• ads (2.3.2 → 2.4.0): new references/reading-google-ads-data.md, covering how Google Ads data misleads before an audit even starts. Search terms are a sample: on small accounts 50–65% of clicks have no disclosed query, so every claim states disclosed vs total clicks, the report is pulled unfiltered first, and nothing (GA4, CRM) recovers the withheld queries. Data traps: conversions vs all_conversions, several final URLs per ad group, click-date attribution, expired experiment arms that look like duplicate campaigns, the 30-day change-history limit, and client_type for who else changes the account. Window vs lifetime and no pooling across config changes. Small numbers: a zero-conversions-in-N-clicks table (what each sample size rules out) plus break-even CVR = CPC ÷ target CPA, and "let it run" only with a date. Stating findings as verified / inferred / stale, the "can the client disprove this in a minute?" test, and a "conclusions that sound right" table. Fixed the audit-guardrails.md Fail example that treated disclosed search terms as a share of all spend. Two new evals (ids 12–13). New triggers: 'search terms report,' 'wasted spend,' 'is this campaign working.'
• tools/google-ads: API URLs moved from the long-sunset v14 to v24 (guide and CLI); added analysis GAQL recipes (campaign inventory, monthly since launch, conversions by action, unfiltered search terms, ads by final URL, change history) and a gotchas table. Closes #601.