
-            return 1
-        if strict and self.warnings:
-            return 1
-        return 0
-
-
-# ---------------------------------------------------------------- helpers
-
-def find_root(root_arg, docs_dir=None):
-    """Resolve the project root: --root wins, else walk up from cwd.
-
-    The docs-root NAME comes from the core (`--docs-dir`, then the env seam, then
-    discovery, then this distribution's `mkt_docs`), so one project mid-migration
-    cannot be validated half under one root and half under the other.
-    """
-    if docs_dir:
-        sdlc_core.set_docs_dir(docs_dir)
-    name = sdlc_core.docs_dir()
-    if root_arg:
-        root = Path(root_arg).resolve()
-        return root if (root / name).is_dir() else None
-    cur = Path.cwd().resolve()
-    for candidate in (cur, *cur.parents):
-        if (candidate / name).is_dir():
-            return candidate
-    return None
-
-
-def read_text(path):
-    return path.read_text(encoding="utf-8", errors="replace")
-
-
-def load_frontmatter(text):
-    """Return (dict, body). Tolerant: missing frontmatter -> ({}, text)."""
-    if not text.startswith("---"):
-        return {}, text
-    lines = text.split("\n")
-    fm = {}
-    for i, line in enumerate(lines[1:], start=1):
-        if line.strip() == "---":
-            return fm, "\n".join(lines[i + 1:])
-        m = re.match(r"^([A-Za-z_][\w-]*):\s*(.*)$", line)
-        if m:
-            fm[m.group(1).strip()] = m.group(2).strip()
-    return {}, text  # unterminated frontmatter: treat as body
-
-
-def parse_num(cell):
-    """Parse a table-cell number: strips currency symbols, spaces, commas,
-    percent signs and [EV-nn] refs; supports k/M suffix. Returns float or None."""
-    if cell is None:
-        return None
-    s = EV_REF.sub("", cell)
-    s = re.sub(r"[^\dkKmM.,%-]", "", s.strip())
-    s = s.replace(",", "").replace("%", "")
-    if not s:
-        return None
-    mult = 1.0
-    if s and s[-1] in "kK":
-        mult, s = 1_000.0, s[:-1]
-    elif s and s[-1] in "mM":
-        mult, s = 1_000_000.0, s[:-1]
-    try:
-        return float(s) * mult
-    except ValueError:
-        return None
-
-
-def parse_tables(text):
-    """Return every markdown table as (header_cells_lower, rows) where each
-    row is a dict {header_lower: cell}."""
-    tables = []
-    lines = text.split("\n")
-    i = 0
-    while i < len(lines):
-        line = lines[i].strip()
-        if line.startswith("|") and i + 1 < len(lines) \
-                and re.match(r"^\|[\s\-:|]+\|$", lines[i + 1].strip()):
-            headers = [h.strip().lower() for h in line.strip("|").split("|")]
-            rows = []
-            j = i + 2
-            while j < len(lines) and lines[j].strip().startswith("|"):
-                cells = [c.strip() for c in lines[j].strip().strip("|").split("|")]
-                if len(cells) >= len(headers) - 1:  # tolerate ragged tail
-                    row = {}
-                    for k, h in enumerate(headers):
-                        row[h] = cells[k] if k < len(cells) else ""
-                    rows.append(row)
-                j += 1
-            tables.append((headers, rows))
-            i = j
-        else:
-            i += 1
-    return tables
-
-
-def find_table(text, required_cols):
-    """First table whose header contains every required column (contains-match)."""
-    req = [c.lower() for c in required_cols]
-    for headers, rows in parse_tables(text):
-        if all(any(r in h for h in headers) for r in req):
-            return headers, rows
-        # exact-name pass for short names like 'kpi'
-        if all(r in headers for r in req):
-            return headers, rows
-    return None, None
-
-
-def col(row, name):
-    """Fetch a cell by contains-match on the header name."""
-    name = name.lower()
-    if name in row:
-        return row[name]
-    for h, v in row.items():
-        if name in h:
-            return v
-    return None
-
-
-def is_example_row(row):
-    return any(EXAMPLE_MARKER in (v or "") for v in row.values())
-
-
-def close_enough(stated, computed, tol):
-    if stated is None or computed is None:
-        return True  # missing cells reported separately
-    if computed == 0:
-        return abs(stated) < 1e-9
-    return abs(stated - computed) / abs(computed) <= tol
-
-
-def canonical_files(root):
-    docs = sdlc_core.ai_path(root)
-    for d in CANONICAL_DIRS:
-        base = docs / d
-        if base.is_dir():
-            for p in sorted(base.rglob("*.md")):
-                yield p
-
-
-def all_md_files(root):
-    docs = sdlc_core.ai_path(root)
-    for p in sorted(docs.rglob("*.md")):
-        yield p
-
-
-# ---------------------------------------------------------------- index
-
-def build_index(root):
-    docs = sdlc_core.ai_path(root)
-    lines = [
-        "# mkt_docs INDEX (generated by mkt_check.py index -- do not edit by hand)",
-        "",
-        "| Path | Description | Status |",
-        "|---|---|---|",
-    ]
-    for p in canonical_files(root):
-        if p.name == "INDEX.md":
-            continue
-        fm, body = load_frontmatter(read_text(p))
-        desc = fm.get("description", "")
-        status = fm.get("status", "")
-        if not desc:
-            h1 = re.search(r"^#\s+(.+)$", body, re.MULTILINE)
-            desc = h1.group(1).strip() if h1 else ""
-        rel = p.relative_to(docs).as_posix()
-        lines.append(f"| {rel} | {desc} | {status} |")
-    return "\n".join(lines) + "\n"
-
-
-def cmd_index(root):
-    content = build_index(root)
-    out = sdlc_core.ai_path(root) / "INDEX.md"
-    out.write_text(content, encoding="utf-8")
-    print(f"[OK] wrote {out.relative_to(root).as_posix()}")
-    # F-028: the workstream registry is generated here too. This entry point does
-    # not reuse the spine's cmd_index, so wiring it explicitly is what keeps the
-    # doctrine ("generated by mkt_check.py index") from being false in this lens.
-    return max(sdlc_core.rc_registry(root), knowledge.kb_refresh_indexes(root / sdlc_core.docs_dir()))
-
-
-# ---------------------------------------------------------------- validate
-
-def run_validate(root):
-    rep = Report()
-    docs = sdlc_core.ai_path(root)
-    # F-028: the generated registry against its HANDOFF_*.md sources. Gated on
-    # source presence, so a project that has not converted sees nothing new.
-    workstreams = sdlc_core.list_workstreams(root)
-    if workstreams:
-        hand = docs / "audit" / "handoff.md"
-        if not hand.is_file():
-            rep.error("audit/handoff.md missing while HANDOFF_*.md sources exist: "
-                      "run 'mkt_check.py index'")
-        elif sdlc_core.norm_text(read_text(hand)) !=                 sdlc_core.norm_text(sdlc_core.build_registry(root)):
-            rep.error("audit/handoff.md not aligned with its HANDOFF_*.md sources: "
-                      "run 'mkt_check.py index'. A merge resolved by hand is exactly "
-                      "what this catches")
-        if len(workstreams) > sdlc_core.REGISTRY_CAP:
-            rep.warn(f"{len(workstreams)} open workstreams: the registry is meant to "
-                     f"stay under {sdlc_core.REGISTRY_CAP}")
-    known = {p.name for p in docs.rglob("*.md")}
-    for p in canonical_files(root):
-        if p.name == "INDEX.md":
-            continue
-        rel = p.relative_to(docs).as_posix()
-        fm, _ = load_frontmatter(read_text(p))
-        status = fm.get("status", "")
-        if not status:
-            rep.warn(f"{rel}: missing frontmatter 'status'")
-        elif status.upper() not in VALID_STATUS:
-            rep.warn(f"{rel}: invalid status '{status}' (expected {'|'.join(VALID_STATUS)})")
-        sup = fm.get("supersedes", "")
-        if sup and Path(sup).name not in known:
-            rep.warn(f"{rel}: supersedes '{sup}' which does not exist under mkt_docs/")
-    index_path = docs / "INDEX.md"
-    if not index_path.exists():
-        rep.warn("mkt_docs/INDEX.md missing -- generate with 'mkt_check.py index'")
-    elif read_text(index_path) != build_index(root):
-        rep.warn("mkt_docs/INDEX.md stale -- regenerate with 'mkt_check.py index'")
-    return rep
-
-
-# ---------------------------------------------------------------- ledger
-
-def load_ledger(root):
-    """Return (rows, report). Rows: list of dicts with id/claim/class/... keys."""
-    rep = Report()
-    path = sdlc_core.ai_path(root) / "research" / "evidence_ledger.md"
-    if not path.exists():
-        rep.error("research/evidence_ledger.md missing")
-        return [], rep
-    headers, rows = find_table(read_text(path), ["id", "claim", "class", "source"])
-    if rows is None:
-        rep.error("evidence_ledger.md: ledger table not found "
-                  "(need columns ID | Claim | Class | ... | Source | ...)")
-        return [], rep
-    entries = []
-    seen = set()
-    for row in rows:
-        rid = (col(row, "id") or "").strip()
-        if not rid:
-            continue
-        if is_example_row(row):
-            continue
-        if not re.fullmatch(r"EV-\d+", rid):
-            rep.error(f"ledger: bad ID '{rid}' (expected EV-<n>)")
-            continue
-        if rid in seen:
-            rep.error(f"ledger: duplicate ID {rid}")
-        seen.add(rid)
-        entries.append(row)
-    return entries, rep
-
-
-def run_ledger(root):
-    entries, rep = load_ledger(root)
-    ids = set()
-    for row in entries:
-        rid = col(row, "id").strip()
-        ids.add(rid)
-        cls = (col(row, "class") or "").strip().upper()
-        source = (col(row, "source") or "").strip()
-        date = (col(row, "date") or "").strip()
-        confidence = (col(row, "confidence") or "").strip().upper()
-        if cls not in LEDGER_CLASSES:
-            rep.error(f"ledger {rid}: class '{cls}' invalid (FACT|BENCHMARK|ASSUMPTION)")
-            continue
-        if cls == "BENCHMARK":
-            if "http" not in source.lower():
-                rep.error(f"ledger {rid}: BENCHMARK without a source URL")
-            if not date:
-                rep.error(f"ledger {rid}: BENCHMARK without a date")
-        if cls == "FACT" and not source:
-            rep.error(f"ledger {rid}: FACT without a source -- name the origin "
-                      "('user, Wave 1' or the client's data file)")
-        if (cls == "FACT" and source and not CLIENT_ORIGIN_RE.match(source)
-                and INTERNAL_SOURCE_RE.search(source)):
-            # FACT means "the client told me" or "the client's own primary data" — the
-            # one class that legitimately has no URL. Pointing it at another document
-            # of this engagement is how a researched observation gets in without the
-            # URL its class would owe: the row cites a file that cites the source, the
-            # validator resolves nothing, and "see VOC.md" is exactly the form the
-            # BENCHMARK rule rejects. Observed in a cold-agent field test, 2026-08-02.
-            rep.error(f"ledger {rid}: FACT sourced to an internal document ('{source}') -- "
-                      "a FACT is what the client stated or their own primary data; an "
-                      "observation gathered by research is a BENCHMARK and owes its URL")
-        if cls == "ASSUMPTION":
-            if confidence not in CONFIDENCE_VALUES:
-                rep.error(f"ledger {rid}: ASSUMPTION without confidence (HIGH|MED|LOW)")
-            value = (col(row, "value") or "").strip()
-            if value and not re.search(r"\d\s*[-–—]\s*\d|\bto\b", value):
-                rep.warn(f"ledger {rid}: ASSUMPTION value looks like a point estimate, expected a range")
-    # reference resolution across every md file (ledger file itself excluded)
-    referenced = set()
-    ledger_path = sdlc_core.ai_path(root) / "research" / "evidence_ledger.md"
-    for p in all_md_files(root):
-        if p == ledger_path:
-            continue
-        text = read_text(p)
-        for line in text.split("\n"):
-            if EXAMPLE_MARKER in line:
-                continue
-            for m in EV_REF.finditer(line):
-                referenced.add(m.group(1))
-                if m.group(1) not in ids:
-                    rel = p.relative_to(sdlc_core.ai_path(root)).as_posix()
-                    rep.error(f"{rel}: reference [{m.group(1)}] not found in the ledger")
-    for rid in sorted(ids - referenced, key=lambda x: int(x.split("-")[1])):
-        rep.warn(f"ledger {rid}: never referenced by any document")
-    return rep
-
-
-# ---------------------------------------------------------------- budget
-
-def default_tactical_files(root, file_arg):
-    if file_arg:
-        p = Path(file_arg)
-        if not p.is_absolute():
-            p = root / file_arg
-        return [p] if p.exists() else []
-    tactics = sdlc_core.ai_path(root) / "tactics"
-    files = []
-    if (tactics / "TACTICAL_PLAN.md").exists():
-        files.append(tactics / "TACTICAL_PLAN.md")
-    if tactics.is_dir():
-        files.extend(sorted(tactics.glob("CAMPAIGN_*.md")))
-    return files
-
-
-def run_budget(root, file_arg=None):
-    rep = Report()
-    files = default_tactical_files(root, file_arg)
-    if not files:
-        rep.warn("no tactical plan found (tactics/TACTICAL_PLAN.md) -- budget check skipped")
-        return rep
-    for path in files:
-        rel = path.name
-        text = read_text(path)
-        m = re.search(r"^Total budget:\s*(.+)$", text, re.MULTILINE | re.IGNORECASE)
-        if not m:
-            rep.error(f"{rel}: 'Total budget:' line missing")
-            continue
-        total = parse_num(m.group(1))
-        if total is None or total <= 0:
-            rep.error(f"{rel}: cannot parse Total budget '{m.group(1).strip()}'")
-            continue
-        headers, rows = find_table(text, ["channel", "budget", "share"])
-        if rows is None:
-            rep.error(f"{rel}: Budget Allocation table not found (Channel | Budget | Share)")
-            continue
-        rows = [r for r in rows if not is_example_row(r)]
-        if not rows:
-            rep.error(f"{rel}: Budget Allocation table has no rows")
-            continue
-        alloc = 0.0
-        for r in rows:
-            v = parse_num(col(r, "budget"))
-            if v is None:
-                rep.error(f"{rel}: unparseable budget cell '{col(r, 'budget')}' "
-                          f"for channel '{col(r, 'channel')}'")
-            else:
-                alloc += v
-        if not close_enough(alloc, total, BUDGET_TOLERANCE):
-            rep.error(f"{rel}: allocations sum to {alloc:g}, Total budget is {total:g} "
-                      f"(tolerance 1%)")
-    return rep
-
-
-# ---------------------------------------------------------------- funnel
-
-FUNNEL_CHAIN = (
-    # (stated_col, compute, label)
-    ("clicks", lambda b, cpc, cvr, close, r: b / cpc if cpc else None, "Budget/CPC"),
-    ("leads", lambda b, cpc, cvr, close, r: (r["clicks"] * cvr / 100.0)
-        if (r["clicks"] is not None and cvr is not None) else None, "Clicks*CVR%"),
-    ("customers", lambda b, cpc, cvr, close, r: (r["leads"] * close / 100.0)
-        if (r["leads"] is not None and close is not None) else None, "Leads*Close%"),
-    ("cac", lambda b, cpc, cvr, close, r: (b / r["customers"])
-        if (r["customers"] not in (None, 0) and b is not None) else None, "Budget/Customers"),
-)
-
-
-def run_funnel(root, file_arg=None):
-    rep = Report()
-    files = default_tactical_files(root, file_arg)
-    if not files:
-        rep.warn("no tactical plan found -- funnel check skipped")
-        return rep
-    for path in files:
-        rel = path.name
-        text = read_text(path)
-        headers, rows = find_table(text, ["channel", "budget", "cpc", "cvr", "customers", "cac"])
-        if rows is None:
-            rep.error(f"{rel}: Funnel Model table not found "
-                      "(Channel | Budget | CPC | Clicks | CVR % | Leads | Close % | Customers | CAC)")
-            continue
-        rows = [r for r in rows if not is_example_row(r)]
-        if not rows:
-            rep.error(f"{rel}: Funnel Model table has no rows")
-            continue
-        for r in rows:
-            channel = col(r, "channel") or "?"
-            budget = parse_num(col(r, "budget"))
-            cpc = parse_num(col(r, "cpc"))
-            cvr = parse_num(col(r, "cvr"))
-            close = parse_num(col(r, "close"))
-            stated = {
-                "clicks": parse_num(col(r, "clicks")),
-                "leads": parse_num(col(r, "leads")),
-                "customers": parse_num(col(r, "customers")),
-                "cac": parse_num(col(r, "cac")),
-            }
-            if budget is None or cpc is None:
-                rep.error(f"{rel} [{channel}]: Budget/CPC missing or unparseable")
-                continue
-            # verify each stated cell against computation from STATED upstream
-            # cells (so a single wrong cell does not cascade into four errors)
-            chain_inputs = dict(stated)
-            for name, compute, label in FUNNEL_CHAIN:
-                computed = compute(budget, cpc, cvr, close, chain_inputs)
-                if stated[name] is None:
-                    rep.error(f"{rel} [{channel}]: '{name}' cell missing")
-                    continue
-                if computed is None:
-                    continue  # upstream missing, already reported
-                if not close_enough(stated[name], computed, FUNNEL_TOLERANCE):
-                    rep.error(f"{rel} [{channel}]: {name}={stated[name]:g} but {label} "
-                              f"= {computed:g} (tolerance 5%)")
-    return rep
-
-
-# ---------------------------------------------------------------- trace
-
-def run_trace(root):
-    rep = Report()
-    docs = sdlc_core.ai_path(root)
-    obj_path = docs / "strategy" / "OBJECTIVES.md"
-    if not obj_path.exists():
-        rep.warn("strategy/OBJECTIVES.md missing -- trace check skipped")
-        return rep
-    objectives = set(OBJECTIVE_HEADING.findall(read_text(obj_path)))
-    if not objectives:
-        rep.error("OBJECTIVES.md: no '### O<n> --...' headings found")
-        return rep
-
-    tactical = docs / "tactics" / "TACTICAL_PLAN.md"
-    served = set()
-    if tactical.exists():
-        headers, rows = find_table(read_text(tactical), ["channel", "objective", "kpi"])
-        if rows is None:
-            rep.error("TACTICAL_PLAN.md: Channel Plan table not found "
-                      "(Channel | Objective | KPI | Budget | Owner)")
-        else:
-            for r in rows:
-                if is_example_row(r):
-                    continue
-                tokens = set(OBJECTIVE_TOKEN.findall(col(r, "objective") or ""))
-                if not tokens:
-                    rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: no objective (O<n>) named")
-                for t in tokens:
-                    if t not in objectives:
-                        rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: objective {t} "
-                                  "not defined in OBJECTIVES.md")
-                served |= tokens & objectives
-                if not (col(r, "kpi") or "").strip():
-                    rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: KPI cell empty")
-        for o in sorted(objectives - served):
-            rep.warn(f"objective {o}: no tactic serves it in TACTICAL_PLAN.md")
-    else:
-        rep.warn("tactics/TACTICAL_PLAN.md missing -- tactic trace skipped")
-
-    measure = docs / "tactics" / "MEASUREMENT_PLAN.md"
-    if measure.exists():
-        headers, rows = find_table(read_text(measure), ["objective", "kpi", "target"])
-        if rows is None:
-            rep.error("MEASUREMENT_PLAN.md: KPI table not found (Objective | KPI | Target | ...)")
-        else:
-            measured = set()
-            for r in rows:
-                if is_example_row(r):
-                    continue
-                measured |= set(OBJECTIVE_TOKEN.findall(col(r, "objective") or ""))
-            for o in sorted(objectives - measured):
-                rep.error(f"objective {o}: no KPI row in MEASUREMENT_PLAN.md")
-    else:
-        rep.warn("tactics/MEASUREMENT_PLAN.md missing -- KPI trace skipped")
-    return rep
-
-
-# ---------------------------------------------------------------- check
-
-def run_check(root):
-    rep = Report()
-    for title, fn in (("validate", run_validate), ("ledger", run_ledger),
-                      ("budget", run_budget), ("funnel", run_funnel),
-                      ("trace", run_trace)):
-        sub = fn(root)
-        sub.print(title)
-        rep.merge(sub)
-    return rep
-
-
-# ---------------------------------------------------------------- main
-
-# --- portable checks this distribution exposes -------------------------------
-# The marketing arithmetic, offered to documents owned by ANOTHER domain: a
-# distilled spec that carries a budget table can import `marketing.budget` and have
-# it actually checked, instead of the table sitting in a document whose own
-# validator has no opinion about it (UC8). Imported checks may only ADD findings.
-
-def _report_findings(rep):
-    return ([("error", m) for m in rep.errors]
-            + [("warning", m) for m in rep.warnings])
-
-
-@sdlc_core.portable_check("marketing.ledger")
-def _check_ledger(rel, meta, text):
-    root = sdlc_core.find_project_root()
-    return _report_findings(run_ledger(root)) if root else []
-
-
-@sdlc_core.portable_check("marketing.budget")
-def _check_budget(rel, meta, text):
-    root = sdlc_core.find_project_root()
-    return _report_findings(run_budget(root)) if root else []
-
-
-@sdlc_core.portable_check("marketing.funnel")
-def _check_funnel(rel, meta, text):
-    root = sdlc_core.find_project_root()
-    return _report_findings(run_funnel(root)) if root else []
-
-
-def main(argv=None):
-    argv = list(sys.argv[1:] if argv is None else argv)
-    # Spine commands are the core's, identical in every distribution: hand over
-    # every subcommand this overlay does not explicitly intercept, untouched,
-    # rather than keeping a hand-copied list of spine names here -- which is how
-    # `migrate` got dropped while SKILL.md kept promising it.
-    if argv and not argv[0].startswith("-") and argv[0] not in OVERLAY_COMMANDS:
-        return knowledge.main(argv)
-
-    parser = argparse.ArgumentParser(prog="mkt_check.py", description=__doc__,
-        epilog="Shared memory commands: recall QUERY, graph, corpus, claim-id, anchor, "
-               "export, import, stale, orient. Use COMMAND --help for options.")
-    parser.add_argument("command", choices=list(OVERLAY_COMMANDS))
-    parser.add_argument("file", nargs="?", default=None,
-                        help="optional target file for budget/funnel")
-    parser.add_argument("--root", default=None, help="project root (contains the docs root)")
-    parser.add_argument("--docs-dir", dest="docs_dir", default=None,
-                        help="name of the documentation root (default: mkt_docs; "
-                             "use --docs-dir ai_docs on a project that has migrated "
-                             "to the family's single tree)")
-    parser.add_argument("--strict", action="store_true",
-                        help="warnings also fail the exit code")
-    args = parser.parse_args(argv)
-
-    # Resolve the docs root fresh on EVERY invocation, from this distribution's own
-    # default: the core keeps it in module state, and an entry point that inherited
-    # whatever a previous call left behind would be answering the wrong question.
-    try:
-        discovered, name = sdlc_core.resolve_docs_dir(args, args.root)
-    except sdlc_core.AmbiguousDocsRoot as exc:
-        print("[ERROR] %s" % exc)
-        return 1
-    root = find_root(args.root or discovered, name)
-    if root is None:
-        print(f"[ERROR] no {sdlc_core.docs_dir()}/ directory found "
-              "(use --root or run mkt-sdlc-init)")
-        return 2
-
-    if args.command == "index":
-        return cmd_index(root)
-
-    runners = {
-        "validate": lambda: run_validate(root),
-        "ledger": lambda: run_ledger(root),
-        "budget": lambda: run_budget(root, args.file),
-        "funnel": lambda: run_funnel(root, args.file),
-        "trace": lambda: run_trace(root),
-    }
-    if args.command == "check":
-        rep = run_check(root)
-    else:
-        rep = runners[args.command]()
-        rep.print(args.command)
-
-    code = rep.exit_code(strict=args.strict)
-    if args.command in ("validate", "check"):
-        code = max(code, knowledge.kb_validate_surface(root / sdlc_core.docs_dir(),
-                                                       full=args.command == "check"))
-    if code == 0:
-        print(f"[OK] {args.command} clean "
-              f"({len(rep.warnings)} warning(s))" if rep.warnings else f"[OK] {args.command} clean")
-    else:
-        print(f"[FAIL] {args.command}: {len(rep.errors)} error(s), {len(rep.warnings)} warning(s)")
-    if args.command == "check":
-        # This overlay REPLACES the spine's cmd_check, so the F-041 wiring note
-        # must be re-attached here: it is check-layer behaviour, whoever owns
-        # the check (fail-open and informational -- never the exit code).
-        sdlc_core.print_orient_hook_note(root)
-    return code
-
-
-if __name__ == "__main__":
-    sys.exit(main())
+#!/usr/bin/env python3
+"""Domain entry point with shared project memory.
+
+Ship this file, sdlc_core.py and knowledge.py together. The core alone omits
+knowledge integrity and this entry point\'s domain-specific behavior.
+"""
+
+import argparse
+import re
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+try:
+    import sdlc_core
+except ImportError as exc:  # pragma: no cover - exercised by the copied-file test
+    sys.stderr.write(
+        "[ERROR] mkt_check.py cannot find sdlc_core.py next to it: " + str(exc) + "\n"
+        "        The validator needs entry point, sdlc_core.py and knowledge.py.\n"
+        "        Copy all three together; core alone omits memory and overlay checks.\n")
+    sys.exit(1)
+
+# This distribution IS the marketing lens. The default docs root is its own; the
+# core still resolves `--docs-dir` and the env seam on top of it, so a project that
+# has migrated to the family's single `ai_docs/` tree is reachable from here too.
+DOMAIN = "marketing"
+import knowledge
+MKT_DOCS_DIR = "mkt_docs"
+sdlc_core.set_entry_point(DOMAIN, provides=("marketing",), script="mkt_check.py")
+sdlc_core.set_docs_dir(MKT_DOCS_DIR)
+
+sdlc_core.set_profile(
+    skill_name="mkt-agentic-sdlc",
+    unit_noun="engagement",
+    support_files=("templates.md", "frameworks.md", "research.md", "elicitation.md",
+                   "review.md", "routing.md", "guides.md", "vision.md", "hybrid.md", "dispatch.md",
+                   "ENFORCEMENT.md", "memory.md"),
+    capabilities=(
+        # spine
+        "triage", "write_triggers", "workstream_registry", "vision_gate",
+        "design_review_gate", "guide_router", "worktree_hygiene",
+        # marketing overlay
+        "subagent_dispatch", "question_discipline",
+    ),
+    remind_line=(
+        "first decide if this skill governs the work; if yes, LOAD it once -- an "
+        "unread protocol cannot be applied. Declare the engagement level and the "
+        "guide-router verdict before acting. Every claim about the market, the "
+        "audience or a competitor carries its evidence -- an unsourced assertion "
+        "is a finding, not a premise. Replay governing text -- a brief, a "
+        "positioning line, a template -- against the case that motivated it."
+    ),
+    # The strategy is reviewed before any tactic is executed: nine phases, but the
+    # ordering the gate cares about is the same one every domain owes.
+    design_gate_between=("### 6. Strategy", "### 8. Action"),
+)
+
+# The subcommands this overlay implements itself. Anything else on the command
+# line is handed straight to the shared core, so a spine command this file never
+# heard of (as `migrate` once was) cannot be silently dropped by a stale list here.
+OVERLAY_COMMANDS = ("check", "validate", "ledger", "budget", "funnel", "trace", "index")
+
+CANONICAL_DIRS = ("vision", "strategy", "tactics", "deliverables")
+VALID_STATUS = ("CURRENT", "SUPERSEDED", "DRAFT", "DEPRECATED")
+LEDGER_CLASSES = ("FACT", "BENCHMARK", "ASSUMPTION")
+CONFIDENCE_VALUES = ("HIGH", "MED", "MEDIUM", "LOW")
+EV_REF = re.compile(r"\[(EV-\d+)\]")
+# A source that points INSIDE this engagement instead of at the world: a document of
+# OUR OWN under a canonical dir, or the "see X" form. Only FACT can carry it today,
+# because FACT is the one class the URL rule does not reach.
+INTERNAL_SOURCE_RE = re.compile(
+    r"(?:\b(?:mkt_docs|ai_docs|research|strategy|tactics|deliverables|vision)/\S*\.md\b"
+    r"|\b(?:see|vedi|cfr\.?)\s)", re.IGNORECASE)
+# The two origins research.md sanctions for a FACT. A cell that opens by naming the
+# client is exempt: their own primary data may well be a Markdown file they handed
+# over, and telling its author to "reclassify as BENCHMARK and add a URL" would be
+# advice to misclassify.
+CLIENT_ORIGIN_RE = re.compile(
+    r"^\s*(?:user|client|owner|utente|cliente|titolare|proprietario)\b", re.IGNORECASE)
+OBJECTIVE_HEADING = re.compile(r"^###\s+(O\d+)\b", re.MULTILINE)
+OBJECTIVE_TOKEN = re.compile(r"\bO\d+\b")
+EXAMPLE_MARKER = "(example)"
+
+BUDGET_TOLERANCE = 0.01   # ±1% on allocation sum
+FUNNEL_TOLERANCE = 0.05   # ±5% on recomputed funnel cells
+
+
+class Report:
+    def __init__(self):
+        self.errors = []
+        self.warnings = []
+
+    def error(self, msg):
+        self.errors.append(msg)
+
+    def warn(self, msg):
+        self.warnings.append(msg)
+
+    def merge(self, other):
+        self.errors.extend(other.errors)
+        self.warnings.extend(other.warnings)
+
+    def print(self, title):
+        for w in self.warnings:
+            print(f"[WARN]  {title}: {w}")
+        for e in self.errors:
+            print(f"[ERROR] {title}: {e}")
+
+    def exit_code(self, strict=False):
+        if self.errors:
+            return 1
+        if strict and self.warnings:
+            return 1
+        return 0
+
+
+# ---------------------------------------------------------------- helpers
+
+def find_root(root_arg, docs_dir=None):
+    """Resolve the project root: --root wins, else walk up from cwd.
+
+    The docs-root NAME comes from the core (`--docs-dir`, then the env seam, then
+    discovery, then this distribution's `mkt_docs`), so one project mid-migration
+    cannot be validated half under one root and half under the other.
+    """
+    if docs_dir:
+        sdlc_core.set_docs_dir(docs_dir)
+    name = sdlc_core.docs_dir()
+    if root_arg:
+        root = Path(root_arg).resolve()
+        return root if (root / name).is_dir() else None
+    cur = Path.cwd().resolve()
+    for candidate in (cur, *cur.parents):
+        if (candidate / name).is_dir():
+            return candidate
+    return None
+
+
+def read_text(path):
+    return path.read_text(encoding="utf-8", errors="replace")
+
+
+def load_frontmatter(text):
+    """Return (dict, body). Tolerant: missing frontmatter -> ({}, text)."""
+    if not text.startswith("---"):
+        return {}, text
+    lines = text.split("\n")
+    fm = {}
+    for i, line in enumerate(lines[1:], start=1):
+        if line.strip() == "---":
+            return fm, "\n".join(lines[i + 1:])
+        m = re.match(r"^([A-Za-z_][\w-]*):\s*(.*)$", line)
+        if m:
+            fm[m.group(1).strip()] = m.group(2).strip()
+    return {}, text  # unterminated frontmatter: treat as body
+
+
+def parse_num(cell):
+    """Parse a table-cell number: strips currency symbols, spaces, commas,
+    percent signs and [EV-nn] refs; supports k/M suffix. Returns float or None."""
+    if cell is None:
+        return None
+    s = EV_REF.sub("", cell)
+    s = re.sub(r"[^\dkKmM.,%-]", "", s.strip())
+    s = s.replace(",", "").replace("%", "")
+    if not s:
+        return None
+    mult = 1.0
+    if s and s[-1] in "kK":
+        mult, s = 1_000.0, s[:-1]
+    elif s and s[-1] in "mM":
+        mult, s = 1_000_000.0, s[:-1]
+    try:
+        return float(s) * mult
+    except ValueError:
+        return None
+
+
+def parse_tables(text):
+    """Return every markdown table as (header_cells_lower, rows) where each
+    row is a dict {header_lower: cell}."""
+    tables = []
+    lines = text.split("\n")
+    i = 0
+    while i < len(lines):
+        line = lines[i].strip()
+        if line.startswith("|") and i + 1 < len(lines) \
+                and re.match(r"^\|[\s\-:|]+\|$", lines[i + 1].strip()):
+            headers = [h.strip().lower() for h in line.strip("|").split("|")]
+            rows = []
+            j = i + 2
+            while j < len(lines) and lines[j].strip().startswith("|"):
+                cells = [c.strip() for c in lines[j].strip().strip("|").split("|")]
+                if len(cells) >= len(headers) - 1:  # tolerate ragged tail
+                    row = {}
+                    for k, h in enumerate(headers):
+                        row[h] = cells[k] if k < len(cells) else ""
+                    rows.append(row)
+                j += 1
+            tables.append((headers, rows))
+            i = j
+        else:
+            i += 1
+    return tables
+
+
+def find_table(text, required_cols):
+    """First table whose header contains every required column (contains-match)."""
+    req = [c.lower() for c in required_cols]
+    for headers, rows in parse_tables(text):
+        if all(any(r in h for h in headers) for r in req):
+            return headers, rows
+        # exact-name pass for short names like 'kpi'
+        if all(r in headers for r in req):
+            return headers, rows
+    return None, None
+
+
+def col(row, name):
+    """Fetch a cell by contains-match on the header name."""
+    name = name.lower()
+    if name in row:
+        return row[name]
+    for h, v in row.items():
+        if name in h:
+            return v
+    return None
+
+
+def is_example_row(row):
+    return any(EXAMPLE_MARKER in (v or "") for v in row.values())
+
+
+def close_enough(stated, computed, tol):
+    if stated is None or computed is None:
+        return True  # missing cells reported separately
+    if computed == 0:
+        return abs(stated) < 1e-9
+    return abs(stated - computed) / abs(computed) <= tol
+
+
+def canonical_files(root):
+    docs = sdlc_core.ai_path(root)
+    for d in CANONICAL_DIRS:
+        base = docs / d
+        if base.is_dir():
+            for p in sorted(base.rglob("*.md")):
+                yield p
+
+
+def all_md_files(root):
+    docs = sdlc_core.ai_path(root)
+    for p in sorted(docs.rglob("*.md")):
+        yield p
+
+
+# ---------------------------------------------------------------- index
+
+def build_index(root):
+    docs = sdlc_core.ai_path(root)
+    lines = [
+        "# mkt_docs INDEX (generated by mkt_check.py index -- do not edit by hand)",
+        "",
+        "| Path | Description | Status |",
+        "|---|---|---|",
+    ]
+    for p in canonical_files(root):
+        if p.name == "INDEX.md":
+            continue
+        fm, body = load_frontmatter(read_text(p))
+        desc = fm.get("description", "")
+        status = fm.get("status", "")
+        if not desc:
+            h1 = re.search(r"^#\s+(.+)$", body, re.MULTILINE)
+            desc = h1.group(1).strip() if h1 else ""
+        rel = p.relative_to(docs).as_posix()
+        lines.append(f"| {rel} | {desc} | {status} |")
+    return "\n".join(lines) + "\n"
+
+
+def cmd_index(root):
+    if sdlc_core.docs_dir() == "ai_docs":
+        return max(sdlc_core.cmd_index(root),
+                   knowledge.kb_refresh_indexes(root / "ai_docs"))
+    content = build_index(root)
+    out = sdlc_core.ai_path(root) / "INDEX.md"
+    out.write_text(content, encoding="utf-8")
+    print(f"[OK] wrote {out.relative_to(root).as_posix()}")
+    # F-028: the workstream registry is generated here too. This entry point does
+    # not reuse the spine's cmd_index, so wiring it explicitly is what keeps the
+    # doctrine ("generated by mkt_check.py index") from being false in this lens.
+    return max(sdlc_core.rc_registry(root), knowledge.kb_refresh_indexes(root / sdlc_core.docs_dir()))
+
+
+# ---------------------------------------------------------------- validate
+
+def run_validate(root):
+    rep = Report()
+    docs = sdlc_core.ai_path(root)
+    shared = sdlc_core.docs_dir() == "ai_docs"
+    # F-028: the generated registry against its HANDOFF_*.md sources. Gated on
+    # source presence, so a project that has not converted sees nothing new.
+    workstreams = sdlc_core.list_workstreams(root)
+    if workstreams:
+        hand = docs / "audit" / "handoff.md"
+        if not hand.is_file():
+            rep.error("audit/handoff.md missing while HANDOFF_*.md sources exist: "
+                      "run 'mkt_check.py index'")
+        elif sdlc_core.norm_text(read_text(hand)) !=                 sdlc_core.norm_text(sdlc_core.build_registry(root)):
+            rep.error("audit/handoff.md not aligned with its HANDOFF_*.md sources: "
+                      "run 'mkt_check.py index'. A merge resolved by hand is exactly "
+                      "what this catches")
+        if len(workstreams) > sdlc_core.REGISTRY_CAP:
+            rep.warn(f"{len(workstreams)} open workstreams: the registry is meant to "
+                     f"stay under {sdlc_core.REGISTRY_CAP}")
+    known = {p.name for p in docs.rglob("*.md")}
+    for p in canonical_files(root):
+        if p.name == "INDEX.md":
+            continue
+        rel = p.relative_to(docs).as_posix()
+        if shared and not rel.startswith(("strategy/", "tactics/", "deliverables/")):
+            continue  # core owns all vision states and shared canonical documents
+        fm, _ = load_frontmatter(read_text(p))
+        status = fm.get("status", "")
+        if not status:
+            rep.warn(f"{rel}: missing frontmatter 'status'")
+        elif status.upper() not in VALID_STATUS:
+            rep.warn(f"{rel}: invalid status '{status}' (expected {'|'.join(VALID_STATUS)})")
+        sup = fm.get("supersedes", "")
+        if sup and Path(sup).name not in known:
+            rep.warn(f"{rel}: supersedes '{sup}' which does not exist under mkt_docs/")
+    index_path = docs / "INDEX.md"
+    if shared:
+        if (index_path.is_file() and read_text(index_path).startswith(
+                "# mkt_docs INDEX (generated by mkt_check.py index")):
+            rep.error("ai_docs/INDEX.md uses the legacy marketing format: "
+                      "run 'mkt_check.py index --docs-dir ai_docs' to migrate it")
+    elif not index_path.exists():
+        rep.warn("mkt_docs/INDEX.md missing -- generate with 'mkt_check.py index'")
+    elif read_text(index_path) != build_index(root):
+        rep.warn("mkt_docs/INDEX.md stale -- regenerate with 'mkt_check.py index'")
+    return rep
+
+
+def has_marketing_engagement(root):
+    """A shared tree only owes marketing checks when it actually has marketing work."""
+    docs = sdlc_core.ai_path(root)
+    if (docs / "research/evidence_ledger.md").is_file():
+        return True
+    for dirname in ("strategy", "tactics", "deliverables"):
+        base = docs / dirname
+        if base.is_dir() and any(base.rglob("*.md")):
+            return True
+    default = sdlc_core.project_default_domain(root)
+    features = docs / "vision/features"
+    if features.is_dir():
+        for path in features.glob("*.md"):
+            meta, _ = load_frontmatter(read_text(path))
+            if sdlc_core.resolve_domain(meta, default)[0] == "marketing":
+                return True
+    return False
+
+
+# ---------------------------------------------------------------- ledger
+
+def load_ledger(root):
+    """Return (rows, report). Rows: list of dicts with id/claim/class/... keys."""
+    rep = Report()
+    path = sdlc_core.ai_path(root) / "research" / "evidence_ledger.md"
+    if not path.exists():
+        rep.error("research/evidence_ledger.md missing")
+        return [], rep
+    headers, rows = find_table(read_text(path), ["id", "claim", "class", "source"])
+    if rows is None:
+        rep.error("evidence_ledger.md: ledger table not found "
+                  "(need columns ID | Claim | Class | ... | Source | ...)")
+        return [], rep
+    entries = []
+    seen = set()
+    for row in rows:
+        rid = (col(row, "id") or "").strip()
+        if not rid:
+            continue
+        if is_example_row(row):
+            continue
+        if not re.fullmatch(r"EV-\d+", rid):
+            rep.error(f"ledger: bad ID '{rid}' (expected EV-<n>)")
+            continue
+        if rid in seen:
+            rep.error(f"ledger: duplicate ID {rid}")
+        seen.add(rid)
+        entries.append(row)
+    return entries, rep
+
+
+def run_ledger(root):
+    entries, rep = load_ledger(root)
+    ids = set()
+    for row in entries:
+        rid = col(row, "id").strip()
+        ids.add(rid)
+        cls = (col(row, "class") or "").strip().upper()
+        source = (col(row, "source") or "").strip()
+        date = (col(row, "date") or "").strip()
+        confidence = (col(row, "confidence") or "").strip().upper()
+        if cls not in LEDGER_CLASSES:
+            rep.error(f"ledger {rid}: class '{cls}' invalid (FACT|BENCHMARK|ASSUMPTION)")
+            continue
+        if cls == "BENCHMARK":
+            if "http" not in source.lower():
+                rep.error(f"ledger {rid}: BENCHMARK without a source URL")
+            if not date:
+                rep.error(f"ledger {rid}: BENCHMARK without a date")
+        if cls == "FACT" and not source:
+            rep.error(f"ledger {rid}: FACT without a source -- name the origin "
+                      "('user, Wave 1' or the client's data file)")
+        if (cls == "FACT" and source and not CLIENT_ORIGIN_RE.match(source)
+                and INTERNAL_SOURCE_RE.search(source)):
+            # FACT means "the client told me" or "the client's own primary data" — the
+            # one class that legitimately has no URL. Pointing it at another document
+            # of this engagement is how a researched observation gets in without the
+            # URL its class would owe: the row cites a file that cites the source, the
+            # validator resolves nothing, and "see VOC.md" is exactly the form the
+            # BENCHMARK rule rejects. Observed in a cold-agent field test, 2026-08-02.
+            rep.error(f"ledger {rid}: FACT sourced to an internal document ('{source}') -- "
+                      "a FACT is what the client stated or their own primary data; an "
+                      "observation gathered by research is a BENCHMARK and owes its URL")
+        if cls == "ASSUMPTION":
+            if confidence not in CONFIDENCE_VALUES:
+                rep.error(f"ledger {rid}: ASSUMPTION without confidence (HIGH|MED|LOW)")
+            value = (col(row, "value") or "").strip()
+            if value and not re.search(r"\d\s*[-–—]\s*\d|\bto\b", value):
+                rep.warn(f"ledger {rid}: ASSUMPTION value looks like a point estimate, expected a range")
+    # reference resolution across every md file (ledger file itself excluded)
+    referenced = set()
+    ledger_path = sdlc_core.ai_path(root) / "research" / "evidence_ledger.md"
+    for p in all_md_files(root):
+        if p == ledger_path:
+            continue
+        text = read_text(p)
+        for line in text.split("\n"):
+            if EXAMPLE_MARKER in line:
+                continue
+            for m in EV_REF.finditer(line):
+                referenced.add(m.group(1))
+                if m.group(1) not in ids:
+                    rel = p.relative_to(sdlc_core.ai_path(root)).as_posix()
+                    rep.error(f"{rel}: reference [{m.group(1)}] not found in the ledger")
+    for rid in sorted(ids - referenced, key=lambda x: int(x.split("-")[1])):
+        rep.warn(f"ledger {rid}: never referenced by any document")
+    return rep
+
+
+# ---------------------------------------------------------------- budget
+
+def default_tactical_files(root, file_arg):
+    if file_arg:
+        p = Path(file_arg)
+        if not p.is_absolute():
+            p = root / file_arg
+        return [p] if p.exists() else []
+    tactics = sdlc_core.ai_path(root) / "tactics"
+    files = []
+    if (tactics / "TACTICAL_PLAN.md").exists():
+        files.append(tactics / "TACTICAL_PLAN.md")
+    if tactics.is_dir():
+        files.extend(sorted(tactics.glob("CAMPAIGN_*.md")))
+    return files
+
+
+def run_budget(root, file_arg=None):
+    rep = Report()
+    files = default_tactical_files(root, file_arg)
+    if not files:
+        rep.warn("no tactical plan found (tactics/TACTICAL_PLAN.md) -- budget check skipped")
+        return rep
+    for path in files:
+        rel = path.name
+        text = read_text(path)
+        m = re.search(r"^Total budget:\s*(.+)$", text, re.MULTILINE | re.IGNORECASE)
+        if not m:
+            rep.error(f"{rel}: 'Total budget:' line missing")
+            continue
+        total = parse_num(m.group(1))
+        if total is None or total <= 0:
+            rep.error(f"{rel}: cannot parse Total budget '{m.group(1).strip()}'")
+            continue
+        headers, rows = find_table(text, ["channel", "budget", "share"])
+        if rows is None:
+            rep.error(f"{rel}: Budget Allocation table not found (Channel | Budget | Share)")
+            continue
+        rows = [r for r in rows if not is_example_row(r)]
+        if not rows:
+            rep.error(f"{rel}: Budget Allocation table has no rows")
+            continue
+        alloc = 0.0
+        for r in rows:
+            v = parse_num(col(r, "budget"))
+            if v is None:
+                rep.error(f"{rel}: unparseable budget cell '{col(r, 'budget')}' "
+                          f"for channel '{col(r, 'channel')}'")
+            else:
+                alloc += v
+        if not close_enough(alloc, total, BUDGET_TOLERANCE):
+            rep.error(f"{rel}: allocations sum to {alloc:g}, Total budget is {total:g} "
+                      f"(tolerance 1%)")
+    return rep
+
+
+# ---------------------------------------------------------------- funnel
+
+FUNNEL_CHAIN = (
+    # (stated_col, compute, label)
+    ("clicks", lambda b, cpc, cvr, close, r: b / cpc if cpc else None, "Budget/CPC"),
+    ("leads", lambda b, cpc, cvr, close, r: (r["clicks"] * cvr / 100.0)
+        if (r["clicks"] is not None and cvr is not None) else None, "Clicks*CVR%"),
+    ("customers", lambda b, cpc, cvr, close, r: (r["leads"] * close / 100.0)
+        if (r["leads"] is not None and close is not None) else None, "Leads*Close%"),
+    ("cac", lambda b, cpc, cvr, close, r: (b / r["customers"])
+        if (r["customers"] not in (None, 0) and b is not None) else None, "Budget/Customers"),
+)
+
+
+def run_funnel(root, file_arg=None):
+    rep = Report()
+    files = default_tactical_files(root, file_arg)
+    if not files:
+        rep.warn("no tactical plan found -- funnel check skipped")
+        return rep
+    for path in files:
+        rel = path.name
+        text = read_text(path)
+        headers, rows = find_table(text, ["channel", "budget", "cpc", "cvr", "customers", "cac"])
+        if rows is None:
+            rep.error(f"{rel}: Funnel Model table not found "
+                      "(Channel | Budget | CPC | Clicks | CVR % | Leads | Close % | Customers | CAC)")
+            continue
+        rows = [r for r in rows if not is_example_row(r)]
+        if not rows:
+            rep.error(f"{rel}: Funnel Model table has no rows")
+            continue
+        for r in rows:
+            channel = col(r, "channel") or "?"
+            budget = parse_num(col(r, "budget"))
+            cpc = parse_num(col(r, "cpc"))
+            cvr = parse_num(col(r, "cvr"))
+            close = parse_num(col(r, "close"))
+            stated = {
+                "clicks": parse_num(col(r, "clicks")),
+                "leads": parse_num(col(r, "leads")),
+                "customers": parse_num(col(r, "customers")),
+                "cac": parse_num(col(r, "cac")),
+            }
+            if budget is None or cpc is None:
+                rep.error(f"{rel} [{channel}]: Budget/CPC missing or unparseable")
+                continue
+            # verify each stated cell against computation from STATED upstream
+            # cells (so a single wrong cell does not cascade into four errors)
+            chain_inputs = dict(stated)
+            for name, compute, label in FUNNEL_CHAIN:
+                computed = compute(budget, cpc, cvr, close, chain_inputs)
+                if stated[name] is None:
+                    rep.error(f"{rel} [{channel}]: '{name}' cell missing")
+                    continue
+                if computed is None:
+                    continue  # upstream missing, already reported
+                if not close_enough(stated[name], computed, FUNNEL_TOLERANCE):
+                    rep.error(f"{rel} [{channel}]: {name}={stated[name]:g} but {label} "
+                              f"= {computed:g} (tolerance 5%)")
+    return rep
+
+
+# ---------------------------------------------------------------- trace
+
+def run_trace(root):
+    rep = Report()
+    docs = sdlc_core.ai_path(root)
+    obj_path = docs / "strategy" / "OBJECTIVES.md"
+    if not obj_path.exists():
+        rep.warn("strategy/OBJECTIVES.md missing -- trace check skipped")
+        return rep
+    objectives = set(OBJECTIVE_HEADING.findall(read_text(obj_path)))
+    if not objectives:
+        rep.error("OBJECTIVES.md: no '### O<n> --...' headings found")
+        return rep
+
+    tactical = docs / "tactics" / "TACTICAL_PLAN.md"
+    served = set()
+    if tactical.exists():
+        headers, rows = find_table(read_text(tactical), ["channel", "objective", "kpi"])
+        if rows is None:
+            rep.error("TACTICAL_PLAN.md: Channel Plan table not found "
+                      "(Channel | Objective | KPI | Budget | Owner)")
+        else:
+            for r in rows:
+                if is_example_row(r):
+                    continue
+                tokens = set(OBJECTIVE_TOKEN.findall(col(r, "objective") or ""))
+                if not tokens:
+                    rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: no objective (O<n>) named")
+                for t in tokens:
+                    if t not in objectives:
+                        rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: objective {t} "
+                                  "not defined in OBJECTIVES.md")
+                served |= tokens & objectives
+                if not (col(r, "kpi") or "").strip():
+                    rep.error(f"TACTICAL_PLAN.md [{col(r, 'channel')}]: KPI cell empty")
+        for o in sorted(objectives - served):
+            rep.warn(f"objective {o}: no tactic serves it in TACTICAL_PLAN.md")
+    else:
+        rep.warn("tactics/TACTICAL_PLAN.md missing -- tactic trace skipped")
+
+    measure = docs / "tactics" / "MEASUREMENT_PLAN.md"
+    if measure.exists():
+        headers, rows = find_table(read_text(measure), ["objective", "kpi", "target"])
+        if rows is None:
+            rep.error("MEASUREMENT_PLAN.md: KPI table not found (Objective | KPI | Target | ...)")
+        else:
+            measured = set()
+            for r in rows:
+                if is_example_row(r):
+                    continue
+                measured |= set(OBJECTIVE_TOKEN.findall(col(r, "objective") or ""))
+            for o in sorted(objectives - measured):
+                rep.error(f"objective {o}: no KPI row in MEASUREMENT_PLAN.md")
+    else:
+        rep.warn("tactics/MEASUREMENT_PLAN.md missing -- KPI trace skipped")
+    return rep
+
+
+# ---------------------------------------------------------------- check
+
+def run_check(root):
+    rep = Report()
+    checks = [("validate", run_validate)]
+    if sdlc_core.docs_dir() != "ai_docs" or has_marketing_engagement(root):
+        checks += [("ledger", run_ledger), ("budget", run_budget),
+                   ("funnel", run_funnel), ("trace", run_trace)]
+    for title, fn in checks:
+        sub = fn(root)
+        sub.print(title)
+        rep.merge(sub)
+    return rep
+
+
+# ---------------------------------------------------------------- main
+
+# --- portable checks this distribution exposes -------------------------------
+# The marketing arithmetic, offered to documents owned by ANOTHER domain: a
+# distilled spec that carries a budget table can import `marketing.budget` and have
+# it actually checked, instead of the table sitting in a document whose own
+# validator has no opinion about it (UC8). Imported checks may only ADD findings.
+
+def _report_findings(rep):
+    return ([("error", m) for m in rep.errors]
+            + [("warning", m) for m in rep.warnings])
+
+
+@sdlc_core.portable_check("marketing.ledger")
+def _check_ledger(rel, meta, text):
+    root = sdlc_core.find_project_root()
+    return _report_findings(run_ledger(root)) if root else []
+
+
+@sdlc_core.portable_check("marketing.budget")
+def _check_budget(rel, meta, text):
+    root = sdlc_core.find_project_root()
+    return _report_findings(run_budget(root)) if root else []
+
+
+@sdlc_core.portable_check("marketing.funnel")
+def _check_funnel(rel, meta, text):
+    root = sdlc_core.find_project_root()
+    return _report_findings(run_funnel(root)) if root else []
+
+
+def main(argv=None):
+    argv = list(sys.argv[1:] if argv is None else argv)
+    # Spine commands are the core's, identical in every distribution: hand over
+    # every subcommand this overlay does not explicitly intercept, untouched,
+    # rather than keeping a hand-copied list of spine names here -- which is how
+    # `migrate` got dropped while SKILL.md kept promising it.
+    if argv and not argv[0].startswith("-") and argv[0] not in OVERLAY_COMMANDS:
+        return knowledge.main(argv)
+
+    parser = argparse.ArgumentParser(prog="mkt_check.py", description=__doc__,
+        epilog="Shared memory commands: recall QUERY, graph, corpus, claim-id, anchor, "
+               "export, import, stale, orient. Use COMMAND --help for options.")
+    parser.add_argument("command", choices=list(OVERLAY_COMMANDS))
+    parser.add_argument("file", nargs="?", default=None,
+                        help="optional target file for budget/funnel")
+    parser.add_argument("--root", default=None, help="project root (contains the docs root)")
+    parser.add_argument("--docs-dir", dest="docs_dir", default=None,
+                        help="name of the documentation root (default: mkt_docs; "
+                             "use --docs-dir ai_docs on a project that has migrated "
+                             "to the family's single tree)")
+    parser.add_argument("--strict", action="store_true",
+                        help="warnings also fail the exit code")
+    args = parser.parse_args(argv)
+
+    # Resolve the docs root fresh on EVERY invocation, from this distribution's own
+    # default: the core keeps it in module state, and an entry point that inherited
+    # whatever a previous call left behind would be answering the wrong question.
+    try:
+        discovered, name = sdlc_core.resolve_docs_dir(args, args.root)
+    except sdlc_core.AmbiguousDocsRoot as exc:
+        print("[ERROR] %s" % exc)
+        return 1
+    root = find_root(args.root or discovered, name)
+    if root is None:
+        print(f"[ERROR] no {sdlc_core.docs_dir()}/ directory found "
+              "(use --root or run mkt-sdlc-init)")
+        return 2
+
+    if args.command == "index":
+        return cmd_index(root)
+
+    runners = {
+        "validate": lambda: run_validate(root),
+        "ledger": lambda: run_ledger(root),
+        "budget": lambda: run_budget(root, args.file),
+        "funnel": lambda: run_funnel(root, args.file),
+        "trace": lambda: run_trace(root),
+    }
+    if args.command == "check":
+        rep = run_check(root)
+    else:
+        rep = runners[args.command]()
+        rep.print(args.command)
+
+    code = rep.exit_code(strict=args.strict)
+    if args.command in ("validate", "check"):
+        if sdlc_core.docs_dir() == "ai_docs":
+            code = max(code, sdlc_core.cmd_validate(root, strict=args.strict))
+        code = max(code, knowledge.kb_validate_surface(root / sdlc_core.docs_dir(),
+                                                       full=args.command == "check"))
+    if code == 0:
+        print(f"[OK] {args.command} clean "
+              f"({len(rep.warnings)} warning(s))" if rep.warnings else f"[OK] {args.command} clean")
+    else:
+        print(f"[FAIL] {args.command}: {len(rep.errors)} error(s), {len(rep.warnings)} warning(s)")
+    if args.command == "check":
+        # This overlay REPLACES the spine's cmd_check, so the F-041 wiring note
+        # must be re-attached here: it is check-layer behaviour, whoever owns
+        # the check (fail-open and informational -- never the exit code).
+        sdlc_core.print_orient_hook_note(root)
+    return code
+
+
+if __name__ == "__main__":
+    sys.exit(main())