Phased Plan — Rev 2

Drafted: May 5, 2026
Owner: Blair Williams, CEO
References: MemberIntel SPEC v1, Cindy & Seth JDs, Decision-Rights Matrix, Phased Plan v1, May 2026 architecture working sessions
Supersedes: phased-plan.md (v1, April 30 2026)

---

What changed from v1

V1 of this plan was built from the SPEC and the JDs. Rev 2 incorporates the architectural commitments worked through in the May infrastructure sessions — GCP project structure, per-tenant isolation, cross-pollination boundary, CI/CD, observability, cost controls, auth, sync pipeline, secrets management, and the AI eval suite. Most of the changes are additive: new milestone items inside existing phases. Three changes are structural and worth reading the friction-points one-pager (phase-1-2-friction-points.md) before signing off:

1. A brain-authoring content lead is recommended for Phase 1, separate from Sarah Olaleye’s marketing/launch content role in Phase 3. This is one additional hire vs the v1 plan.
2. Phase 2 scope is calibrated honestly. “Initial chat advisor” in Phase 2 means a route that takes a question and returns a Haiku answer — no citations yet, no eval coverage yet. Polish is Phase 3. V1’s milestone language was optimistic about a brand-new Senior AI Engineer’s ramp speed.
3. The differentiation eval subset is a Phase 3 milestone, not a Phase 5 one. If the gap-vs-baseline isn’t widening by month 3, the team has time to course-correct. If it isn’t measured until Phase 5, the discovery is too late.

Everything else is additive. The phase shape, team ramp, and GA target (mid-October 2026) are unchanged.

---

Phase Overview (unchanged from v1)

Phase	Months	Team Size	Focus
1 — Scoping & Foundation	May 2026	2 (3 with brain content lead)	Customer interviews, PRDs, architecture, privacy counsel, foundation infra
2 — Build Kickoff	June 2026	3–4	Engineering core forms; pipelines start
3 — Design & Content	July 2026	5–6	Design + content workstreams; differentiation eval baseline
4 — Website + Beta	August 2026	6–7	Site build; closed beta opens
5 — Launch Ramp	September – mid-October 2026	7+ PR firm	All-hands GA prep
6 — Post-Launch	November 2026+	4 ongoing	Matrixed contributors taper back

---

Phase 1 — Scoping & Foundation (May 2026, ~4 weeks)

Dedicated team (2, recommended 3)

• Cindy Thoennessen — Product Lead, MemberIntel
• Seth Shoultes — Lead Architect, MemberIntel
• Brain-authoring content lead (recommended new hire — see friction-points one-pager). Note: this is distinct from Sarah Olaleye, who joins Phase 3 for marketing/launch content.

Cindy’s deliverables (unchanged from v1, except as noted)

• 10–15 customer discovery interviews completed (per SPEC §13 Phase 0)
• Engage outside privacy counsel by June 1 (hard deadline — if this slips, GA slips)
• NEW: Schedule a substantive architecture review with privacy counsel + Seth in late May, not deferred to Phase 3. Counsel’s input shapes the cross-pollination boundary, the audit dataset, and the secrets management approach.
• PRDs drafted, in priority order:
  • Free → Pro upgrade flow + chat experience (treat as one PRD — they’re intertwined)
  • Free vs Pro entitlement matrix (every feature × every tier — input to Seth’s entitlement service)
  • Dashboard insight-card system
• Beta cohort selection criteria proposal
• Decision-rights matrix signed off by Cindy, Seth, and Blair
• Cross-functional partner kickoff meetings (Curt, Wray, Russ, Thomas, Paul C, Ally, Danielle) scheduled
• NEW: Raise the meta-decisions from the architecture risk review with Blair: brain-content lead hire timing, differentiation eval as a tracked metric, infrastructure-leaning hire on the V1.5 roadmap, SOC 2 placeholder on the calendar.

Seth’s deliverables (substantially expanded from v1)

Architecture decisions land first. By end of week one of Phase 1, the following are decided and committed as Architectural Decision Records (ADRs):

• Hosting: GCP, project-per-environment with four projects (memberintel-prod, memberintel-staging, memberintel-dev, memberintel-shared) under a Google Cloud Organization with production and non-production folders.
• Database: Cloud SQL Postgres with CMEK enabled.
• Per-tenant isolation: shared-schema with Row-Level Security, FORCE on every customer-data table, dual database roles.
• Vector store: pgvector (resolves SPEC Open Q9).
• CI/CD: GitHub Actions with Workload Identity Federation, separate pipelines for code, infrastructure (Terraform), and database migrations.
• Observability: Cloud Logging + Cloud Trace + Cloud Monitoring + OpenTelemetry instrumentation in app code; BigQuery for analytics; locked-down BigQuery dataset for audit.
• Secrets management: Secret Manager with KMS-backed encryption, path-prefixed naming convention (system/, customer/, app/), 90-day rotation cadence.
• Stripe integration: customer-OAuth-only in V1 (resolves SPEC Open Q4).
• Hosting choice (resolves SPEC Open Q8 in favor of GCP).

Schema design. The canonical data schema is committed as a markdown document with ER diagram and column-level annotations by end of Phase 1. Includes:

• Platform-agnostic fields (source_platform, source_id) on canonical tables, even though only MP exists in V1, so V2.1 platform abstraction is additive rather than a refactor.
• NEW: Per-customer brain entries are versioned from day one (every update is a versioned write, prior versions retained).
• Audit log structure.
• Entitlement table shape with V1.5 trial-state extension fields present in V1.

Foundation infrastructure scaffolding. By end of Phase 1, the GCP projects exist, Workload Identity Federation is configured, the KMS keyring is set up with separate keys for app-data, secrets, backups, and audit. No application code yet — this is Terraform scaffolding that Phase 2 builds on.

Vendor decisions logged. Cloud Tasks + Cloud Scheduler + Cloud Run for the worker stack. Atlassian Statuspage or Better Stack Status for the status page. Anthropic API directly per SPEC, with NEW: a thin model-provider abstraction layer wrapping the SDK (one day of work in Phase 1; one-month refactor if deferred and a strategic question forces a second provider).

MP-side API surface confirmation with Paul Carter. The dedicated MemberIntel REST endpoint approach (/wp-json/memberintel/v1/sync and sub-endpoints), with versioning and incremental-sync support (since=<timestamp> parameter), needs Paul Carter’s team’s commitment by end of May. Otherwise the sync pipeline is blocked in Phase 2. Resolves SPEC Open Q7.

Lightweight prototyping to de-risk material architectural choices that ADRs alone won’t validate. Specifically: a throwaway prototype of the RLS isolation pattern with two synthetic tenants, confirming the FORCE policy, the dual-role pattern, and the middleware-enforced tenant context all behave as expected. One day of work; pays back the first time anyone questions the isolation story.

Senior AI Engineer hiring. Search active. Danielle running cultural/operational fit interviews. NEW: Interview process tests for cloud infra comfort, not just AI engineering — by V1.5 this person is sharing ops load with Seth.

Cross-functional engagement (light, not full-time)

• Santiago — sets up GitHub-based ticket tracking for MemberIntel
• Danielle — Senior AI Engineer search posted, cultural fit interview process defined
• Paul Carter — confirms MP-side API endpoint approach for V1 sync
• Outside privacy counsel — engaged by June 1, with substantive architecture review scheduled in late May

Phase 1 milestone (gate to Phase 2)

• ☐ ADRs committed for hosting, database, isolation, vector store, CI/CD, observability, secrets, Stripe integration, model abstraction
• ☐ Schema design reviewed and approved by Blair
• ☐ At least 2 PRDs drafted and approved by Blair
• ☐ Privacy counsel engaged by June 1; architecture review held
• ☐ Decision-rights matrix signed
• ☐ Senior AI Engineer hire pipeline has 3+ qualified candidates in late-stage interviews
• ☐ MP API surface approach confirmed with Paul Carter
• ☐ GCP projects, KMS keyring, WIF, and basic Terraform scaffolding deployed
• ☐ RLS isolation prototype validated
• ☐ Brain content lead hire decision made (yes/no, and if yes, candidate identified)

If these aren’t met by May 31, do not kick off Phase 2. Slip the schedule rather than pull people into work that isn’t ready.

---

Phase 2 — Build Kickoff (June 2026, ~4 weeks)

Dedicated team grows to 3–4

• Cindy + Seth (continuing)
• Brain content lead (continuing if hired in Phase 1)
• Ronald Reymundo — joins MemberIntel as primary engineering focus
• Senior AI Engineer (new hire, ramping — target close mid-June, productive by July)

Engineering scope (calibrated from v1)

Infrastructure foundation completion (weeks 1–2). Terraform modules for the VPC, Cloud SQL with CMEK, Cloud Run services, Cloud Tasks queues, Secret Manager structure, Cloud Logging sinks, BigQuery datasets including audit. The CI/CD pipeline with the RLS test harness as a required PR check. The kill-switch infrastructure for LLM calls, cross-pollination, agent (V1.5+), and eval suite. Two engineers focused, with Claude Code doing scaffolding and Seth reviewing.

Schema and migrations (weeks 1–2). Tables created with the migration tooling. RLS policies with FORCE on every customer-data table. Two database roles. Middleware for tenant context. Integration tests that verify isolation, running on every PR.

Auth layer (weeks 2–3).

• MP-license-based authentication via per-license signing keys, banner-click flow.
• Unified user model with multiple identity sources (single user_id, multiple linkable credentials).
• Standalone email/password fallback (Argon2id, NIST 800-63B password rules, server-side sessions, optional TOTP MFA).
• Account-merge prevention via email-matching at link time.

Entitlement service skeleton (weeks 2–3). Postgres tables for tier and trial state, Memorystore for hot quota counters, the check_and_consume API. Tier model wired up to Cindy’s Free vs Pro entitlement matrix from Phase 1. Trial state fields present in V1 even though V1.5 doesn’t ship yet (additive, not a refactor in V1.5).

MP sync pipeline (weeks 3–4). Cloud Tasks queue, per-customer concurrency=1, basic timeouts (full adaptive logic deferred to Phase 3). Pulls from Paul Carter’s team’s MP-side endpoint with versioning. Writes to per-customer warehouse with RLS context set. Schema validation on responses.

Stripe sync pipeline (weeks 3–4). Customer-OAuth flow with read-only scopes. Refresh token storage in Secret Manager keyed by tenant_id. Webhook listener with signature verification (handles multiple active signing keys for rotation overlap). Idempotent event processing via stripe_events_processed table per tenant.

Initial chat skeleton (week 4). A route exists. Takes a question, calls Claude Haiku with a basic prompt and minimal retrieval (search_global_brain only, since the global brain seed lands in Phase 3 with the content lead). Returns a response. No citations, no eval coverage, no per-customer brain integration yet. This is honest about what “initial chat” means in 4 weeks with a brand-new senior hire ramping.

Cindy’s scope (Phase 2)

• Continue PRD authoring (cross-pollination flow, weekly digest, advanced reports)
• Beta cohort outreach begins
• Privacy compliance work in flight (ToS / DPA / Privacy Policy drafts with counsel)
• Brand identity decision (with Blair approval, with Russ’s input)
• Marketing site planning (begin)

Phase 2 milestone (gate to Phase 3)

• ☐ Auth working end-to-end: MP customer can click banner, sign in to MemberIntel
• ☐ Data ingestion working: a real MP customer’s data syncs into the per-customer warehouse with RLS isolation verified
• ☐ Senior AI Engineer onboarded and contributing
• ☐ Initial chat skeleton: a route returns a Haiku response (citations, eval, per-customer brain → Phase 3)
• ☐ Entitlement service operational with V1.5-ready trial state fields
• ☐ Kill-switch infrastructure deployed and tested in staging
• ☐ ToS / Privacy Policy drafts in legal review

---

Phase 3 — Design & Content (July 2026, ~4 weeks)

Dedicated team grows to 5–6

• Cindy + Seth + Ronald + Senior AI Engineer (continuing)
• Brain content lead (continuing if hired in Phase 1)
• Meo LaTao — joins MemberIntel full-time for design through Phase 5
• Sarah Olaleye — joins MemberIntel full-time through Phase 5 for marketing site copy, launch content, AEO content, customer education content, and email campaigns (NOT brain authoring — that’s the dedicated brain content lead’s role)

Design scope (Meo, with Russ design-system review)

Same as v1: dashboard mockups, brand visual identity, marketing site mockups, chat panel UI, brain editor UX, tier upgrade-prompt surfaces, in-MP-admin banner integration design.

Brain content scope (brain content lead)

• 30+ hand-written playbooks committed by end of Phase 3, on track to 50+ by GA per SPEC §13
• Indexed MP docs + KB + blog seeded into global brain
• Playbook authoring tooling defined (markdown source, vector indexing pipeline, content review workflow)
• Cross-pollination opt-out copy and flow drafted (the job itself doesn’t run until Phase 4 when there’s beta data)

Marketing/launch content scope (Sarah, with Curt’s brand-consistency review)

• Marketing site copy
• Launch blog series (5–10 posts)
• AEO-optimized content (overlaps with Rock 1)
• Customer education: FAQs, knowledge base, onboarding email sequence
• Email campaigns for existing customer base
• Post-launch case studies from beta cohort (begin in Phase 4)

Engineering scope (Phase 3)

Chat experience polish.

• Citations enforced in prompt and validated in eval suite
• Feedback capture (thumbs up/down) wired to audit dataset
• Tool calling stabilized: query_customer_metrics, search_global_brain, search_customer_brain, update_customer_brain, analyze_site all working
• Sonnet routing for Pro paths, Haiku for Free, server-side enforced via entitlement layer
• Per-customer brain integrated into chat retrieval

Eval suite — first deliverable (this is new in Rev 2 as a Phase 3 milestone).

• ~80–100 scenarios at end of Phase 3, distributed across categories: chat-grounding, recommendation quality, MP-specific knowledge, refusal/scope, citation discipline, tool calling, tier-routing safety
• Judge model (Sonnet) scoring with rubric per criterion
• NEW: Differentiation subset (30–50 scenarios) shipped with baseline measurement against vanilla Sonnet. This is the leading indicator for the SPEC’s #1 risk. Reviewed monthly with Blair starting Phase 3.
• Eval suite versioned, runnable locally, integrated into CI for PRs that touch LLM paths

Self-serve billing. Stripe integration with Ally — plan upgrades, downgrades, cancellation, failed-payment dunning. Tested across trial-to-paid, trial-to-failed, Pro-to-Free, Free-to-Pro paths.

Public site analysis pipeline. Weekly cached for both tiers, on-demand for Pro with daily rate limit. The cost-discipline target: $0.20/customer/month per SPEC §5.4. Cost circuit breaker integrated.

Customer-facing sync state visibility. Settings panel shows last sync, status, error message per pipeline. Without this, sync failures become silent churn.

Standard dashboard with tier-gated cards. Free shows monthly snapshot with teaser cards; Pro shows live data with full insight prose.

Cindy’s scope (Phase 3)

• PRDs continue (cross-pollination flow, weekly digest, advanced reports, downgrade flow)
• Beta cohort identification finalized
• Marketing site coordination intensifies
• PR plan: decide internal vs outside agency by end of July; if outside, source and engage
• NEW: Quarterly architecture review #1 scheduled for early Phase 4 (Cindy + Seth + Blair + privacy counsel)

Phase 3 milestone (gate to Phase 4)

• ☐ Chat experience working with Haiku (Free) and Sonnet (Pro) routing, citations enforced
• ☐ Dashboard visible with tier-gated cards
• ☐ Brand identity locked
• ☐ Marketing site copy approved
• ☐ Beta cohort identified (10–20 customers, mix of Free + Pro)
• ☐ Outside PR agency engaged (if going that route)
• ☐ Differentiation eval subset shipped with baseline measurement
• ☐ 30+ playbooks committed, brain content quality reviewed by content lead

---

Phase 4 — Website + Beta Launch (August 2026, ~4 weeks)

Dedicated team grows to 6–7

• Cindy + Seth + Ronald + Senior AI Engineer + brain content lead + Meo + Sarah (continuing)
• Kalpesh Baghat — joins MemberIntel through Phase 5 for website / microsite build

Website scope (Kalpesh + Meo + Sarah)

Same as v1: memberintel.com build, pricing page, in-MP-admin banner integration, marketing site sections, honest copy per SPEC §4.

Engineering scope (Phase 4)

Cross-pollination job goes live. This is the right time because by Phase 4 the beta cohort has real per-customer brain entries. The three-roles model from the security architecture:

• cross_pollination_role — read-only access to per-customer brains, write access to staging table only
• candidates_role — staging table read/write
• content_lead_role — review and approval, writes to global brain
• k-anonymity floor of N=5 customers exhibiting a pattern before it’s eligible
• Per-customer consent versioning enforced at job-start time
• Monthly cadence during V1, weekly post-launch when steady
• Runs on Cloud Run Jobs with dedicated service account, separate from the API

Weekly digest email generation. Haiku for Free, Sonnet for Pro. Hooks into Caseproof’s existing transactional email infrastructure.

AI eval suite to ~150 scenarios. Nightly drift detection in staging-as-prod. Differentiation subset reviewed monthly with Blair. Eval failures generate Slack alerts.

Cost monitoring dashboards. Per-customer token caps live. Daily global cost circuit breaker active. Weekly cost-per-cohort dashboard for Cindy.

Privacy controls. Data export (CSV/JSON), deletion pathway, consent flow at signup, cross-pollination opt-out surface — all production-quality before beta customers test them.

i18n shipping. English (default), Spanish, German per SPEC §11. UI translation, AI chat handling input/output language.

Performance hardening, observability tightening. OTel instrumentation now showing real production-shape requests; latency baselines emerge; alert thresholds tuned from real data rather than guesses.

Closed beta launches (mid-Phase 4). Per SPEC §13: 10–20 hand-picked beta customers across Free and Pro tiers. Weekly feedback collection cadence. Bug triage workflow with engineering. Feedback-loop refinement based on findings.

Cindy’s scope (Phase 4)

• Beta operations daily
• Sales + support enablement materials in draft
• Curt + Sarah finalize launch content + email campaigns
• Press release + press kit drafted (with PR agency if engaged)
• Quarterly architecture review #1 with Seth, Blair, privacy counsel

Phase 4 milestone (gate to Phase 5)

• ☐ Closed beta running with weekly feedback flowing
• ☐ memberintel.com live
• ☐ AI eval suite passing on representative scenarios; differentiation gap widening vs Phase 3 baseline
• ☐ Cost-per-user dashboards operational
• ☐ Privacy compliance posture clean (counsel sign-off)
• ☐ Cross-pollination job running monthly with content lead review queue functional
• ☐ Sales + support enablement materials drafted
• ☐ Quarterly architecture review #1 completed

---

Phase 5 — Launch Ramp (September – mid-October 2026, ~6 weeks)

Full team (7) + outside PR agency engaged

All members from Phase 4 plus:

• Outside PR agency (3-month engagement, ~$5–10K/mo)
• Curt Noble (matrixed) — paid acquisition campaigns ramp; GA4 / Meta Pixel / server-side analytics live
• Wray Cullimore (matrixed) — CS team trained on MemberIntel; support runbooks live

GA prep (per SPEC §13 Phase 2, with architectural additions)

Operational readiness.

• Status page configured before GA, integrated with Cloud Monitoring
• Runbooks written for the 5 highest-stakes failure modes:
  • API down
  • Cloud SQL down
  • RLS violation detected
  • Cross-pollination leak suspected
  • Payment failure spike
• On-call rotation: Seth + Senior AI Engineer primary, Ronald secondary, Cindy customer-facing escalation
• Kill switches tested in staging by every team member
• Rollback procedure tested end-to-end (under 60 seconds for code rollback via Cloud Run revisions)

Final security and compliance review.

• Privacy counsel architecture sign-off (this is a “review” not “first look” — counsel has been engaged since June)
• Audit log queryable; 90-day query exercise completed
• Encryption-at-rest verified (Cloud SQL CMEK, audit dataset, backups)
• Secret rotation cadences documented
• Data deletion pathway tested end-to-end with synthetic tenant

Final eval suite full run.

• ~150 scenarios passing at threshold
• Differentiation subset showing measurable advantage over baseline LLM
• Hallucination rate <1% on financial-data scenarios per SPEC §8.4

Final cost projections at projected adoption volumes.

• Stress-tested at 5K, 10K, 50K free users per SPEC Phase 0 deliverable
• Cost-per-Free-user grounded in beta cohort data, not just SPEC estimates
• Daily global circuit breaker threshold reviewed and tuned

Launch comms and content (per v1). Pre-launch email sequence to MP customer base, paid acquisition starting mid-September, press embargo and journalist outreach, launch-day social/blog/email coordination.

Launch operations (per v1, with additions).

• Day-one go-live checklist owned by Cindy
• Rollback plan documented (different from technical rollback — this is the “should we pull the launch” decision tree)
• Customer comms templates pre-written (Cindy posts with one click)
• Internal launch comms — Caseproof team briefed week of launch
• Status page operational

Phase 5 milestone (the launch)

• ☐ Mid-October: V1 GA shipped publicly
• ☐ Free + Pro tiers live
• ☐ MP customer base notified
• ☐ Press coverage secured
• ☐ Eval suite differentiation gap defensible vs baseline LLM
• ☐ Cost-per-Free-user tracking against SPEC target ($1.10/mo or below)

---

Phase 6 — Post-Launch (November 2026 onward)

Team taper (mostly unchanged from v1)

Person	Status post-launch
Cindy Thoennessen	Stays dedicated — ongoing Product Lead
Seth Shoultes	Stays dedicated — ongoing Lead Architect
Ronald Reymundo	Stays dedicated — ongoing engineer
Senior AI Engineer	Stays dedicated — ongoing engineer
Brain content lead	Stays dedicated — ongoing brain authoring + cross-pollination review + eval scenario authoring
Meo LaTao	Tapers back to Russ’s team late October if dashboard design is locked; some matrixed work continues
Sarah Olaleye	Tapers back to Katelyn’s team late October if launch content is shipped; some matrixed work continues
Kalpesh Baghat	Tapers back to Thomas / IPJ in November if site is stable
PR agency	Engagement ends late October

Steady-state team (5 dedicated, up from v1’s 4)

Cindy + Seth + Ronald + Senior AI Engineer + brain content lead.

Post-launch focus (per SPEC §13 V1.5 prep, with additions)

Performance and conversion (per v1).

• Free → Pro conversion rate monitoring; rapid A/B iteration
• Cross-pollination job tuning based on real customer brain content
• Cost monitoring at real volumes (5K, 10K, 50K Free users)
• Beta cohort grad → paying Pro conversion analysis
• Quarterly free-tier review per SPEC Open Q14
• V1.5 planning (agent / write-actions, greenfield onboarding wizard)

Architectural cadence (new in Rev 2).

• Quarterly architecture review with Cindy + Seth + Blair + privacy counsel
• Monthly differentiation-eval review with Blair (the SPEC’s #1 risk indicator)
• 90-day secret rotation cadence executed
• Eval suite versioning maintained; new scenarios added from production thumbs-down review

V1.5 hiring planning (new in Rev 2).

• An infrastructure-leaning engineer is the next hire after Senior AI Engineer, in V1.5 ramp rather than deferred to V2. Once the agent ships in V1.5, two engineers carrying ops on top of feature work is a burnout risk.

SOC 2 placeholder (new in Rev 2).

• Q3 2027 review of SOC 2 Type 1 readiness. The architecture we’ve built (CMEK, audit logs, IAM discipline, encryption-at-rest, secrets management) makes SOC 2 much easier when the time comes, but doesn’t make it free — 3–6 months of work once the decision is made.

---

Critical assumptions (the load-bearing ones — updated from v1)

1. Senior AI Engineer hire closes mid-June. Phase 2 build pace depends on it. (Unchanged from v1.)
2. Phase 1 architecture decisions are made in 4 weeks. ADRs, schema, infrastructure foundation, MP API surface, prototype validation. If May becomes 6 weeks, everything downstream slides.
3. Russ and Thomas are notified by mid-May that Meo and Kalpesh are going full-time on MemberIntel July–October. (Unchanged from v1.)
4. Blair holds the line. When Russ has competing demands for Meo, or Thomas for Kalpesh, Blair tie-breaks toward MemberIntel during their dedicated windows. (Unchanged from v1.)
5. Privacy counsel engagement is non-negotiable by June 1, with substantive architecture review in late May. (Strengthened from v1.)
6. Phase 1 is 2 or 3 people. No pulling in additional resources during scoping beyond the optional brain content lead. Resist the urge.
7. NEW: Brain content lead in Phase 1 is the recommended structure. If Blair declines this hire, the SPEC’s #1 risk (differentiation from generic AI) becomes materially harder to mitigate. The friction-points one-pager covers this in detail.
8. NEW: Phase 2 milestone language is calibrated. “Initial chat” means a Haiku route returns a response. No citations, no eval coverage, no per-customer brain integration in Phase 2. Polish is Phase 3.

---

Cross-functional matrix (unchanged from v1, with one addition)

Person	Role	Approximate load
Santiago Perez Asis	Cross-Caseproof PM (incl. MemberIntel sprint tracking)	20–30%
Curt Noble	Marketing positioning, AEO, paid acquisition, brand consistency	Light through Phase 3, heavier Phase 4–5
Wray Cullimore	Support enablement	Light throughout, ramping at launch
Carl Jessee	Educational video content	Light throughout
Russ Williams	Design system / brand consistency review (Meo’s manager)	Light throughout
Thomas Levy	Data layer engineering consultation; Kalpesh’s manager	Light throughout, ad hoc
Paul Carter	MemberCore data access; MP-side endpoint development for sync (Phase 1–2 substantive)	Heavier Phase 1–2, light afterward
Ally Roger	Stripe billing, dunning, plan upgrades/downgrades	Heavier Phase 2–4
Mahmoud Saeed	Senior engineering / code review backstop (ad hoc)	Only if needed
Danielle Lea-Jones	Senior AI Engineer hire — cultural/operational fit + onboarding	Phase 1–2
Outside privacy counsel	Compliance workstream, substantive architecture review in late May	Phase 1–4, heavier Phase 1–2
Outside PR agency	PR execution	Phase 4–5 only

---

Risk register (top items per SPEC §14, with architecture-aware mitigations)

Risk	Mitigation in this plan
Senior AI Engineer hire slips	Danielle prioritizes Phase 1; Mahmoud as ad hoc backstop
Phase 1 scoping takes 6 weeks instead of 4	Hard milestone gate; slip schedule rather than start Phase 2 with bad foundations
Free-tier cost spiral (per SPEC Risk #2)	Entitlement layer hard-enforced server-side; per-call max_tokens; daily global circuit breaker; weekly cost-per-cohort review during Phase 5
Free → Pro conversion underperforms (Risk #3)	Multiple upgrade-moment surfaces designed into PRDs; A/B testing plan in Phase 5
Differentiation from generic AI (Risk #1)	Differentiation eval subset shipped Phase 3 with baseline measurement; monthly review with Blair; brain content lead in Phase 1 if approved
Russ or Thomas pull Meo / Kalpesh back mid-window	Blair tie-breaks; written commitment in mid-May
Privacy counsel engagement slips past June 1	Cindy’s Phase 1 explicit deliverable, monitored weekly; substantive review scheduled late May
Cross-pollination privacy slip (per SPEC Risk #10)	Three-roles model with dedicated service account; k-anonymity floor N=5; per-customer consent versioning; content-lead review queue; cross-pollination on dedicated Cloud Run Jobs separate from API
RLS isolation gap	FORCE policies on every customer-data table; dual database roles; middleware-enforced tenant context; isolation tests required on every PR; Phase 1 prototype validation
Anthropic dependency	Thin model abstraction layer in Phase 1; kill switch infrastructure; degradation path designed

---

Open questions to resolve in Phase 1 (subset of SPEC §14)

#	Question	Owner	Resolution by
Q1	Final naming + domain + trademark check	Cindy + Blair	May 15
Q4	Stripe Connect vs customer-OAuth split	Seth + Ally	Resolved: customer-OAuth only in V1
Q5	Customer support model for free users	Cindy + Wray	May 31
Q7	MP-side API surface for V1 sync	Seth + Paul Carter	Resolved: dedicated /wp-json/memberintel/v1/sync endpoint, versioned, supports since= parameter
Q8	Hosting choice (GCP vs Heroku)	Seth	Resolved: GCP
Q9	Vector store choice	Seth	Resolved: pgvector
Q12	Free-tier opt-in flow specifics	Cindy + Blair	May 22
Q13	”Pro for 30 days” launch promo for existing MP customers?	Blair	May 31
Q15	Brand presentation of free tier	Cindy + Blair	May 22
NEW: Brain content lead hire?	Blair (with Cindy + Seth input)	May 12 — friction-points one-pager covers this

---

Document version: Rev 2 — supersedes v1 (April 30 2026). To be reviewed alongside the friction-points one-pager (phase-1-2-friction-points.md) before sign-off by Cindy, Seth, and Blair.