ALL MEMOS Download .docx

Plan: INNERVERSE Release — Build Backwards, UI First

Context

Harnoor instruction 2026-04-19: "new release of innerverse first and build backwards · ui first usage first · then backend · reduce friction of use · make it so every person in the world has it in many forms · its so seamless"

This plan applies Amazon working-backwards to INNERVERSE. The PRFAQ is the anchor. Every task here exists because a specific PRFAQ claim depends on it.

The build order is deliberate: UI and usage validation come BEFORE any backend wiring. That order reduces risk (you find UX problems with a click-through in hours, not weeks) and enforces friction-reduction (you literally cannot ship a signup wall because the UI isn't wired to a backend that could enforce one).

What must be true on launch day (reverse-engineered from PRFAQ)

The press release claims will be honest on 2026-05-03 if and only if all of these hold:

| # | PRFAQ claim | Must be true |

|---|---|---|

| 1 | "Open the URL and you are already in the Mirror" | Zero-signup first-session flow deployed |

| 2 | "No email, no password, no credit card, no onboarding screen" | No forms, modals, or tutorials in v1 |

| 3 | "The conversation starts immediately" | First message input is on the page at load time |

| 4 | "Built on a carefully authored system prompt" | System Prompt v1.0 validated 6/7 on test conversations |

| 5 | "Mirror doesn't flatter. It helps you see." | Real beta user quote collected |

| 6 | "Free forever at the Mirror tier" | No paywall, no credit card, no friction anywhere in v1 |

| 7 | "No app to download" | Mobile-first PWA, not native |

| 8 | "Available worldwide today" | CloudFront global distribution live |

| 9 | "Send this to someone" button | One-click anonymous share flow working |

| 10 | Privacy + crisis resources page | Static page + 988 link |

Anything that is NOT on this list is not v1.

---

Build order (reverse from launch)

← Launch day (2026-05-03) — Day 20

• Social post from Harnoor: 1 post, silentinfinity.com link, one sentence
• Monitor: CloudWatch + simple session count, no heavy analytics
• Crisis protocol and 988 link visible
• 3 beta quotes live on the site (if launch page has them)

← Pre-launch (2026-05-01 → 2026-05-02) — Days 18-19

• Final 3 beta sessions → quotes collected
• Privacy Policy + Terms + Crisis resources page live
• Friction audit: time from URL → first message ≤ 3 seconds on 4G, ≤ 3 taps
• Load test: 100 concurrent Mirror sessions sustained (Bedrock throttle + Lambda concurrency check)

← Backend wire + beta (2026-04-26 → 2026-04-30) — Days 13-17

• API Gateway + Lambda → Bedrock InvokeModel w/ Sonnet 4.6 + System Prompt v1.0 + Guardrails
• Lambda streams SSE back to the UI (already built in Phase 2) so the UX is identical to the static click-through
• Optional Cognito passwordless magic-link (email only, never blocks first session)
• Stripe checkout page for $29/mo Path tier (inactive — just the URL)
• 3 invited beta users do real sessions, quotes collected
• Anonymous share-link endpoint (creates a new session UUID, no account needed)

← Usage validation (2026-04-24 → 2026-04-25) — Days 11-12

• 1–3 real humans sit down with the static UI. They type. A canned script (or Harnoor typing on the other side) produces responses. This is a Wizard-of-Oz test — we measure the UX, not the model.
• Collect friction data: where did they hesitate, misread, tap twice?
• Iterate the UI based on observations (no backend work yet)
• GO/NO-GO gate: if the UI produces silence or confusion in >1 of 3 testers, iterate, do not proceed to backend

← UI click-through (2026-04-20 → 2026-04-23) — Days 7-10 (starts tomorrow)

• Next.js + Amplify scaffold on silentinfinity.com
• Zero-signup entry: URL loads → input field focused → one placeholder sentence
• Conversation UI: user bubble + mirror bubble + streaming dots. No header, no footer, no nav.
• "Send this to someone" button: one-tap share-sheet with a pre-filled anonymous link
• Styling: Co-Star × Aesop × A24 — deep cosmic blue-black, liquid gold accents, Söhne/Tiempos type
• No backend yet. Messages stored in localStorage; mirror responses are canned (5–10 hand-authored replies for the demo).
• Mobile-first: tested on iOS Safari and Android Chrome before desktop
• Deploys to silentinfinity.com via Amplify CI on every push
• End-of-phase deliverable: URL that a real human can open and use for 5 minutes without a backend

← Today (2026-04-19) — Day 6 — Prompt validation (Move 1)

• Run System Prompt v1.0 against 7 scripted scenarios in Claude directly (no Bedrock, no AWS)
• Pass criteria: 6/7 warm, boundaried, non-flattering responses
• Author canned replies for the UI click-through from the best outputs
• If <6/7: iterate prompt TODAY before any build starts
• Deliverable: F:/TITAN/knowledge/research/innerverse-prompt-validation-2026-04-19.md

---

Forward execution order (same plan, chronological)

| Day | Date | Phase | Owner | Deliverable |

|---|---|---|---|---|

| 6 | 2026-04-19 (today) | Prompt validation | SCOUT (dispatched) | innerverse-prompt-validation-2026-04-19.md with 6/7 pass |

| 7 | 2026-04-20 | UI scaffold begins | FORGE / Harnoor | Next.js + Amplify shell on silentinfinity.com |

| 8 | 2026-04-21 | UI click-through | FORGE | Conversation UI rendered, canned replies wired |

| 9 | 2026-04-22 | UI polish + mobile | FORGE | Mobile-first test pass on iOS + Android |

| 10 | 2026-04-23 | Share + friction audit | FORGE | Share button working; friction ≤ 3 taps verified |

| 11 | 2026-04-24 | Usage test 1–3 | Harnoor + 1 friend | Observations doc; iteration list |

| 12 | 2026-04-25 | UI iteration | FORGE | Fixes from usage test applied |

| 13 | 2026-04-26 | Backend scaffold | FORGE | Lambda + API GW + Cognito (optional) deployed |

| 14 | 2026-04-27 | Bedrock wire | FORGE | Sonnet 4.6 streaming responses end-to-end |

| 15 | 2026-04-28 | Guardrails + crisis | FORGE | Crisis keywords → 988 page; Guardrails on |

| 16 | 2026-04-29 | Share link backend | FORGE | Anonymous session UUID + share URL live |

| 17 | 2026-04-30 | Beta invite 1 | Harnoor | Beta user 1 completes session; quote collected |

| 18 | 2026-05-01 | Beta 2 + legal | Harnoor | Privacy/Terms live; beta user 2 done |

| 19 | 2026-05-02 | Beta 3 + load test | FORGE + Harnoor | Beta user 3 done; 100 concurrent sessions held |

| 20 | 2026-05-03 | LAUNCH | Harnoor | Social post live; silentinfinity.com public |

---

The friction audit (runs at every phase)

Before any merge, these questions are answered YES:

• [ ] Does the user need to type anything to START the conversation? (Must be NO)
• [ ] Does any field have a placeholder that requires the user to read it to proceed? (NO)
• [ ] Is there a modal, popup, tooltip, or overlay on the first session? (NO)
• [ ] Does anything load between the URL tap and the input being focused? (NO, <1s cached shell)
• [ ] Is there a "Sign up to continue" anywhere in v1? (NO)
• [ ] Is there a paywall anywhere in v1? (NO)
• [ ] Is there a streak, badge, level, XP, reward, or other gamification? (NO)
• [ ] Is there a push permission request? (NO)
• [ ] Does the first session REQUIRE the user's name, age, or any field? (NO)
• [ ] Can they share the experience with one tap? (YES)

If any answer flips, stop and refactor before shipping.

---

Distribution roadmap (post-launch — "many forms, every person")

Referenced from PRFAQ Q6b. These are downstream of v1 but flagged now so Phase-1 code does not block them.

| Phase | Form | Lead time | Key dependency |

|---|---|---|---|

| v1 | Web PWA (launch) | — | This plan |

| v1.1 | Anonymous share links | v1 day 1 | Already in v1 scope |

| v2 | SMS / WhatsApp companion | +60d | Twilio + WhatsApp Business API |

| v3 | Voice phone line | +120d | Twilio Programmable Voice + Eleven Labs |

| v4 | iMessage app + widget embed | +180d | Apple dev account + JS embed SDK |

| v5 | Slack / Discord / Teams bots | +240d | Bot framework per platform |

| v6 | iOS / Android native apps | +270d | Expo or RN wrapper around the PWA |

| v7 | Smart-mirror + car integrations | Year 2+ | Hardware partnerships |

Architectural rule (reduces friction across forms): the Mirror is ONE Bedrock-backed service exposed via ONE API. Every form-factor is a thin adapter. The prompt, memory model, and safety guardrails are shared. No fork, ever.

---

Risks

| Risk | Mitigation | Owner |

|---|---|---|

| Founder cycles to a new empire before Day 14 | PRFAQ + this plan are the pre-commitment | Harnoor |

| System Prompt fails Move 1 validation today | Iterate prompt same-day; Layer I depends on it | SCOUT → Harnoor |

| Usage test on Day 11 shows UI doesn't work | Iterate UI Days 12–13, compress backend phase | FORGE |

| Bedrock throttles at launch | Pre-warm Lambda concurrency; Sonnet 4.6 has high RPM | FORGE |

| Founder-face commitment not made by Day 12 | Decision gate Day 12; if no, launch as anonymous-brand | Harnoor |

| Launch day sees 0 traffic | Harnoor social post is the sole distribution primitive — that post must be authentic + Day 12 drafted | Harnoor |

---

Parallelizable tracks

Track A (Harnoor):   Founder-face decision → Social post draft → Beta recruitment → Launch post
Track B (FORGE):     UI scaffold → Click-through → Usage-test fixes → Backend → Bedrock → Guardrails → Share link
Track C (SCOUT):     Prompt validation (today) → Beta quote collation → Launch-day analytics spec

Harnoor is the single-threaded owner of founder-face + launch post + beta outreach. Everything else is delegable.

---

Open questions

• Opus 4.6 vs Sonnet 4.6 for v1 Mirror model tier? (Leaning Sonnet 4.6 — 40% cheaper; quality likely sufficient. Test in prompt validation today.)
• Wizard-of-Oz human-on-the-other-side for Day 11 usage test vs canned replies? (Leaning canned + Harnoor observing.)
• Founder-face commitment — decide by Day 12.
• Beta user pool — who are the 3? (Recruit from Harnoor's network Days 13–17.)

---

Change log

• 2026-04-19: initial plan, reverse-engineered from PRFAQ; build-backwards + UI-first + friction-audit + many-forms vision all locked