Daily Audit Reports

Performance snapshots from tools/performance_snapshot.py. Current report shows active issues — mark resolved to hide. Older reports are archived below.

Wave 0.12 — Account-Level Negatives Scope Review

10 sub-waves proposed from 180-day 0-conv search-term mining (£2,715 waste across 500 terms). Empty account-level list 10439392410 is the mutation target. Review each sub-wave, approve/modify/reject, and add notes. Decisions auto-save to browser. Claude reads this state before executing the batch mutation.

Compliance Q1–Q4 — Wave 1 Unpause Blockers

Four risk-appetite decisions blocking Wave 1 campaign unpauses across UK, FR, and DE. Each question is tied to a browser-verified landing-page compliance issue from the 6 April audit. Answering all four unblocks ~11 PAUSED Search + PMax campaigns for unpause.

Go Live 1 — Urgent Decisions to Move Forward

Live, in-play issues blocking Phase 1 launch. Each subsequent Go Live N tab will hold the next set of pending blockers. Answer the dropdowns + add notes — auto-saves to browser. Export JSON when done and paste back in chat.

⚠ POLICY UPDATE 2026-04-25 — Shopify-only LPs

All landing pages ship on Shopify going forward. GitHub Pages LPs (apps.nutritionalproducts.org/landing-pages/*) are DEPRECATED as legacy reference. No new GitHub Pages LPs. Memory ref: feedback_lp_deployment_policy_shopify_only.md.

Phase 1 launch order: 1) UK ADHD blog FIRST (Session 38 ports, ~95% ready). 2) Existing ENABLED brand catchers ranked by 90d revenue: PMax DE £29.9k → PMax UK £21.6k → PMax FR £10.3k → PMax US £3.7k → SERP brand cluster. 3) Wave 1 Quick Wins. 4) Wave 2 GABA/Dopamine (single-LP launch, A/B deferred). 5) DG last.

Critical Findings (2026-04-25)

• Session 38 ADHD Shopify port ~95% ready — 7 blog variants (A–F final + base) published 2026-04-23/24. Going live very soon. First campaign to launch. Default variant B = f-final.
• Session 21 image template library final — scaled FLUX gen ready (heroes, illustrations, wides, blog).
• Session 25 marcom matrix complete — scaled copy gen with lensing, matrix, lexicon, localisation.
• The 8 expected GABA/Dopamine educational Shopify LPs were never built. What was built (2026-04-20→24) is 6 alternate-URL product pages: UK /products/buy-gaba-supplements + /products/buy-dopamine-supplements; US /products/gaba-supplements + /products/dopamine-supplements; DE /products/gaba-supplement + /products/dopamin-supplement. Same ELITE/PRO products, different SEO handles. These are Phase 1 launch destinations; Shopify blog versions go in build queue.
• FR has zero GABA/Dopamine content of any kind. No product handle, no Shopify blog. FR is the laggard.
• GitHub Pages educational LPs deprecated — precursor-education.html, dopamine-guide.html, pre-port adhd-supplements-guide.html. Legacy reference only. Do not route new ad traffic.
• Day-Zero baseline captured 2026-04-24: 90d account totals £80,258 spend / 3,080 conv / £26.06 blended CPA. PMax = 82% of spend at £26.87 CPA. Search at £8.71 CPA most efficient.
• UTM schema v2 shipped 2026-04-24 — SoT at v4.0/utm-url-rules.md. Per-campaign register in sheet tab "2026 Launch UTMs", 38 rows.
• 2 DE Ritalin PMax campaigns ENABLED at £0.01 budget, zero spend in 90d. Effectively dead. In Phase 1 scope. Decision needed.
• store_urls.json refreshed 2026-04-24 19:41:54. UK 2,858 / US 123 / DE 142 / FR 119.

Go Live 1.1 — Action Blocks (rearranged for speed)

Reframed from Go Live 1 (decision tracker) into blocks of work that can be done quickly, ordered by speed-to-launch. Each block lists the action, who's blocked, and the step-by-step. Calum's directive 2026-04-27: ADHD ad campaign go-live + refresh of existing PMax brand campaigns are top priority because they're easy wins. Built from v4.0/quiz-rebuild/MORNING-BRIEF-2026-04-27.md + PENDING-TASKS-REGISTRY-2026-04-27.md + Opus 4.7 pre-launch review.

Layering note: Go Live 1 (left of this tab) remains the canonical decision-tracker for A–K + B2/C2. This tab is the execution view — sequence + steps, not options.

Block 1 — Quick Wins (do FIRST · ~1–2 hours total)

These two unlock the highest immediate revenue with the least effort. Both are mechanically ready — only Calum sign-off is needed before Claude executes.

Block 2 — Calum-blocked approvals (15–30 min each · can be done in any order)

Each item below needs only Calum's call. Once approved, Claude executes immediately. Do these in parallel with Block 1.

Block 3 — Hard blockers before any market unpause (~45 min total)

From the Opus 4.7 pre-launch review (banked in session 32 file). These don't block Block 1.1 ADHD swap or Block 1.2 brand stamping (those are edits to already-running campaigns). They DO block any new market unpause + DG launch.

Block 4 — Post-publish QC (after each Block 2.2 publish · ~10 min per article)

Block 5 — Build queue (parallel work · can start any time)

These don't block Phase 1 launch but turn single-LP launches into proper A/B tests + extend coverage to deferred lanes/markets.

• P8 — Engine v1.1: DEV-03 processing micro-question + DEV-04 editorial modules below quiz outcome.
• P9 — Spec calibration: ORIGINAL routing reachability in Variant A, hormones threshold tune (35% → 30%), per-variant DEV-01 recalibration.
• P10 — Phase 2 localisation: US/DE/FR clones of winning UK quiz variant (after ~3 weeks of A/B/C data).
• P11 — Phase 3 quiz lanes: Peri symptom checker, Dopamine/GABA recommender, Student/Worker focus, US-only ADHD quiz.
• P12 — Phase 4 collection-vs-quiz A/B: per lane (GABA UK, Dopamine UK, Brain Fog UK, Peri UK).
• P13 — Lane format build queue: US ADHD aggregating collection, US GABA/Dopamine FDA blogs, DE Konzentration/Gut-Hirn/GABA/Dopamine, FR Anti-Fatigue Mentale + Ritalin TDAH compliance review.
• K (Go Live 1): UK GABA + UK Dopamine Shopify blog LP build (mirrors ADHD pattern, fastest path to A/B in highest-volume market).

Block 6 — Infrastructure (parallel · not launch-gating)

• P14 — Python relay to Coolify VPS: spec at v4.0/quiz-rebuild/python-relay-spec.md; FastAPI service at relay.nutritionalproducts.org/quiz/submit; replaces console.log + localStorage queue.
• P15 — GoHighLevel workflows: inbound webhook + tag schema + market-specific email nurture sequences.
• P16 — A/B test registry entries: TEST-UK-BRAINFOG-QUIZ-{A,B,C}, TEST-UK-GABA-{ALT,BLOG}, TEST-UK-DOPAMINE-{ALT,BLOG} in v4.0/agent/ab-tests.json.
• P17 — A/B traffic split mechanism: random JS redirect 33/33/33 with localStorage persistence on the canonical entry article, fanning out to A/B/C handles.
• P18 — Dashboard filter dropdown: add new recommendation_v2.choice values to campaign-lp-strategy Overview tab filter (line ~99 in index.html).
• P19 — Go Live 2 tab trigger: when next batch of launch decisions arrives (post-Phase-1 launch data, Phase 2 localisation kick-off).
• schedules.json auto-sync hook (S26 #1): dashboard-freshness only.

Block 7 — Older session 32 residual (P23 · not launch-gating)

• 2 DE Ritalin PMax campaigns ENABLED at £0.01 budget (Decision D in Go Live 1).
• Wave 8.4 / 8.5 / 8.6 (tROAS legacy brand PMax / audience signal refresh / product feed QA).
• Wave 0.13 / 8.8 campaign-specific negatives.
• Search theme expansion (40–50 themes/AG on UK Peri / UK GABA / US GABA / US Brain Fog).
• 12 legacy DG campaigns bleeding £2,937/mo at 0.21x ROAS — separate retire decision.
• smarter-brain LP: zero paid traffic but still indexable. Recommend noindex or rewrite within 2 weeks of launch.

Suggested order of operations for today

1. Morning (Calum): approve Block 1.1 ADHD swap + Block 1.2 brand stamping. Open the 5 article previews from Block 2.2. Triage Interact 503 (Block 2.1).
2. Morning (Claude): execute Block 1.1 (~10 min) + Block 1.2 (~25 min) once approved. Run Block 3.1 H4 test purchases (~30 min, can be done in parallel by Calum).
3. Midday (Calum): approve Dopamine AMBER (2.3) + 85% claim decision (2.4) + illustration budget (2.5).
4. Midday (Claude): publish each approved article (Block 2.2). Trigger Block 4.1 QC after each.
5. Afternoon (Calum): 15 min in Google Ads UI for Block 3.2 DG purchaser exclusions.
6. Afternoon (Claude): kick off Block 5 build queue (UK GABA + Dopamine blog ports first, mirroring ADHD pattern).
7. Evening: Block 6 infrastructure work begins in background.

Intel — What the Data Says Right Now

Intel is the current-state layer of the agent's three-layer operating model. It synthesises the three data rails — Google Ads, GA4, and Shopify — into a single reconciled view of account health. Because no single platform is the source of truth, Intel exists to resolve discrepancies and surface the real numbers before any tactical or strategic decision is made. Sources: memory/ref-tracking-triage.md, memory/CANONICAL_intel_sop.md, memory/feedback_ga4_canonical.md. Latest dataset: v4.0/shopify-build/v1-firstsync/.

Where Intel Lives

Triage SoT: memory/ref-tracking-triage.md — 3-rail source-of-truth ownership table, empirical reconciliation tolerances, 7 canned triage queries, 4 known anomalies.
Intel SOP: memory/CANONICAL_intel_sop.md — step-by-step protocol for producing a reconciled snapshot.
GA4 canonical: memory/feedback_ga4_canonical.md — OAuth Desktop credential, 4 GA4 property IDs, conversion-vs-purchase inflation rule, legacy property exclusion.
Shopify datasets: v4.0/shopify-build/v1-firstsync/ — raw order exports used for reconciliation.
BNS admin token: tools/bns_sheets.py — use bns-token.json for all GA4 and Sheets reads (not legacy brainzyme@ credential).

Tactics — This-Quarter Plays & Kill/Scale Calls

Tactics is the short-horizon layer: in-flight tests, immediate optimisation actions, kill/scale decisions, and Calum-action items that need to be resolved within the current quarter. Currently sourced from session files 32 and 38B. This section will be promoted to a canonical home in an upcoming agent-playbook session — for now it is the holding ground for open items surfaced by the Intel layer.

Strategy — Growth Thesis & Channel-Mix Doctrine

Strategy is the long-horizon layer: the brand's growth thesis, market prioritisation, channel-mix doctrine, and year-ahead vision. Strategy memories are currently fragmented across BUILD_STRATEGY.md, session 26 (playbook), session 28 (blueprint), and session 30 (competitor review). Consolidation is queued for an upcoming agent-playbook session — this tab is the canonical destination once that work is done.

Second-Pass QC REVIEWER: Opus 4.7

Single-shot audit of Session 38B Cross-Rail Triage syntheses (UK / US / DE / FR 25-month + cross-store 6yr Shopify intel). Cross-checked against raw historical-{6yr,2yr}-* JSON datasets via Python helpers (no line-by-line file reads). Full report: v4.0/shopify-build/v1-firstsync/SECOND-PASS-QC-2026-04-26.md.

Verdict: ACCEPT WITH CORRECTIONS — All numbers in the 5 syntheses are defensible against raw data. Two headline framings hide methodology choices that change reader interpretation. Two HIGH-severity items patched into canonical memory; original syntheses left unchanged.

Cross-store 6yr §8 reports refund rate as refund_count / orders ("UK refund rate doubled from 2024 1.02% → 2025 2.78% → 2026 YTD 3.67%"). UK per-market §5 reports it as refund_amount / total_revenue (UK 2025 = 0.98%, flat against 5-year 0.93%–1.27% band). Same term, different metric, different magnitude.

Reading: UK customers file refund events at 2.5–3× the historical rate, but the £ value is flat. Pattern is consistent with line-item partial refunds (single units inside multi-unit orders), not full-order returns. The "refund rate doubled" framing reads as a returns crisis; £ impact is unchanged.

Canonical wording (recommended): "UK refund-event rate has risen from ~1.0% to 2.78–3.67% on a count basis, while refund value as a share of revenue stays at the historical ~1% band — consistent with rising line-item partial refunds."

memory/ref-tracking-triage.md §2a quotes DE empirical GA4↔Shopify dropout at −32.4% ±4.0% alongside UK / US / FR. The DE synthesis correctly discloses this excludes 4 outlier months (1× Apr 2024 partial launch + 3× Feb–Apr 2026 Key Event inflation window). The full unfiltered 25-month DE series is −26.3% ±26.3% — a 6.5× wider variance band.

Risk: Quoting −32.4% ±4.0% without the curation note makes DE look 3-rail-clean; an ops reader using the table to set alert thresholds will calibrate too tight and trigger false positives.

Fix applied: ref-tracking-triage.md §2a inline marker added ("21-month curated subset… full 25-month is −26.3% ±26.3%. Per Second-Pass QC 2026-04-26."). Other 3 markets do not need the same caveat — UK/US/FR figures use the full window or only exclude a single startup month explicitly called out.

MED-1: DE Apr 2026 refund line (426 events / €347 amount / €0.81 avg) is correctly flagged in DE synthesis as a Shopify line-item refund artifact. The cross-store 6yr file silently rolls those events into DE 2026 totals without the asterisk — flag only, no rewrite.

LOW-1: DE/FR true ROAS quotes Shopify-EUR-revenue ÷ Google-Ads-GBP-spend without spelling out FX conversion. Magnitude ~5% at current EUR/GBP ≈ 1.18, consistent across both markets so comparative ranking unaffected. Standardise on a single quoted currency per-market in the next pass.

1. UK refund-event acceleration is recent and concentrated. 470 events across 2025 (up from 187 in 2024) and 175 events in just 4 months of 2026 implies ~1.6× annualised rate vs 2025. Worth a session: which SKU / channel mix is driving line-item partials? (Possibility: subscription cancellations partial-refunding the latest cycle.)
2. FR is the only market with positive YoY revenue trajectory in the 25-month window, AND its brand-search cost is just £4.4K for £51K revenue. Underfunding more acute than UK by ratio. Strategic implication: incremental £ to FR brand search is currently the highest-marginal-ROAS lever in the portfolio.
3. YT FR campaigns have ~£2.4K spend at zero recorded conversions in the audit window. Either the conversion attribution is broken (likely — view-through-only campaigns wouldn't show in standard conv columns) or the spend is genuinely unattributed. One-hour audit before next FR YT push.

1. Lock the refund-rate definition at the SoT layer. Add to ref-tracking-triage.md: refund_rate = refund_amount / total_revenue (default for £-impact framing); refund_event_rate = refund_count / orders (default for behaviour-pattern framing). Synthesis docs MUST name which one they're quoting.
2. When publishing empirical tolerance bands, always quote the unfiltered series alongside the curated one. Format: −X% ±Y% (curated, n=21) / −P% ±Q% (full, n=25).
3. Cross-currency true ROAS: standardise on a single quoted currency per-market, document FX assumption inline.
4. For Anomaly-A-style reversals (ghost vs canonical streams): when a per-market synthesis applies a property-ID exclusion, the methodology block should restate the canonical property ID and the excluded ID — currently inferred only from feedback_ga4_canonical.md.

2026-04-30 SESSION COMPLETE — 100% PUBLISHED, ZERO FAILS

State of the world after 2026-04-30 mass-publish + QC + fix wave: 58 / 58 pages published across UK/US/DE/FR. 53 pass / 5 warn / 0 fail / 0 pending. Compliance: 44 green / 10 amber / 4 cleared / 0 RED-blocked. Pages Registry SSoT live at campaign-lp-strategy → "Pages Registry" tab + mirrored to tier-list Sheet (gid 285543633).

What shipped today (chronological):
1. Pages Registry SSoT — 58-row registry built across 4 markets via parallel Sonnet agents. Schema v1.0 locked. Tab 6 added to Creative LP Matrix dashboard. Mirrored to tier-list Google Sheet with conditional formatting.
2. Mass-publish CEP wave — 31 + 5 = 36 pages published via 4 parallel Sonnet agents (UK 14 + US 10 + DE 3 + FR 4) + 1 conversion agent (DE/FR 5 held cities). All HTTP 200, all postchecks confirmed.
3. DE/FR per-city template conversion — 5 held pages (Berlin/Hamburg/Köln/Lyon/Lille) converted from shared-template-plus-setting to per-city thin templates. Eliminated manual admin step.
4. 5 Liquid syntax errors fixed — Edinburgh + London + Paris + Lyon + Lille had apostrophe-in-single-quoted-string bugs. Auto-fixer script + live theme PUTs.
5. render → include swap on 4 market section files (bz-exam-city.liquid) — fixed empty H1 across all 22 city pages. Root cause: {% render %} scope isolation.
6. DE/FR meta description metafields set + body_html cleared on 7 pages — eliminated visible SERP placeholder ("Diese Seite nutzt das bz-exam-city Template").
7. Quiz B + C fixed — missing template + data snippet uploaded to UK live theme. Root cause was different from QC agent's diagnosis: assets existed in repo but were never deployed.
8. 4 RED articles published — legal sign-off received. DE ADHS/Ritalin + FR TDAH/Ritaline. Full SEO metafields set + ANSM/HWG-compliant disclaimers verified.

Lessons banked permanently (8 new feedback memories + 4 SOP gotchas): G23 Liquid {% render %} scope isolation · G24 single-quoted Liquid + apostrophe = parser bug · G25 body_html as auto meta description fallback · G26 per-variant thin templates beat shared+setting · CEP bulk-write 4-bucket grouping (canonical) · SSoT registry pattern (canonical) · gws CLI --params vs --json + +append auto-routing gotchas · tools/fix_liquid_apostrophes.py auto-fixer · SHOPIFY_PAGE_PUBLISH_SOP v2.0 + THEME_COMPONENT_SOP G4 + asset-deploy-vs-repo.

5 remaining warns (no regression, separate workstreams):
• US Wave-2: PERI10 + BF10 discount codes missing on live us-peri-wave2 + us-bf-wave2 (article-template body injection)
• 5 amber-compliance pages (Dopamine collections UK/US/DE/FR + FR Anti-Fatigue) — content sign-off gates only, no customer-visible bugs

Scheduled follow-up: Remote agent SCHED-010 fires 2026-05-07 09:00 BST — re-runs QC sweep on all 58 pages, generates report, emails Calum via Gmail MCP. Catches drift, stale state, or new pages added without going through hooks (Task #7 not yet built).

Overnight Builds Review — 2026-04-27 (POST-CUTOVER + QC SWEEP)

22 articles + 13 collections (31 total) deployed to LIVE themes via Path B file-copy across UK/US/DE/FR. 22 articles are PUBLISHED + public; 13 collections remain published=false. Theme assets (sections, snippets, templates, CSS, JS) are LIVE-side; admin preview works without staff cookies.

What landed tonight (in chronological order):
1. 16 GABA + Dopamine illustrations via GPT-Image-2 (Calum new directive: GPT-Image-2 = default for illustrations, FLUX only for product-alongside scenes); £0.50 cost, 16/16 first-pass; cross-store push to all 4 stores' assets/.
2. UK Perimenopause editorial (greenfield) — article id 1001007186301, handle perimenopause-focus-guide, F2 pattern, ELITE-targeted, compliance CLEAN.
3. UK Brain Fog editorial (greenfield) — article id 1001007153533, handle brain-fog-explained-and-what-helps, advertorial pre-click warm-up, CTAs route to quiz + collection.
4. 7 Perimenopause + 8 Brain Fog illustrations via GPT-Image-2; £0.47 cost combined; cross-store push.
5. US GABA + Dopamine ports built on Session 38 framework, FDA-lensed (£→$, EFSA→FDA, etc.). Articles 570713669824 + 570713702592, blog focus-learning-center.
6. DE GABA + Dopamin ports, HWG-translated, "kein Arzneimittel" disclaimer placed prominently 2x. Articles 627993772296 + 627993805064, blog adhs-medikamente-zum-lernen, handles gaba-vorstufen-leitfaden + dopamin-leitfaden.
7. FR GABA + Dopamine ports, ANSM-translated, "complément alimentaire pas un médicament" disclaimer placed 7-8x. Articles 612915544406 + 612915609942, blog news, handles guide-precurseurs-gaba + guide-dopamine.
8. SEO meta sweep on all 13 articles (UK 7 + US 2 + DE 2 + FR 2) — meta title (≤60c) + meta description (~140c) per article via metafields. H1/H2/H3 hierarchy verified clean across all sections.
9. Critical bug fix shipped: G22 (bz-component-carousel) {%- doc -%} → {%- comment -%} on UK + DE (Session 38 framework had Theme-Check-only doc tag that breaks article render at runtime). US + FR already clean.
10. Path B cutover — all 13 publish configs deployed to LIVE themes via shopify_publish_page.py --skip-publish. New --skip-publish flag added to script.

Calum's morning checklist (per article):
1. Open admin preview on each store's LIVE theme → Articles → preview unpublished article. (Or use the article admin URLs in the cards below.)
2. Compliance pass per article — AMBER items (UK Dopamine + FR Dopamine + 1 small "soutenir mesurably" typo) need explicit sign-off.
3. Create discount codes per market: GABA10/DOPAMINE10 (UK + US + FR), GABA10/DOPAMIN10 (DE), PERI10 (UK), BF10 (UK).
4. Per article: PUT {"article": {"id": <id>, "published": true}} via Admin API or use Shopify admin "Publish" toggle.
5. Then run Browserbase QC per published URL (was blocked pre-publish by G13 staff-cookie).

Heroes still missing: The full-bleed product-shelf hero shots (bz-{topic}-hero-shelf-wide.jpg) for GABA / Dopamine / Peri / Brain Fog need separate FLUX product-shelf shoots — they're not part of the GPT-Image-2 ligne-claire batch. Articles render with broken hero placeholder until those land.

Coordination note: Session 38 finished multi-store framework rollout (bz-component-carousel + Quiz engine + component registry) to US/DE/FR mid-session — that unblocked the US/DE/FR ports. Earlier DE WIP from a halted pre-framework port agent preserved in stash halted-de-port-agent-2026-04-27-pending-session38-framework — safe to drop after confirming the clean re-build looks right.

WAVE 2 EXPANSION — 9 more pages built (22 total tonight)

Per Calum evening directives: build 2-3 versions per LP slot for A/B/C testing (Collection · Blog · Quiz). 9 wave-2 pages built + deployed to LIVE themes via Path B --skip-publish. All UNPUBLISHED.

UK Collection variants (A/B vs Blogs): gaba-precursor-supplements (id 682565927293) · dopamine-support-supplements (id 682565992829) · perimenopause-focus-supplements (id 682566091133). All UNPUBLISHED. Default Shopify collection template (body_html intro above product grid). Pre-existing duplicate handles `buy-gaba-supplements-uk` + `buy-dopamine-supplements-uk` still present — you decide keep-vs-archive.

UK GABA+Dopamine combined Recommender Quiz: find-your-focus-path-quiz (id 1001007513981, template bz-quiz-uk-gd). 8 questions / 3 phases. Engine-compatible 5-cluster scoring. Recommendations: ELITE for GABA-dominant (stress) · PRO for Dopamine-dominant (overload/nutrition) · BUNDLE for mixed (tied within 10%) · ORIGINAL only for child route (Q1=No, age 10-15) per your locked logic. Decisions D28-D30 added to decisions.md.

US gap-fill triple: US Peri editorial (id 570713866432, handle perimenopause-focus-guide) · US ADHD aggregating Collection (id 351133040832, handle adhd-focus-supplements, 3 products linked) · US Brain Fog editorial (id 570713931968, handle brain-fog-explained-and-what-helps, dual-pathway CTAs).

DE Konzentrationsplaner extension: konzentrationsplaner (id 627994886408). PRO primary (Motivation/Dopamin-Wege), ELITE secondary. 5-day week planner format. HWG disclaimer 2x. Discount code KONZ10 needed.

FR Anti-Fatigue Mentale editorial: guide-anti-fatigue-mentale (id 612915904854). PRO primary, ELITE secondary (existing surmenage-fatigue-mentale handle). ANSM disclaimer 6x. UFC-trigger language ABSENT. Discount code FATIGUE10 needed.

UTM tracker: 13 wave-1 pages logged to Google Sheet 1f1Q43vPnOjZEsruFWq5R015beJeI9VBvWTE8ttHwQVk tab "2026 Launch UTMs" (range A40:AA52, 27-column UTM v2 schema). Wave-2 batch (9 pages) being appended below by parallel UTM agent.

Discount codes still needed (per market): UK GABA10 / DOPAMINE10 / PERI10 / BF10 · US GABA10 / DOPAMINE10 (+ optional US peri/bf codes) · DE GABA10 / DOPAMIN10 (no E) / KONZ10 · FR GABA10 / DOPAMINE10 / FATIGUE10. Create in Shopify admin per store before publish.

Hero shelf images still pending for: UK Peri / UK BF / DE Konzentration / FR Anti-Fatigue / US Peri / US BF. Separate FLUX product-shelf shoot (not part of tonight's GPT-Image-2 ligne-claire batches).

Wave-2 illustrations needed for: DE Konzentration (8 illust + 1 hero, scenes around focused work / planner / circadian rhythm) and FR Anti-Fatigue (8 illust + 1 hero, scenes around mental fatigue / cognitive recovery). UK Quiz GD also needs illustrations.

Wave-2 SEO sweep needed — the 2 SEO agents tonight ran BEFORE wave-2 builds. Wave-2 pages don't yet have meta_title/meta_description metafields. Schedule a third SEO sweep next session covering: UK Quiz GD, UK 3 Collections (note: collection metafields differ from article), US Peri/ADHD-Coll/BF, DE Konzentration, FR Anti-Fatigue.

Scoring weights review for UK Quiz GD — per-question weights are author judgments. Recommend manual test of 4 paths: child route, GABA-dominant, Dopamine-dominant, mixed.

WAVE 3 EXPANSION + Campaign Audit — 9 more collections + ad violations report (31 total tonight)

Per Calum directive 2026-04-27 evening: complete LP coverage gaps + audit live ad copy against Marcom Matrix v3.2.

Wave 3 Collections (9): US 4 (gaba-precursor-supplements 351168757952 / dopamine-support-supplements 351168790720 / perimenopause-focus-supplements 351168856256 / brain-fog-supplements 351168889024) · UK ADHD aggregating-2 (adhd-focus-supplements 682616390013) · DE 2 (gaba-praparate 576598212872 / dopamin-praparate 576598245640) · FR 2 (complements-precurseurs-gaba 695193633110 / complements-soutien-dopamine 695193665878). All UNPUBLISHED. FR has dedicated GABA/Dopamine product handles (`complement-gaba` + `complement-dopamine`) — earlier FR product-handle gap notes were stale.

LP coverage post-wave-3: UK ✅ all lanes ≥3 variants · US ✅ all 5 lanes ≥2 variants (Quiz Phase 2/3 still queued) · DE ✅ Konzentration ≥3 + GABA/Dopamin at 3 · FR ✅ GABA + Dopamine + Anti-Fatigue at 2-3 variants · RED gates remain on DE ADHS / FR TDAH / Ritalin lanes.

⚠️ CAMPAIGN AUDIT — CORRECTED 2026-04-27 22:00 (POST-METHODOLOGY-FIX). The original audit ran without filtering on the joint condition campaign.status=ENABLED AND asset_group.status=ENABLED AND asset_group_asset.status=ENABLED — producing false positives on PMax-asset claims. After re-run with the proper 3-status filter, the TRUE-violation queue is:

FALSE POSITIVES (zero rows on triple-ENABLED filter): Black Friday assets (all 4 PMax) · "BF15" code · "nootropic" in any PMax description · "Performance Mentale" / "Effets Max" (FR PMax) · "Unlock your full potential" (UK PMax) · "Peak Mental" (UK PMax) · "is design to" typo (UK PMax) · "boosters" (UK/US PMax). All previously claimed as live; verified REMOVED at the asset-link level.

CONFIRMED LIVE violations (post-3-status verification):
• "Trusted by 2,000+ Doctors" / "Loved by Doctors" — 2 assets ENABLED in PMax USA (asset_group 6448919526). NOT in PMax UK (which audit had wrongly cited).
• "Über 2.000 Ärzte vertrauen Brainzyme®" — 1 asset ENABLED in PMax DE Main (asset_group 6558136289).
• "Mit Brainzyme® FOCUS rezeptfrei & legal Leistung steigern" + 2 variants — 9 ENABLED rows across PMax DE Main + DE Ritalin Germany + DE Ritalin Austria. HWG legal scope review needed (DE Ritalin allowed per exception; DE Main exposure questionable).
• "powerful, plant-powered brain supplements" — 1 asset ENABLED in PMax UK (asset_group 6509497072). Banned optimiser language per matrix (H13).
• "Game-Changer" English idiom in DE copy — 1 asset ENABLED in PMax DE Main (asset_group 6521203827).
• SERP UK ADHD Bundle (campaign 11666978147, ad_group 126235035533, ad 706882072278) ALL CONFIRMED LIVE in single ENABLED ad group: "Attention Deficit Issues?" headline + "Manage Attention Fast" headline + "Scientifically proven natural solution to support attention issues fast." description. £1.8k/mo at ASA-exposure risk. NatAlts-DKI + Focus&Attention-DKI ad groups (originally cited for "Doctors" headlines) are PAUSED — not live-serving.
• "Brain Nootropic Supplements" headline — 1 ENABLED RSA in SERP UK Brand (single ad 618559319501). LEARNING status, not pinned. C1 confirmed.
• Brand-mark missing ® — SERP DE: "Brainzyme FOCUS Pro™" + "Brainzyme FOCUS Elite™". SERP FR: "Brainzyme FOCUS ORIGINAL™" + "Brainzyme FOCUS PRO™" + "Brainzyme FOCUS ELITE™". L1+L2 confirmed live.
• SERP US Brand RSA (campaign 13522300536, ad 527830378007) is verbatim UK copy-paste — no FDA structure-function localisation. H2+H18 confirmed.

Full corrected priority queue + per-violation evidence at v4.0/campaign-content-audit-2026-04-27.md. Markets-by-true-cleanliness rank: US (low: 2 confirmed) → FR (low: brand-mark only) → UK (mid: ADHD ASA + nootropic + powerful) → DE (high: HWG scope + Ärzte + Game-Changer + brand-mark).

Master multi-phase plan banked: 7 phases (A→G) at v4.0/go-live-rundown.md. Phase A (LP gap-fill) + Phase B (audit) DONE tonight. Phases C-G defer to next sessions: C=copy refresh, D=image rerun, E=video selection (Session 50), F=ad-group retrofit, G=final QC. Sequencing + Calum gates per phase documented.

WAVE 6 — 16 city LPs + 4 RED articles + 36 illustrations + 3 multi-store framework ports (POST-OPUS-QC-SWEEP)

✅ OPUS QC SWEEP PASS (2026-04-27 → 2026-04-28). Full-spectrum verification: Block A 16/16 LPs verified UNPUBLISHED + correct handles via Admin API. Block C 40/40 framework files (CSS/JS/Liquid/templates/snippets) on each store's live theme. Block D 4/4 RED articles compliance-clean (0 banned phrases in customer copy after stripping Liquid+HTML comments; all required disclaimers present). Block E 36/36 city illustrations present + sized 90-170 KB. Block F preflight: 21/22 cities PASS after fixes (4 confirmed FAILs surfaced + corrected — see below).

What landed in Wave 6 (chronological):
1. US: 5 city LPs + framework port (Boston / NYC / LA / Chicago / Austin) — page IDs 120522539200/571968/604736/637504/670272. Per-city thin templates (admin-config-free). FDA-lensed. Commits US `979d4f3` + UK xref `8670820`. All UNPUBLISHED.
2. FR: 3 city LPs + framework port + 2 RED articles — Paris/Lyon/Lille (160561201494/234262/267030). 2 RED: TDAH guide 613444747606 + Ritaline comparaison 613445599574. ANSM-defensive disclaimers. Commit FR `30d71f3`.
3. DE: 4 city LPs + framework port + 2 RED articles — München/Berlin/Hamburg/Köln (164673487112/519880/552648/585416). 2 RED: ADHS-Leitfaden 628058685704 + Ritalin-Vergleich 628058718472. HWG-konform + management exception. Commits DE `ad113b5` + `4ef2adb`.
4. 36 city illustrations (US 15 / DE 12 / FR 9). GPT-Image-2 medium, 36/36 first-pass QC. £1.13 spend.
5. Compliance fixes shipped (Opus QC sweep finding): 4 files PUT to live themes — UK + US sections (`Wired but tired` → `Restless but exhausted`; `no withdrawal` rewritten); UK Manchester snippet (`No crash. No withdrawal.` → `No crash, no jitters.`); FR section (`cafféiner` → `avant le café`; `soutenu nutritionnellement` → `soutenu par la nutrition`; `measurablement` typo). All preflight PASS post-fix. Commits UK `3920eca` / US `66a2337` / FR `a352a8a`.

⚠️ Architecture finding — TWO PATTERNS in Wave 6 city deployment:
• Per-city thin templates (works without admin config): UK 10 cities (incl. all Wave 5.5) + US 5 cities = 15 pages. templates/page.bz-exam-city-{slug}.json with city_slug hardcoded.
• Shared template + section setting (needs Calum admin step): DE 4 + FR 3 = 7 pages. templates/page.bz-exam-city.json shared, all 7 default to munich/paris until city_slug set per page in admin.
Recommendation: KEEP shared+setting for now (works, just needs 7 admin clicks). Convert later as a focused refactor task. Per CLAUDE.md L5 — don't bundle preventative architecture refactor with confirmed compliance fix in same deploy round. Calum's morning to-do: theme editor → each DE/FR page → bz-exam-city section → set city_slug → Save (7 pages × 30s).

Calum morning to-do (Wave 6):
1. Set city_slug per page in admin for 7 DE/FR pages (Berlin/Hamburg/Köln/Lyon/Lille — Munich + Paris are template defaults).
2. RED article legal review — both DE (ADHS / Ritalin-Vergleich) + FR (TDAH / Ritaline-Comparaison). All 4 are NEEDS-LEGAL-REVIEW status (default for RED). 0 banned phrases in customer copy verified by Opus.
3. Discount codes — verify WELCOME10 active in UK + US + FR Shopify admin; verify GABA10/DOPAMINE10/DOPAMIN10/FATIGUE10/WELCOME10 per-market.
4. Publish toggle — per page/article via admin "Publish" toggle when ready. Suggested order: city LPs first (lower compliance risk), RED articles after legal review.
5. RED article illustrations still needed — DE ADHS 8 illust + DE Ritalin 8 illust + FR TDAH 1 hero + 8 illust + FR Ritaline 8 illust = 33 illustrations. Section files have onerror="this.style.display='none'" so missing illustrations gracefully hide.
6. UTM Sheet — Wave 4-6 URLs (~28-30) not yet logged to Sheet 1f1Q43vPnOjZEsruFWq5R015beJeI9VBvWTE8ttHwQVk tab "2026 Launch UTMs". Schedule a follow-up UTM batch.

Lessons-learned register: G22 doc-tag · G23 JSON template comment strip · Splide→Embla · --skip-publish · PYTHONIOENCODING=utf-8 · pull-rebase-before-push · drift-quadrant · per-market compliance · GPT-Image-2 default · L5 fix-isolation (Wave 6 confirmed) · architecture-divergence detection. Apply to ALL new work.

Plan summary — what's IN vs DEFERRED

Image generation — model routing decision (banked tonight)

Per Calum 2026-04-27: GPT-Image-2 (gpt-image-1) is the default for illustrations. FLUX is reserved for scenes where products need to be shown alongside something else to make a point (e.g. product-in-context heroes). Tonight's 16 ligne-claire scene illustrations are all conceptual/atmospheric — all generated via GPT-Image-2.