Lifecycle Audit — Portfolio Review

Score an existing set of journeys against the same rules the engine uses to generate them. Output: an Audit Report with per-journey findings and a portfolio-level verdict.

When NOT to use this

• No journeys exist anywhere yet — nothing described, generated, or imported — there's nothing to score; that's a lifecycle-connect → lifecycle-journeys conversation instead.
• The user wants the findings acted on, not just diagnosed — this skill never rewrites journeys itself (see Never do below); generation is a separate, offered step via lifecycle-journeys.
• Real performance or holdout data exists and the ask is whether a journey is actually working — that's lifecycle-results, which judges measured incremental outcomes; this skill judges structure and methodology without live results.

Inputs

Journeys as: prior engine output, user description ("we have a welcome email and a cart reminder"), or CRM exports/screenshots. Sparse descriptions are fine — audit what is known, list what could not be assessed. If available, also use the DQS + stage map; without them, stage-coverage and depth-vs-data-fit findings are marked "data-blind" — dimension 3 specifically cannot be judged without knowing what the data supports, so guessing "over-engineered" or "under-leveraged" from journey shape alone is a fabrication risk (CLAUDE.md rule 3), not a finding.

Audit dimensions (score each 0–5, with evidence)

Dimension 4 also covers sector-benchmark fit. When a running pattern has sector-specific timing guidance in knowledge/industries/<sector>.md ("Sector-specific timing & cadence"), compare the journey's actual/stated trigger timing and channel escalation order against that guidance and cite the specific range verbatim (e.g. "fires at 48h against the sector's 1–4h cart-abandonment window") — never an invented industry average (CLAUDE.md rule 3: ranges from knowledge files may be cited, nothing may be fabricated). A deviation is a finding scaled to its degree, not an automatic Critical — existing-but-slow still beats missing (dimension 2), and severity comes from revenue/compliance impact, not distance from the benchmark alone.

Scoring posture (anti-inflation)

Score as an adversarial reviewer of someone else's work, not as the author: if you cannot name a concrete gap for a dimension, you have not looked hard enough to score it high. When torn between two adjacent scores, take the lower. Mostly-top scores on a first pass are a signal to re-examine the scoring, not evidence of quality. Never reverse-engineer scores from a desired overall verdict.

Two failure modes get confused under this posture and must be told apart: no evidence provided (the description simply didn't cover it — mark "not assessable," exclude it from the average, but count it in coverage) is not the same as evidence of absence (the journey was described and plainly has no exit criteria, no holdout, etc. — that scores low). Defaulting ambiguous cases to "not assessable" rather than a guessed-low score keeps the adversarial posture from tipping into fabrication.

When the verdict gates a real decision (budget, headcount, killing a program), treat one pass as provisional: a second independent pass (fresh session, or a different model) that disagrees by ≥ 2 points on any dimension is the same "two judgments to gate" discipline evals/rubric.md already applies to copy and journey judging — one adversarial pass catches more than no review, but it is still one reviewer's read.

Report format

1. Verdict — one paragraph + overall score (avg of the scored dimensions, /5) + evidence coverage stated inline, the same way the DQS reports its flags in the score line itself rather than burying them — 6/8 dimensions scored, 2 not assessable (measurement, copy compliance — no config visibility) — never a bare average silently computed over however many dimensions happened to be visible. A verdict built on fewer than half the dimensions says so in the paragraph itself, not just in the coverage tag.
2. Findings table — severity (Critical/High/Medium/Low) | journey | finding | fix. Critical = revenue leaking or compliance risk.
3. Coverage map — same stage table as the portfolio template §3.
4. Recommended actions — ordered; each one maps to a concrete next step (run /lifecycle journeys for the missing P0s, re-run copy for flow X, and when the user has performance data: run /lifecycle results to score these against holdouts).
5. Audit trail — when a brand config exists, check output/<brand>/audit-history.md (create it if absent, same convention as results-log.md): a Critical/High finding matching one from the prior run's log (same journey + same dimension) is flagged repeat finding, open since <date> instead of reported as new; a prior Critical/High that doesn't reappear gets one line resolved since last audit. Append this run's Critical/High findings before closing out. No brand config (ad hoc/anonymous audit) → skip this step and note once that findings aren't being tracked across runs.

Common Pitfalls

Pitfall 1: "Not assessable" used to dodge a hard call. Symptom: several dimensions come back "not assessable" on a review where the user actually described the journeys in reasonable detail — the ambiguity is judgment-shaped, not evidence-shaped. Consequence: the audit is technically defensible (nothing was fabricated) but useless (nothing was decided) — this is the exact gap the "no evidence provided vs evidence of absence" distinction above exists to close, restated as a self-check on the finished report. Fix: before marking a dimension "not assessable," confirm it is truly unstated rather than stated-but-bad — a journey with a described exit that just happens to be weak scores low, it doesn't get exempted.

Pitfall 2: Severity flattens to Medium. Symptom: the findings table reads Medium down the entire severity column. Consequence: a compliance-class issue and a wording nitpick become indistinguishable, and the one output an audit exists to produce — what to fix first — disappears. Fix: apply "Critical = revenue leaking or compliance risk" literally per finding, not as a description of the audit's overall tone; most audits should have a mix, and an all-Medium table is a signal to re-examine severity, the same way an all-high-score pass is a signal to re-examine scores.

Pitfall 3: Score creep without a named gap. Symptom: a dimension lands at 4/5 or 5/5 with a one-line justification that doesn't point at anything specific the journey is missing. Consequence: this is the fabrication risk the scoring-posture section already names — a high score with no gap attached is indistinguishable from a score that was never actually checked. Fix: if the write-up can't name a concrete gap, the score is not yet earned — drop to the next tier down or go back and look again, per the "when torn, take the lower" rule already stated above.

Never do

• Never score dimensions without stating the evidence ("no exit criteria mentioned for 3 of 5 journeys").
• Never fail a journey for missing information the user didn't provide — mark "not assessable" instead.
• Never rewrite the user's journeys inside the audit — the audit diagnoses; generation is a separate, offered step.