Go-Live Journey — Cycle 1

The 6-phase journey

We've been using "Phase 1 / Phase 2" loosely for the cascade determinism tooling (5 gates + cascade-qc-gate skill) — those are infrastructure phases. The cycle-1 go-live journey is its own track, mapped below. Each phase has a clear deliverable + gate to the next.

Phase 0 — Pre-cycle-1 foundation Done early May

v4.0 LP cascade template stood up · 11 UK city LPs deployed · 3 FR city LPs deployed · PageFly subscriptions A-baseline live · brand-lifecycle PMax PAUSED with audience signals wired

Phase 1 — Cycle-1 cascade dispatch Done 13 May

6 tests planned (T1-T5 + #65 SERP Search co-run) · 8 subagent streams · ~840KB local markdown briefs at projects/briefs/cascade-2026-05-13/ · Creative Production Tracker bypassed (gap surfaced + closed in Phase 3)

Phase 2 — Retrospective + determinism tooling Done 14 May AM

7-audit retrospective surfaced 14 findings (2 CRITICAL + 6 HIGH + 4 MED + 2 LOW) · Phase 1 determinism tooling shipped: 5 deterministic gates (check_canonical_facts · cascade_rename · power_analysis · check_test_id_uniqueness · verify_subagent_claims) at commit 90210d4

Phase 3 — Path B re-scope + ARC retro-fit Done 14 May PM/eve

Cycle-1 cut 6 tests → 4 (T2 + #65 dropped per Opus + Phase 1 power-analysis BLOCKs) · T1/T3/T4 single-arm B-only · T5 A/B preserved with 42-56d extended window LOCKED · arch_id retro-fit complete for 4 retained tests · AGENTS.md "Before dispatching a cascade" MANDATORY rule + Gate 6 SOP + Phase 2 cascade-qc-gate skill all SHIPPED · 9 commits across 2 repos

YOU ARE HERE

Phase 4 — Cycle-1 ENABLE Current Phase 5.5: 4/4 NO-GO

All cascade artefacts + both v4.0 B-side LPs (/pages/welcome-back-v4, /pages/subscribe-and-save-v4) are SHIPPED. The Phase 5.5 Targeting + Intent QC gate ran — wave 1 self-audit + wave 2 Codex/Gemini adversarial review — and returned all 4 retained tests NO-GO. ENABLE is now gated on 18 decision cards (A–R) on the Go-Live Dashboard, which Calum is working through.

⚠️ Phase 5.5 QC gate — 4/4 NO-GO; ENABLE held on 18 decision cards

Phase 5.5 verdict: T1 / T3 / T4 / T5 all NO-GO. Wave 2 (Codex 5.5 + Gemini Pro) confirmed every verdict and surfaced 6 HIGH findings wave-1 missed — mostly Google Ads control-plane mechanics (PMax doesn't enforce per-AG budgets; audience signals aren't hard fences; the T5 A/B is statistically underpowered).
[Calum] Wave-2 strategic decisions — Cards H–L · T3 PMax structure · T5 statistical power (re-opens E) · T1/T4 medication-comparison intent gating · 18-24 exclusion.
[Calum] ENABLE-gate decisions — Cards M–R · T1 budget · T1 technical fixes · PageFly A-page freeze · 3 non-UK caffeine PUTs · US L241 AMBER · T3 6 Customer Match lists.
[Calum] Card A ASA neurodivergent fix & Card G image library — both IN PROGRESS.
[Claude] Parked execution tasks — T4 URL expansion fix (5 FR handles), T3 ENABLE-CHECKLIST spec-drift fix, v4.0 LP image-slot fills. Unblocked; awaiting go-ahead.
[Claude] + [Calum CEP] 4 retained tests PAUSED → ENABLED — fires once the decision cards clear.

Phase 5 — Smoke test live (cycle 1) Next

4 retained tests run live · T1 + T4 single-arm B (~28d standard) · T3 single-arm B (~28d) · T5 A/B (42-56d extended) · Power analysis at data midpoint · Drift detection · Early-kill or extend per Phase 1 power_analysis.py thresholds (50 conv/wk floor for PMax)

Phase 6 — Cycle-1 readout + Cycle-2 planning Future

Win/loss declaration per test · Cycle-2 scope decisions · FR city expansion lives here

Cycle-2 queue (locked-in scope)

FR city expansion — 3 → 25 cities + Bac-specific + parent variants (task #22). First cascade under new arch_id-mandatory regime; Phase 2 cascade-qc-gate Gate A will fire pre-dispatch.
UK city expansion — 11 → 15 cities (Oxford 910/mo · Reading 250 · Cambridge 180 · Brighton 90 per DataForSEO audit) + parent-variant template work (task #21)
UK tier differentiation Option B — URL parameter ?tier=gcse|alevel|higher driving Liquid template hero/testimonial/CTA swap (task #65). Decided Option A for cycle 1; Option B queued here.
Campaign Planner ↔ Go-Live one-way sync — Sheet (canonical SoT) → dashboard read cache, mirroring Pages Registry pattern (task #47)
Re-evaluate #65 SERP Search co-run — preserved as cycle-2 reassessment input; with more budget + cleaner architecture, dedicated Search can be re-added cleanly
Re-evaluate T2 UK DG-Parent DG — dropped from cycle 1 for ASA risk + thin budget; reassess at cycle 2

Recent developments (14–15 May)

15 May Go-live-decisions board freshness pass + K31/K33 deep dive — all 56 open CRITICAL/HIGH items verified against 6 days of activity (Codex-QC'd subagent). 1 resolved (I01), 2 superseded (K31/K33 — a Pinecone + campaign-architecture deep dive confirmed they are the old framing of the cascade exam-season workstream, not a parallel build). Board confirmed substantially accurate.
15 May Go-Live Dashboard — 18 decision cards (A–R) — Phase 5.5 panel added (4 NO-GO test cards with detail drop-downs + comment boxes). Pending Decisions expanded to 18: 5 wave-2 strategic cards (H–L) + 6 ENABLE-gate cards (M–R). Card F resolved (ELITE-lapsed combine); Card E annotated SUPERSEDED by Card I.
15 May Phase 5.5 wave 2 — 4/4 NO-GO confirmed — Codex 5.5 + Gemini Pro adversarial review of the 4 Phase 5.5 QC reports. Both confirm all 4 verdicts STAND; surfaced 6 HIGH findings wave-1 missed (PMax control-plane mechanics) + 6 decisions for Calum. Synthesis: cascade-2026-05-13/phase-5-5-wave-2/wave-2-synthesis.md.
14–15 May Both v4.0 B-side LPs SHIPPED — /pages/welcome-back-v4 (T3 — Option C built as client-side JS cohort branching; Shopify Liquid can't read request.query at the cache layer) + /pages/subscribe-and-save-v4 (T5). Both live, awaiting real-browser QC. Multi-market caffeine sweep: Calum chose Option A — all 11 UK exam-season city LPs revised in one template edit; US/DE/FR PUTs queued (Card P).
14 May 21:30 Option C locked for welcome-back-v4 — Calum decision · URL-param dynamic modules approach. Single Shopify Page at /pages/welcome-back-v4 with Liquid template reading request.query['cohort'] = elite / pro / original, swapping hero / featured-SKU / testimonials / CTA per cohort. Same canonical URL for SEO. PMax asset groups (one per LAPSED × Product cohort) carry the param in final_url. Mirrors the UK tier-differentiation pattern decided for cycle 2.
14 May 21:00 Cardiff QC PASS — pages_registry row 93 flipped qc_result=pass · 5/5 functional phases passed (HTTP / compliance / Liquid+schema / Playwright mobile+desktop / assets) · 13 Cardiff-specific mentions confirmed (ASSL library, Cathays Park, Cardiff Uni, Cardiff Met) · 0 broken images.
14 May 21:00 "no caffeine" template-wide flag — RESOLVED — Cardiff QC surfaced "no caffeine, no banned substances" in all 11 UK exam-season city LPs (factually false — PRO/ELITE contain 80mg Guarana). Calum chose Option A: all 11 UK LPs revised in one shared-template edit. US/DE/FR equivalents queued for approval (Go-Live Card P).
14 May 20:30 subscribe-and-save-v4 LP build — SHIPPED — T5 B-side live at /pages/subscribe-and-save-v4. Built + canonical gate + Shopify Dev MCP validation + registered in pages_registry. Image slots carry placeholders until Card G lands; awaiting Calum real-browser QC.
14 May 18:00 T5 42-56d extended window LOCKED — Calum decision · NOT the £40/d budget bump alternative. Test window: 42-56d (vs the 28d cycle-1 default for T1/T3/T4). Retro window: ~2026-06-24 to 2026-07-08. Budget unchanged ~£20/d. Rationale: lower cash exposure while surfacing significance through duration.
14 May 18:00 UK tier differentiation: Option A for cycle 1, Option B for cycle 2 — Calum decision · Cycle 1 keeps tier-agnostic UK LPs (GCSE / A-level / Higher all see same /pages/exam-season-{city} copy). PMax custom-intent + keyword clusters provide tier-level signal at audience layer only. Cycle 2: URL parameter approach for Liquid template swap (task #65).
14 May 17:30 Gate 6 v1 + v2 SHIPPED — tools/check_cascade_arch_id_tracker.py 699 LOC + 420-line test suite (16/16 PASS) + fixtures README. Real-cascade smoke test PASS for 4 retained tests. v2 SKIP_DIR_NAMES extension for LP/template support dirs.
14 May 17:30 Phase 2 cascade-qc-gate skill SHIPPED — 5 files: SKILL.md (10,306 chars frontmatter 901 / body 160 lines) + 3 references docs + run_gate_a.sh 186-line aggregator. Validator tools/check-tool-map.py exits 0; skill count 39 → 40. Two-gate workflow: Gate A pre-dispatch deterministic + Gate B pre-enable multi-agent review (all GREEN required).
14 May 17:30 FR segment 365d → 540d propagated — bz_2plus_purchases_12mo_no_sub renamed bz_2plus_purchases_18mo_no_sub · verified 1,385 members (was 944) · clears 1k Customer Match floor. segment-binding.md filter + criteria_broadened metadata + size estimate all updated.
14 May 16:00 Path B SELECTED + ARC retro-fit committed — 9 parallel agents + 1 independent verifier. 3 Codex HIGHs fixed inline pre-commit. T1 → pmax_uk_attention · T3 → pmax_uk_lapsed · T4 → pmax_fr_anti_fatigue · T5 → pmax_uk_brand (all asset_group_within consolidation per Q1 decision). Master Sheet playbook gets 8 rows for the Phase 1+2 system.
14 May 16:00 AGENTS.md "Before dispatching a cascade" MANDATORY rule + CLAUDE.md mirror — every cascade test MUST resolve arch_id (mint or consolidate via config/campaign-architecture.json) AND populate Creative Production Tracker row with arch_id BEFORE subagent dispatch. No tracker row + arch_id = no cascade dispatch. No bypass.

Cross-links

🎯Financial Audit v2.28 — Path B SELECTED 🚀Go-Live Dashboard — AG-level decisions 📝Decision Board — pending Calum decisions 🏠Command Centre