Freddie Deepak Phase 2 specs
Playable prototype Phase 1 spec Editor brief tab
Overview Constraints Heatmap Eight features Components Execution
Section 1 of 6 · Phase 2 positioning

Freddie Deepak Phase 2 — where Mailchimp wins permanently

Phase 1 closed the creative parity gap vs. generalist design tools. Phase 2 maps the next eight capabilities from the same Canva-gap analysis into Mailchimp primitives: orchestrated sends, live performance feedback, assembly-time personalization, governed multi-channel assets, compliance-first localization, AMP, enterprise Shield, and a cross-domain Freddie agent. This document is the spec pair to Phase 1.
Prototype: email-first
Features: 8
8
capabilities spanning Plan → Send → Govern → Extend
C-98–102
five new baseline proposals (this phase)
ESP-native
telemetry + policy + MIME physics competitors lack
1 loop
performance data → scoped canvas edits
Phase 2 is deliberately not “more stickers.” It is send-time truth: the same object model that powers orchestration is the one that records approvals, evaluates Shield policies, and serves AMP with static fallbacks. Name cue: stakeholders asked for this tab to appear as “Freddie Deepak Phase 2 specs” (editor brief) — file + toolbar title align to that.

1.1Relationship to Phase 1

Phase 1 extended composer, manual rail, Brand Kit, settings, and MCP with ten creative-depth features. Phase 2 assumes those surfaces exist and adds downstream operational AI: agent runs with gates, metric-aware refinement, personalization assembly, derivative assets, locale law packs, interactive MIME, governance suite, and federated tool scopes.

Section 2 · Constraint manifest

Inherited rules + Phase 2 deltas

R1–R6 from Phase 1 still apply (tokens, C-ID discipline, honest channel parity). Phase 2 adds R7-ops and R8-policy.
Spec v1.0 · May 2026

2.1Load-bearing rules

R7-ops
Every “agentic” action must map to a workspace object (campaign, journey step, asset graph entry) with a checkpoint id — no orphan LLM traces.
R8-policy
Personalization and localization features ship with a suppress + explain contract: if a cell is blocked, marketers see the rule name, not a silent omission.
R4-bis
This prototype is email-first. SMS/WhatsApp equivalents are backlog unless an affordance is physically identical (e.g., asset crops).

2.2New component proposals (baseline IDs)

C-IDPrimitivePrimary feature
C-98Orchestration timeline / plan graph row01 Agentic Orchestrator
C-99Live KPI strip (contextual metrics)02 Performance Loop
C-100AMP module shell + validation status06 AMP for Email
C-101Shield evaluation summary panel07 Mailchimp Shield
C-102Cross-domain MCP tool registry row08 Cross-domain agent

Features 03–05 primarily compose Phase 1 primitives (C-46, C-82, C-89, C-23, C-61) with new /v.phase2-* variants (register alongside Phase 1’s variant catalog).

Section 3 · Feature × surface

Where Phase 2 lands in the builder

FeaturePlan / briefComposerEditor canvasSettings / adminReporting
01 Orchestrator
02 Performance loop
03 Send-time personalization
04 Asset studio
05 Localization
06 AMP email
07 Shield
08 Cross-domain agent

● primary surface · ○ secondary touches. Heatmap is directional for staffing, not a maturity score.

Section 4 · Capability definitions

Eight Phase 2 features (Canva AI → Mailchimp)

01

Agentic Campaign Orchestrator

Plan graph · human checkpoints · audit trails

Freddie proposes multi-step graphs (brief → creative → audience → schedule). Each edge requires explicit approval; exports an audit row for Shield. Uses C-98 plus existing composer strategy fields.

Mailchimp wedge

Campaign objects already live in Mailchimp — orchestration attaches to real sends, not slides.

02

Real-Time Campaign Performance Loop

Live KPI strip · anomaly → chat suggestion

While editing or reviewing an active send, a C-99 strip shows cohort KPIs. On regression, Freddie proposes a scoped edit with preview diff (reuses C-43/v.preview-diff).

03

Send-Time AI Personalization

Assembly matrix · policy suppress with explain

Dimensionality (affinity, tier, locale) maps to content cells. Blocked cells surface rule ids (R8-policy). Simulator uses profile fixtures from audience service.

04

Cross-Channel Asset Studio

Master asset → derivative crops

Exports align to channel safe zones; ties into Universal Content / Content Studio roadmap. Changing the master marks derivatives stale with batch refresh.

05

Compliance-Aware Localization

Locale packs · mandatory footers · tone profiles

Packs combine legal text, price/promo rules, and Brand Voice per region. Hard stops prevent schedule when mandatory blocks missing.

06

AMP for Email

AMP components + static fallback pairing

C-100 wraps validated AMP snippets; publishing emits multipart structure with paired static HTML per best practice.

07

Mailchimp Shield (governance suite)

PII · claims · approvals matrix

C-101 aggregates scanners + human roles. Designed for enterprise procurement stories (not bolt-on “AI trust badges”).

08

Cross-Domain Conversational Agent

MCP scopes · Shopify/helpdesk handoff

Extends Phase 1 MCP with C-102 tool registry rows and per-domain credentials. Same campaign graph surfaces in support context.

Section 5 · Variant register (excerpt)

Notable variants on existing baselines

Variant IDBaseFeatureNote
C-43/v.performance-suggestionAssistant body02Embeds KPI delta + proposed action
C-46/v.personalization-matrixGoal / strategy fields03Highlights suppressed cells
C-23/v.locale-blockerBanner05Hard-stop with pack id + fix CTA
C-90/v.scoped-mcp-toolsToken / integrations row08Lists enabled external tools
C-91/v.shield-sectionSettings popover07Labeled Shield policies + approvers
Section 6 · Execution

Build sequence + validation gates

6.1Recommended build order

#StepFeaturesRationale
1Shield policy engine + audit store07, 01Orchestrator checkpoints require durable policy evaluation
2Localization packs + mandatory blocks05Unblocks international sends before personalization expands
3Live metrics ingress + C-99 strip02Feeds the refinement loop early
4Assembly-time personalization engine03Uses packs + Shield outcomes
5Asset derivative graph04Depends on stable asset IDs from Phase 1 content surfaces
6AMP validator + MIME publisher06Isolated pipeline; pair with rendering tests
7Cross-domain MCP registry08External OAuth + scoped tokens last

6.2Gates (block GA)

Cross-refs: Phase 1 execution · Design Audit · Canva-gap analysis