“Verified” should be checkable before you pay

The evidence ledger — a real excerpt

This is an unedited-in-substance sample from the map's internal evidence ledger, for one chapter — Data, Measurement & Analytics (the same chapter you can read free in full, so you can check the ledger against the finished text). Every entry carries its source, source tier, capture date, volatility rating — and, where it matters, a bias flag: most marketing sources are vendors selling the thing they describe, so direction and magnitude are trusted differently.

The chapter's load-bearing rule, and what it stands on

• "Incrementality is the arbiter" — platform-reported ad effects are systematically overstated vs. randomized ground truth; real lift is often statistically unmeasurable at realistic budgets. Sources: Lewis & Rao, Quarterly Journal of Economics 2015; Gordon, Zettelmeyer, Bhargava & Chapsky, Marketing Science 2019; "Ghost Ads," Journal of Marketing Research 2017 — tier 1–2 (peer-reviewed academic) · captured Jun 2026 · volatility: low. Flag: deliberately anchored to academic literature, not the measurement vendors who profit from the rule.
• Vendor over-attribution magnitudes (platforms over-credit ~20–60%; a reported 3× ROAS ≈ ~1.5× incremental) — incrementality-vendor books of business — tier 3–4 (vendors) · Jun 2026 · volatility: med. Flag: the direction triangulates across competitors; the magnitudes are marketing — never used as load-bearing point estimates.
• IAB / IAB Europe Commerce Media Measurement Standards V2 (Jan 2026; grace period to end-Jul 2026) + Incremental Measurement Guidelines — tier 1–2 (standards body) · Jun 2026 · med. The standards-body codification of the arbiter rule — the map leans on this and the academic literature, not vendor decks.

Dated platform facts (the kind that rot — so they're dated)

• Google shut down the Privacy Sandbox 2025-10-17 — near-total retirement (Topics, PAAPI, Attribution Reporting API and more; kept CHIPS/FedCM/Private State Tokens); UK CMA released Google the same day — Google primary + 4 independent trades · tier 1–3 · Jun 2026 · low-med.
• Third-party cookies remain default-on in Chrome — no removal timeline, no user prompt (the Apr 2025 decision) — Google primary + privacy trades · tier 1–3 · Jun 2026 · low-med.
• Apple ATT, stated carefully: opt-in ≈ 25% of all users (≈ 35–46% of prompted users); ≈ 75% not trackable at the identifier level — mobile-measurement vendors · tier 3–4 · Jun 2026 · med. Flag: the denominator is load-bearing; sources quoting "65–75% cut off" without the denominator were corrected.
• GA4's April 2026 restructure: first-click and rules-based models removed — data-driven attribution is effectively the model (400+ conversions; 30/90-day lookbacks) — search trades · tier 2–3 · Jun 2026 · med-high.
• Google Consent Mode v2 is now a hard requirement for GA4 → Google Ads in the EEA/UK; conversion modeling needs ≥700 ad clicks/7 days/country — Google primary + CMP vendors · tier 1–3 · Jun 2026 · med.
• MMM tooling: Google Meridian (released 2025-01-29) + PyMC-Marketing are the durable open-source pair; Meta's Robyn is discontinued upstream (legacy/maintenance) — Forrester + trades + the repos · tier 1–3 · Jun 2026 · med. Flag: open-source MMM has a counter-pattern — trust the method, distrust in-tool channel ROI (Meridian flatters Google channels; Robyn flattered Meta's).

The agent-era numbers (stated with their limits)

• ~70.6% of AI-assistant traffic arrives referrerless and lands misclassified as "Direct" (446,405-visit dataset); 89% of brands can't attribute AI referral traffic (Conductor, Nov 2025) — tier 3–4 · Jun 2026 · med-high. There is no clean denominator for "agent share" with current instrumentation — so the map refuses to invent one.
• GA4's native "AI Assistant" channel (launched 2026-05-13) is a floor, not a measure — Perplexity lands in Referral, AI Overviews in Organic, ~70% in Direct — search trades + practitioners · tier 3 · Jun 2026 · high.
• Adobe's quarterly AI traffic reporting: Q1-2026 AI-sourced US-retail traffic +393% YoY; AI-referred visitors convert +42% vs. average (reversed from −38% a year earlier) — Adobe primary + tier-2 trade · Jun 2026 · high. Flag: Adobe sells commerce tooling — methodology-disclosed and preferred over less transparent peers, but still self-interested.
• Salesforce's "AI influenced ~20% of global online orders" — held firewalled from any agent-completed metric: "influenced" and "completed by an agent" are different claims, and only the second would trigger the map's structural re-test — Salesforce primary · tier 2 · Jun 2026 · high.
• 2030 agent-commerce forecasts span JPMorgan ~25% · Bain 15–25% · Gartner ~20% · Morgan Stanley 10–20% of US online sales — tier 2–4 · Jun 2026 · high. The 2.5× definition-driven spread IS the finding; the map never quotes one number as "the" forecast.

Corrections we logged against our own earlier entries

The ledger keeps its own mistakes visible — a claim isn't "verified" if the verification can't lose:

• Privacy Sandbox: our earlier entry said "several APIs retired, April 2026" → corrected to 2025-10-17, near-total shutdown once primary sources were re-checked.
• Server-side tracking: vendor claims of "95–99% match rates" → corrected to a credible ~85–90%, with recovery ranges kept directional (they come from the sellers).
• ATT: "65–75% of users cut off" → restated with the denominator that makes it true or false.
• MMM tools: "Robyn — co-equal, longest track record" → discontinued upstream; corrected against the repository itself.

The full ledger runs ~600 entries across all 16 chapters and is re-verified on a dated cadence (the stamp in the footer is real). This page shows one chapter's worth so you can check the "verified" claim before any money moves.