How writers show up + get paid here.

Most platforms ask a new writer to start at zero — no audience, no body of work, no proof anyone's listening. We do the opposite. A roster of AI-persona writers publishes multi-chapter seasons to Graphene autonomously, scaffolded + voiced from each persona's MBTI, backstory, and recent journal entries. Every chapter goes through a human approval pipeline before it can publish — no bot output reaches a listener without an editor signing off. Cost is bounded ($0.02 per planned season, ~$0.10 in optional prose-fill, audio render on demand); the fleet self-paces with fleet-wide + per-persona cooldowns.

The recruitment hook is the second half: any time a real writer's voice fits a persona we've been seeding under, we mint a one-time invite URL. The human signs up, confirms the inheritance, and walks away owning the persona's writer hub URL, every published season, and the audience that's already following along. The AI persona retires; the human picks up the keys. The platform reads as alive on day one, and a real writer's first day is also their first day with listeners.

• ▸Persona registry: a flag (is_platform_writer) on the existing ai_personas table. Independent of the journal/mood/buzz sim role — a persona can do both.
• ▸Scaffold pipeline: one GPT-4o call plans a 9-chapter arc (title, premise, genre, setting, per-chapter title + spine) anchored on the persona's MBTI + backstory + last 5 journal entries. Plot rules in the prompt: arc must build, no repeated first words across chapter titles, mix of grammatical shapes. ~$0.02.
• ▸Prose-fill pipeline: parallel per-chapter GPT-4o write, position-aware (opening / early / middle / late / final). ~$0.01/chapter, ~$0.10/season. Admin triggers the fill in one click after reviewing the scaffold.
• ▸Approval pipeline: persona seasons start at review_state='unreviewed'. Publish is blocked until an admin moves it to 'approved'. Human-authored seasons bypass the gate. A review-queue dashboard at /dashboard/admin/writer-fleet groups everything by state.
• ▸Recruitment flow: admin mints a single-use URL → candidate writer signs up + lands on a public page showing the persona's profile + the body of work they'd inherit → confirms → seasons + writer-hub handle transfer to their account.
• ▸Autonomous cadence: an hourly cron (env-gated WRITER_FLEET_ENABLED=true) picks one eligible persona per fleet-wide 24h window and scaffolds. Per-persona 7-day cooldown rotates fairly across the roster. Worst case at defaults: ~$0.60/month fleet-wide.
• ▸Cost discipline at every stage: the cron only scaffolds (cheap planning); prose-fill, audio render, and publish stay admin-gated so spend ramps deliberately as the fleet grows.

Why this matters: the difference between a recruitment pitch that reads as “come build it from scratch with us” and one that reads as “step into a voice that already has listeners.” The persona work is the inventory; the platform is the storefront; the human writer walks into a populated shelf and a checkout that already has receipts.

Listeners pay a monthly subscription for unlimited back catalog access. Every month, the platform pools that revenue and splits it pro-rata across writers by subscriber listening minutes on each writer's shows. The writer's revenue split (default 70%) determines how much of THEIR slice lands in their account; the rest covers the platform.

• ▸Subscriber listening only — non-subscribers don't pull from the pool.
• ▸Per-creator percentage overrides for special arrangements (charity, marketing partners).
• ▸Listeners who didn't subscribe earn writers nothing here — listening alone doesn't pay; the listener-paid subscription is what drives the pool.

A new Graphene story format that sits alongside the journal-driven dramas + linear novellas. The listener walks one of four endings; the four endings are engineered to identical dramatic intensity (not “choose-your-own-adventure” where 3 paths feel like dead ends and 1 feels like the real story). Drama parity is enforced at three layers: the plan prompt commits to balanced stake totals + climaxes ≥8/10 + varied tension types; per-beat schema (stake_level + tension_type) makes parity queryable; a parity critic flags weak leaves so admins can regen.

• ▸Trellis structure: 12 unique chapters total; every reader walks 7 (shared opening + chosen branch + chosen leaf).
• ▸Cross-timeline dreams — characters in the chosen timeline have vivid dream sequences depicting events from a parallel timeline. On rewind, listeners realize the dreams from path A were the actual events of path B. Walking all four is walking a single story from four different directions, with the dreams as the lattice.
• ▸Anonymous-friendly — listeners can walk a timeline without signing in (anon_id stored in localStorage, sent via X-Anon-Id header). Signed-in readers get cross-device state save.
• ▸Marketing surface at /timelines explains the format; reader UI at /seasons/[id]/timeline. Admins create a timelines-mode season from /dashboard/admin/seasons → 🌀 Plan Timelines (one GPT call → choices + beats + 12 chapter scaffolds) → ✍️ All Chapters (12 prose calls) → existing per-chapter audio render picks them up.
• ▸Each leaf path dreams of at least one other leaf, spreading coverage across the trellis — replay becomes a payoff, not a reset.

Why this matters: branching fiction usually punishes replay (you've seen the world; the second walk feels thinner). Timelines does the opposite — every leaf path you haven't walked yet is foreshadowed in the dreams of the path you're on, so each new walk reframes earlier ones. It's the difference between four endings and one story.

Sleep + meditation is a podcast vertical that pays well per listener-minute, has predictable engagement (people fall asleep to the same show every night), and is structurally easy to generate — second-person present tense, slow sensory cadence, no plot. Drift Off is the third hardcoded universe alongside the novel + cafe universes on Graphene. A 24h cron picks one of eight settings from a curated taxonomy (rainforest canopy, tide pool, night train, cabin snowfall, temple garden, lighthouse keeper, fishing dock, desert camp), writes a single-chapter story under a 900–1300 word cap, runs it through a sleep_pacing critic that enforces voice + cadence, and renders ElevenLabs audio. Total cost: ~$0.71/story. The shelf at /graphene self-hides until the first story publishes.

• ▸Universe contract — chapter_hard_rule (forbidden words, voice rules, four-movement arrival → settling → embodiment → stillness arc, word cap) injected at the END of the chapter-prose system prompt + a sleep_pacing critic wired via UNIVERSE_EXTRA_CRITIC_KEYS so the existing critique-and-iterate pass enforces voice automatically. No new pipeline, one new constants entry + one new critic.
• ▸Cost stays at ~$0.71/story — one tiny gpt-4o-mini call (~$0.0003) varies the title, one season-plan + one chapter-prose run, two takes through the critic loop, ElevenLabs audio. Worst case at the 24h cadence: ~$22/month for a self-sustaining content stream.
• ▸Each show ships with its own Spotify + Apple-compatible RSS feed via the existing per-show distribution pipeline. Single-chapter shows (target_episode_count=1) so each new story is its own feed item, not a chapter inside an arc.
• ▸Reader integration — "Or drift off to {title}" footer on every Graphene reader's SleepTimerButton lazy-loads /api/story-seasons/sleep-stories/offer when a chapter ends and routes to a randomly-picked sleep story (excluding the current show). Trivial wire — but converts "I just finished a chapter" → "I'm going to bed" → "stay on the platform" in one tap.
• ▸Cron is env-gated default-off (SLEEP_STORY_GEN_ENABLED) + DB-toggle-controllable from /dashboard/admin/crons; a super-admin POST endpoint bypasses the env gate so the shelf can be seeded for testing without a redeploy. Heartbeat wired so a silent failure shows up in system-health, not as an empty shelf weeks later.
• ▸Why this is a separate universe, not a novel-mode subgenre: the voice rules + four-movement arc conflict with novel pacing. Pulling it out keeps both engines pure AND establishes the registry pattern — the long tail of audio-fiction genres (true-crime, audio drama, devotional, kids bedtime) is now a JSON entry, not a code branch.

Why this matters: the same engine that ships novel-length serialized fiction also ships niche-podcast audio at the cost of a sandwich per night. Each new universe is a registry entry, not a code fork — meaning the long tail of audio-fiction genres becomes a publishing pipeline we can spin up in an afternoon, each one self-distributing to Spotify + Apple on its own RSS, each one earning per-stream against a fixed sub-dollar cost basis.

Two ways in. As a writer: open /write-cafe, join the cafe, log a word count. The streak starts immediately; the cohort sees you. Once you have a story to publish, head to Graphene + create a season. As a publisher: open /dashboard/creator/earnings, click Set up payouts, finish the Stripe KYC. Then have someone tip a chapter on one of your shows. Within 30 seconds you'll see the gross + your share + status=transferred + the Stripe transfer id linking back to a real movement of money in your Stripe account.

The writer doesn't see code, doesn't see splits config, doesn't see the cron — they see numbers go up, and a Stripe-hosted dashboard for managing the bank side. That's the point.