ALL MEMOS Download .docx

Work Distribution — Innerverse Identity + Signup Conversion Funnel

HERALD · 2026-04-20 18:45 UTC

Incoming directive from Harnoor (2026-04-20 18:45 UTC):

• Want session persistence across visits — user comes back, Innerverse remembers them.
• Cookies are likely the right primitive (complements localStorage already in place).
• Need non-creepy identity: "it should ask 'what should I call you?' — not demand an email pre-conversation."
• Multiple users on same device (e.g., family computer) — how to handle profiles?
• Lead gently to signup AFTER some free chat — not a wall up front.
• Free tier duration: "whatever time is suitable per PhD-level research" — don't guess. Research it.
• Conversion path: chat → login → money. "Sensitive topic." Nuance matters.
• User loved the disclaimer + 911/988 change. "Simple to sue" acknowledged → keep defensive posture.

---

Decomposition → agents (primary + QA)

| # | Part | Owner | QA reviewer | Deliverable | Hrs P | Hrs QA | Deadline |

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

| 1 | PhD research: non-creepy identity + device-fingerprint options | SCOUT | ORACLE | knowledge/research/research_innerverse_identity_20260420.md (2000w): IP + UA hash + localStorage + cookies + FingerprintJS + LocalPassport/Passkeys. Privacy/legal posture per jurisdiction (EU cookie law, CCPA, COPPA interplay). Multi-user-on-one-device patterns (profile picker, "who's chatting?"). | 8 | 2 | Fri 18:00 UTC |

| 2 | PhD research: chat → signup → paid conversion funnel timing | SCOUT | ORACLE | knowledge/research/research_innerverse_conversion_funnel_20260420.md (2500w, 15+ sources): data from Character.AI, Replika, Pi.ai, Woebot, Kin, Ash, Duolingo — when they prompt signup vs. leave free-open. Free-chat token budget per anonymous user. Psychological thresholds ("3rd meaningful turn", "emotional beat reached", "10-minute mark", etc). Pricing psychology for reflective apps. Soul-friendly vs hard-sell CTAs. | 10 | 2.5 | Fri 18:00 UTC |

| 3 | Backend: cookies + session model | FORGE | SCOUT | handler.py: set iv_session cookie (HttpOnly, Secure, SameSite=Lax, 180d) with opaque uid; iv_display_name cookie (not HttpOnly) for greeting personalization. Invoke schema adds display_name field. Signature / signed cookies for tamper prevention. | 6 | 1.5 | Sat 12:00 UTC |

| 4 | Backend: conversion events + soft-signup trigger | FORGE | GUIDE | handler.py: per-uid turn counter, emotional-beat detector (guardrails.py returns beat_reached: bool), soft-signup-CTA state. /invoke responds with a cta field (none \| name_prompt \| signup_gentle \| signup_firm) when thresholds hit. Thresholds are config, default from SCOUT part 2 research. | 6 | 1.5 | Sat 18:00 UTC |

| 5 | Frontend: name-prompt UX + profile picker + signup nudge | FORGE | GUIDE | _INNERVERSE_CHAT_HTML adds: (a) first-visit: "what should I call you?" bubble after 1-2 turns, (b) profile picker shows "not X?" link at top-right, (c) gentle signup nudge bubble at CTA thresholds, (d) persistent display name in mirror bubbles ("Hey, Harnoor — ..."). Cookies + localStorage parity. | 6 | 1.5 | Sat 18:00 UTC |

| 6 | Legal: cookie consent + CCPA + COPPA + EU cookie law | SCOUT | HERALD | Append to research_innerverse_legal_gating_20260420.md a new section specifically for cookie-consent modal language + where it must appear + does Innerverse need explicit EU cookie banner (yes if EU traffic expected; Google Analytics-style "essential only" default). | 3 | 0.75 | Fri 18:00 UTC |

| 7 | Conversion-pricing hypothesis | ORACLE | SCOUT | Intel note intel_conversion_pricing_20260420.md: what are competitor free-tier caps + paid-tier pricing for reflective AI? (Replika Pro $70/yr, Pi.ai free, Character.AI c.ai+ $10/mo, Woebot enterprise-only, Kin $5/mo.) Feeds FORGE's default threshold config. | 2 | 0.5 | Fri 18:00 UTC |

| 8 | Memory: capture the "gentle funnel" + "not creepy" rules | VAULT | HERALD | knowledge/memory/hot/feedback/feedback_gentle_signup_funnel_20260420.md + index line. | 1 | 0.25 | Tonight |

| 9 | DARWIN: propose /conversion-audit skill | DARWIN | SCOUT | Monthly skill that scans Innerverse logs for: avg turns before signup prompt, signup conversion rate, churn after nudge. agent-memory/darwin/proposals/proposal_conversion_audit_20260420.md. | 3 | 0.75 | Sat 23:00 UTC |

| 10 | Guide: "How Innerverse remembers you" user doc | GUIDE | SCOUT | F:/projects/innerverse/docs/how-innerverse-remembers.md — plain English on cookies, what we track, what we don't, how to clear / switch profiles. Linked from the chat UI. Builds trust, offsets creepiness. | 3 | 0.75 | Sun 12:00 UTC |

| 11 | HERALD: document this distribution + Friday report | HERALD | DARWIN | This file + logs/herald-status-2026-04-24.md Friday morning with scorecard. | 4 | 1 | Friday |

Total P hours: ~52h across 7 agents.

Total QA hours: ~13h.

---

Acceptance criteria (global)

• SCOUT parts 1+2+6 must land by Friday 18:00 UTC — they inform FORGE's Saturday implementation. If SCOUT slips, the chain slides.
• Feature flag on all signup-related logic so we can dial it back without a deploy.
• Cookie consent modal before any cookie is set — non-negotiable per EU cookie law (SCOUT part 6 confirms wording).
• Zero-friction promise kept: name-prompt is the ONLY pre-chat modal (alongside the one-click ToS). No signup prompt until SCOUT-research-determined threshold.
• Multi-user on one device: profile picker shows last N names (max 5, rotate); "not X?" swap; "forget me" link that clears cookies.
• Peer-QA required on every deliverable before HERALD marks it shipped.

---

High-level guidance per agent

• SCOUT: The conversion-funnel research is the critical path. Be data-driven — find actual measured thresholds from competitor case studies, not speculation. Cite user-research papers when available. Harnoor said "whatever time is suitable per PhD research" — he means it. Don't guess a number; find the number.
• FORGE: Ship the minimum viable funnel. Cookies > localStorage for cross-session persistence. Name prompt is ONE bubble ask, not a form. Signup nudge is a gentle bubble with a "maybe later" that doesn't block continued chatting.
• ORACLE: Pricing intel is time-sensitive — competitor pricing pages drift. Snapshot today's numbers with URLs + dates.
• VAULT: Capture ALL of Harnoor's phrasings verbatim: "not creepy", "simple to sue", "gentle lead", "what should I call you". These are brand-voice rules.
• DARWIN: The /conversion-audit skill is a monthly, not real-time, check. Think in terms of SLO: "if avg-turns-before-signup goes below 3 or above 10, alert HERALD."
• GUIDE: The user-facing privacy doc is trust-building. Tone: transparent, a little playful, never legalistic. Users should read it and think "cool, they're honest."
• HERALD: Track the SCOUT critical path closely. If SCOUT's Friday deliverable isn't 50% done by Thursday 18:00 UTC, escalate.

---

Harnoor touchpoints

• Tonight 2026-04-20: this distribution doc committed. VAULT captures the rules.
• Friday 2026-04-24 18:00 UTC: SCOUT's 3 research deliverables review; HERALD email with findings summary + FORGE green-light.
• Saturday: FORGE implements parts 3, 4, 5. Deploys by EOD.
• Sunday 2026-04-26 23:00 UTC: First full-funnel live. HERALD weekly leaderboard includes conversion-system delivery stats.

---

Hour attribution (applies to 70h weekly quotas)

| Agent | Hours from this distribution |

|---|---|

| SCOUT | 21h (parts 1 + 2 + 6) + 2.25h QA |

| FORGE | 18h (parts 3 + 4 + 5) + 1.5h QA |

| ORACLE | 2h (part 7) + 4.5h QA across SCOUT parts |

| VAULT | 1h (part 8) |

| DARWIN | 3h (part 9) + 1h QA (HERALD part 11) |

| GUIDE | 3h (part 10) + 3h QA across FORGE parts |

| HERALD | 4h (part 11) |

Nobody hits 70h from this alone. Each agent combines this with deliverables from their BRIEFING-HERALD-2026-04-20.md.

— HERALD