/funnel

Quality Gates

Per-stage exit criteria for the funnel premium LP studio chain. A stage that fails its gate either repairs in place (≤1 retry) or escalates back to the user.

Gates are not negotiable. Recipe selection determines which stages run, but every stage that runs must clear its gate.

Gate Mechanics

Gate evaluation is owned by funnel premium (not the delegate). funnel premium reads the delegate's output, checks against the gate's measurable criteria, and chooses outcome.

Oscillation Guard

If the same criterion flips PASS↔FAIL across 2 retries (same axis or cross-axis), funnel premium skips the standard 1-retry rule and escalates immediately to Magi for trade-off arbitration. Examples:

• DESIGN AA contrast PASS → repair for Visual Hierarchy → contrast FAIL again → escalate (don't loop)
• OPTIMIZE Motion ≥15/20 → Bolt INP repair → Motion drops to 14/20 → escalate (Bolt vs Flow ping-pong)
• GEO 14/20 → Growth adds TL;DR → Brand voice variance ≥1.5 → escalate (Growth vs Prose conflict)

Trade-off Ping-Pong Detector

When repairing axis A causes axis B to regress in the same stage, do NOT use the 1-retry budget. Route to Magi for Logos/Pathos/Sophia arbitration. Common ping-pongs (see also "Axis Trade-off Discipline" below):

• Motion richness vs INP budget (Flow ↔ Bolt)
• Hero asset richness vs LCP (Builder/Ink ↔ Bolt)
• SEO keyword density vs Brand voice (Growth ↔ Prose)
• Detail-craft completeness vs release date (Vision/Palette ↔ Launch)

Stage 1 · DISCOVER Gate

Criteria:

• 3+ market insights with citations (Field) — skip if recipe = saas/ecom/event/magnet with existing brand
• Top 3 competitor LPs teardown with feature matrix + 2+ differentiation hooks (Compete)
• Sentiment themes + top-5 pain quotes (Voice) — skip if no existing customers

Repair triggers:

• Generic insights ("the market is growing") — re-dispatch with sharper brief
• Competitor teardown missing CTA / above-fold analysis — re-dispatch
• Pain quotes are summarized, not verbatim — re-dispatch for quotes

Escalate when:

• Field cannot find category data → ask user for category framing
• Compete cannot identify direct competitors → ask user to name them

Stage 2 · AUDIENCE Gate

Criteria:

• 1–3 personas, each with: name, role, goals, frictions, jobs-to-be-done, channel, decision driver (Cast)
• Journey map covering Awareness → Consideration → Decision → Post-purchase (canvas-rendered or text-form)
• 5+ unmet needs ranked by intensity (Echo[demand])
• Echo baseline mandatory: if a draft structure/wireframe exists, Echo cognitive walkthrough produces emotional friction score (target ≤ Medium). If none exists, Echo runs on Funnel's preliminary section-list spec at minimum — never skip entirely, otherwise VERIFY-stage re-walk has no baseline to compare against and design regressions become undetectable.

Repair triggers:

• Personas are too generic ("decision maker, 30–50") — re-dispatch Cast with named ICP attributes
• Journey map skips a stage — re-dispatch
• Unmet needs duplicate published feature list — re-dispatch Echo[demand] with "feature gaps" reframing

Escalate when:

• Cast cannot anchor on a primary persona without a stakeholder decision → ask user

Stage 3 · STRATEGY Gate

Criteria:

• CVR target locked (Median / Top Quartile / Top Decile + chosen target + traffic-source qualifier), calibrated against conversion-playbook.md industry baseline; recipe_alignment_check PASS
• KPI tree: primary (CVR) + 3–5 secondary metrics (Pulse)
• Funnel event taxonomy: page_view → engagement → CTA_click → form_start → form_submit → success (Pulse)
• North-Star metric named and consistent with KPI tree
• GEO measurement plan: Mention Rate ≥15% target, Citation Rate baseline+30%, Share-of-Voice tracked (Pulse + Growth alignment)
• Strategic deliberation logged (Magi) for: primary persona, primary objection, urgency framing
• Brand System triple lock (machine-checkable): all three of vision_archetype_locked, saga_story_locked, compete_positioning_locked are true in the project state. Saga's Strategy-stage output is the brand story arc (Hero/Guide/Problem/Plan/Success/Failure); a separate narrative-copy pass happens at STRUCTURE.
• Above-the-fold asset weight budget declared: hero image ≤200KB / hero video ≤2MB / total font ≤100KB. Hero asset lead-time risk surfaced (lead-time ≤ remaining stage budget).

Repair triggers:

• CVR target not industry-anchored — re-dispatch Pulse with playbook reference
• Event taxonomy missing form fields or success event — re-dispatch
• Magi verdict is "no decision" — re-dispatch with sharper question

Escalate when:

• User has not approved CVR target → ask once before Structure stage
• Strategic deliberation reveals two-promise drift → escalate to enforce one-promise rule

Stage 4 · STRUCTURE Gate

Criteria:

• Primary scan pattern locked (F / Z / Layer-cake / Centered axis / Card grid) per ia-blueprint.md §1.2
• Navigation pattern decided (no-nav / utility bar / sticky CTA / anchor TOC / mobile bottom bar / exit-intent) per ia-blueprint.md §1.6
• Wireframe outline: hero + 5–7 sections (Funnel)
• Above-fold component set explicit: headline + sub-headline + hero CTA + (optional secondary CTA) + hero asset + (optional trust strip) — each present-or-absent decision logged
• Copy v1 covering: headline, sub-headline, hero CTA, 3–5 benefits, 1+ objection handling, social proof block, FAQ (3–5 Q), final CTA (Funnel + Prose)
• Reading level appropriate to persona (FK 6–9 for consumer, 8–11 for B2B) (Prose)
• Saga arc (when invoked): narrative copy applied — before/after frame + customer hero arc on at least 1 testimonial; this is distinct from the Strategy-stage Brand Story arc
• Form consent microcopy drafted (if form-driven LP): consent/unsubscribe/GDPR notice copy below form fields
• Thank-you / post-conversion page sketched for lead-gen / magnet / event recipes

Repair triggers:

• Hero copy fails the 5-second test (cannot answer "what / for whom / why now" in 5s) — re-dispatch Funnel + Prose
• Objection handling absent or generic — re-dispatch Prose with persona objection list from Audience stage
• Social proof block is placeholder ("[insert logos]") — re-dispatch with specific proof types (logos / numbers / quotes / case studies)

Escalate when:

• Brand has no usable social proof yet → ask user to choose "trust mechanic" (founder authority / industry endorsement / press / open-source signals)

Stage 5 · DESIGN Gate

Criteria:

• Vision direction approved (direction.md or equivalent) with: archetype, mood, type direction, palette intent
• Tokens frozen (Muse) — color, type scale, spacing scale, radius, motion duration baseline
• WCAG 2.2 AA contrast ratios verified — 4.5:1 text, 3:1 UI (Palette)
• Hero assets locked (Ink or Builder) — production-ready, license-cleared
• If Figma source exists: Frame extraction complete, Code Connect mapping if needed

Repair triggers:

• AA contrast fails for any text — re-dispatch Muse to adjust palette
• Hero asset is placeholder — re-dispatch Builder/Ink
• Tokens contradict Vision direction (e.g., dark archetype with high-saturation accent) — re-dispatch Muse

Escalate when:

• Vision direction is missing and user brief is ambiguous → route to Vision first, then resume
• Hero asset requires paid API spend > approved budget → ask user

Stage 6 · BUILD Gate

Criteria:

• Working prototype on tokens, no hardcoded values (Forge → Artisan)
• Production code passes type-check, lint, build (Artisan)
• Lighthouse Perf ≥ 80 on prototype (not yet 90 — that's Optimize stage)
• Motion implementation respects prefers-reduced-motion (Flow)
• i18n string extraction complete if Polyglot in scope — no hardcoded user-facing strings

Repair triggers:

• Hardcoded color / spacing values found — re-dispatch Artisan with token reference
• Type errors — re-dispatch Artisan
• Motion breaks INP budget on test devices — coordinate Flow + Bolt

Escalate when:

• Build framework choice was not specified and funnel premium cannot infer (no existing codebase) → ask user (React / Vue / Svelte / static)

Stage 7 · OPTIMIZE Gate

Criteria:

• Lighthouse Mobile: Perf ≥ 90, Accessibility ≥ 95, Best Practices ≥ 95, SEO ≥ 95
• Core Web Vitals: LCP ≤ 2.5s, INP ≤ 200ms, CLS ≤ 0.1 — all Green; also TTFB ≤ 800ms (p75), FCP ≤ 1.8s
• GEO rubric ≥ 15/20 (Growth) — JSON-LD, structured data, AI-search friendly headings, per-platform tactics. The /20 rubric is the canonical scale; ≥90 notation is deprecated.
• First A/B variant designed with hypothesis, sample size (≥1000 conversions/variant, 95% significance, ≥14 days), decision criteria (Experiment)
• Form CRO completed (if form-driven LP): field count justified by lead-quality filter rationale, label clarity, error state copy, segment-aware CTA where traffic permits
• Schema-content consistency check: AI engines verify schema fields match visible page content (e.g., Offer.price matches displayed price)

Repair triggers:

• Perf below 90 — re-dispatch Bolt with specific budget breach (LCP / INP / CLS)
• SEO/GEO below 90 — re-dispatch Growth with missing-element list
• Variant lacks measurable decision criteria — re-dispatch Experiment

Escalate when:

• Perf budget cannot be met with current asset weight (e.g., hero video) → ask user about asset reduction
• A/B test infrastructure missing → ask user about variant deployment strategy

Stage 8 · VERIFY Gate

Criteria:

• Judge: zero unresolved P1 or P2 findings (P3 acceptable with explicit ack)
• Voyager: E2E suite green covering happy path + form submit + CTA click + error states + visual regression baseline
• Attest: spec compliance ≥ 95% AC coverage against Structure-stage wireframe + copy spec
• Sentinel: zero high-severity findings (form input validation, headers, secrets scan, dependency CVE)
• Echo re-walk: friction score ≤ Low (improved from Audience-stage baseline)
• WCAG 2.2 AA: full audit pass (contrast, focus visible, keyboard reachable, ARIA roles, image alt text)

Repair triggers:

• Any P1/P2 — re-dispatch Builder/Artisan with Judge findings
• E2E flake — re-run; if reproducible, re-dispatch Voyager to stabilize
• Attest coverage gap — re-dispatch Artisan to fix or update spec
• Sentinel high finding — re-dispatch Sentinel with fix applied

Escalate when:

• Judge finds an intent misalignment (built feature ≠ briefed feature) → escalate to user
• Echo re-walk friction worsened vs. Audience stage → escalate; design regression

Stage 9 · LAUNCH Gate

Criteria:

• Release plan: version, rollout strategy (full / staged / canary), rollback plan, feature flag if applicable (Launch)
• PR ready: clean commits, conventional-commit messages, no agent names in commit/PR text (Guardian)
• Analytics events live and emitting: at least page_view + CTA_click + form_submit verified in dashboard (Beacon)
• Alert rules armed: 5xx spike, form-submit drop, CWV regression
• First A/B variant queued or "control only" decision documented
• Privacy / Terms / consent banner in place if PII collected (Cloak / Canon checked)

Repair triggers:

• Analytics not firing — re-dispatch Beacon
• Rollback plan missing — re-dispatch Launch
• PR fails Guardian classification (e.g., commit contains agent name) — re-dispatch Guardian

Escalate when:

• Rollback strategy requires infrastructure access funnel premium cannot grant → ask user
• Legal review (Canon) flags unresolved compliance item → escalate

Cross-Stage Continuous Checks

These run continuously, not at a single gate:

6-Axis Rubric Gate

Composite gate enforced at the DESIGN exit (Design / Animation / Branding rubrics) and the OPTIMIZE exit (Marketing / SEO / IA / GEO). All thresholds must clear before VERIFY.

Axis Repair Workflow

When an axis fails:

1. funnel premium flags the failing criterion with score and gap.
2. Re-dispatch the owning agent(s) with a focused repair bundle.
3. Re-score the rubric.
4. If still failing after one retry → escalate to user with a written choice (accept conditional pass / extend scope / cut feature).

Axis Trade-off Discipline

Conflicts between axes are managed, not silenced:

Gate Reporting Format

After each stage, funnel premium emits a stage report in this shape and persists to .agents/funnel-premium/{project}.json:

STAGE_REPORT:
  Stage: STRUCTURE
  Outcome: PASS | FAIL_REPAIR | FAIL_ESCALATE | CONDITIONAL_PASS
  Delegates: [Funnel, Prose, Saga]
  Artifacts: [wireframe-v1.md, copy-v1.md, saga-arc.md]
  Gate_Checks:
    - check: hero answers what/for-whom/why-now in 5s
      result: PASS
    - check: objection handling present and specific
      result: FAIL_REPAIR
      action: re-dispatched Prose with persona objection list
  Open_Items: []
  Next_Stage: DESIGN