Narrly release notes

**CAST:** eight Vogler roles (adds Herald + Trickster) in For / Against creator UI with expand detail + icons. Story `who` labels only — no confidence scores. Default **on** (`NARRLY_ARCHETYPES` / `NEXT_PUBLIC_NARRLY_ARCHETYPES`; set `0` to kill).

**Generate:** OpenAI (`gpt-4o-mini`) cast scan after refine; regex cue assigner is fallback only. Still stripped on publish; mobile feed serializes owner-only.

Creator panel prefers stored/LLM cast over local cue rebuild.

**Why this lands:** public receipt prints the arc once. Elixir is only Return with Elixir — it no longer copies Stakes.

**Stabilize:** generate overlay uses the same Power panel as editor and owned detail. Create uses one cover path (`gpt-image-1`) with 45s server / 55s client timeout and a Retry banner. Refine stays default-on (`NARRLY_NARRATOLOGY_REFINE=0` to kill).

**One beat vocabulary:** Ordinary World → The Call → The Quest → The Reckoning → The Breakthrough → Return with Elixir (`lib/canonicalArc.ts`, `claude/narrly-vogler-map.md`). `NARRLY_ARC_V2` (off) switches generate/parse/stitch to 6 beats and a 7-dimension composite. Simulate eval stays on the 5-metric floor.

**Cast overlay:** Hero, Mentor, Shadow, Threshold Guardian, Shapeshifter, Ally on creator surfaces only (`NARRLY_ARCHETYPES`, off). Mentor is metadata — never cast in the body. Stripped on publish.

**Refusal of the Call:** the likely objection lives inside the arc; Counter stress-tests it (`NARRLY_OBJECTION`, off).

**Power-as-coach:** why a dimension scored that way, the fix, and Sharpen the arc (`NARRLY_POWER_COACH`, off). Do-no-harm unchanged.

**Recast:** Board / Buyer / Analyst / Press applies to the open draft (`NARRLY_AUDIENCE_RETARGET`, off). Not a silent copy.

Analytics (ids and scores only): `arc_v2_scored`, `archetype_assigned`, `objection_generated`, `power_coach_viewed`, `power_coach_applied`, `audience_retargeted`.

**Narrative Power:** owned editor and detail now show a creator-only Power readout (composite, dimensions, 5-beat tension sparkline, turning point + stakes). One-tap **Sharpen** calls targeted refine with the existing do-no-harm tension guard. Published narrative body stays prose-only — no beat labels or panel copy.

**Discovery Engine** is the first-run create path: paste a URL or notes above the fold, score the four-beat draft immediately, then Refine or Publish.

**Vogler alignment (flagged):** `NARRLY_ARC_V2` restores Ordinary World / Ordeal / Return-with-Elixir as first-class optional beats and adds `ordinary_world_contrast` + `elixir_takeaway` scoring. Default off so the 5-dimension composite floor does not move.

**Why this lands:** on publish, a shareable receipt (arc one-liners, turning point, stakes, elixir, Power score) is stored and rendered beside the public narrative — it does not alter the body.

**Refine rollout:** default **on** after live eval (+7.5 composite). Kill with `NARRLY_NARRATOLOGY_REFINE=0`. Optional percent bucket via `NARRLY_NARRATOLOGY_REFINE_PERCENT`. Mobile generate already runs `refineNarrativeDraft` and returns `narratology` scores. Do-no-harm tension guards stay intact.

**Cover generation:** `gpt-image-1` calls time out at 45s; stale pending/generating covers are marked failed so mobile cannot spin forever. Failures return an error (not a silent data-URI “success”).

**Mobile feed detail** now includes `narratology`, `whyThisLands`, `coverImageError`, and `isOwnedByMe` so Power + receipt can render on-device. Mobile `stage_*` ids map onto Campbell beats so Why this lands is actually stored at publish.

**Welcome email** no longer blocks Create. Verified users proceed immediately; Welcome retries in the background.

**Mobile Sharpen** returns `narratology` plus `composite_before` / `composite_after` so the client can refresh Power. Mobile generate writes Call / Quest / Reckoning / Breakthrough / New World as the actual section jobs (ids unchanged for scoring). Headings that are beat labels are stripped. The product is not cast as a Mentor.

Analytics (privacy-safe): `power_panel_viewed`, `sharpen_tapped`, `sharpen_succeeded {dimension, composite_before, composite_after}`, `discovery_engine_intake {scored, composite}`, `receipt_generated`, `receipt_shared`.

Free email signup waits on `/verify-email` before guided Create, so Welcome can send after verification. Unverified password users are gated out of Create. A failed welcome email no longer deletes the account; the user can retry.

Internal narratology scoring on generation: 0–100 `tension_arc`, `turning_point_strength`, `reversal_stakes`, `arc_coherence`, `specificity_evidence` (plus composite). Existing section metrics stay secondary. Nothing is shown in the UI.

Self-refine loop: if a core dimension is below threshold, rewrite the weak section once (escalate stakes / sharpen the reversal), re-score, cap 2 passes / 8s. Flag `NARRLY_NARRATOLOGY_REFINE` defaults **off** until Dane flips it (optional percent A/B). The rewrite instruction is a **system** message; only the model completion is persisted. If the completion contains instruction fragments (`Current section:`, `rewrite the given narrative`, `word count`, `Do not invent`, …) the pass is rejected and the original section is kept. Eval fails the run if any refined section still contains that instruction text.

Privacy-safe `narrative_scored` analytics emit sub-scores only — no body text.

Narrative quality eval: `npm run eval:narrative` runs a fixed 16-case Product/Market/Mission/Intelligence benchmark through generate + 2.0 scoring, prints per-dimension averages and composite, and fails on regression vs `data/eval/last-run.json`. Pass `--update` to freeze a new baseline; `--live` for real OpenAI generation (and refine, unless `--no-refine`).

Refine do-no-harm: rewrite only a dimension below `NARRLY_NARRATOLOGY_REFINE_THRESHOLD` (default 65). After rewrite, re-score all five and keep the original unless the target rose and no other dimension dropped more than `NARRLY_NARRATOLOGY_HARM_TOLERANCE` (default 1). **Protected dimension:** `tension_arc` may not drop more than `NARRLY_NARRATOLOGY_TENSION_EPSILON` (default 0.5) on any pass — a rewrite that lifts stakes/coherence but drops tension beyond epsilon is rejected and the original section is kept. A rejected pass still counts toward the 2-pass cap. Nothing shown in the UI.

Generation model A/B scaffolding (`NARRLY_GEN_FRONTIER_MODEL`) stays in tree for future Claude/o-series or prompt tests. Flag defaults **off**; generation remains **gpt-4**. Live eval (n=16, seed 1688) showed gpt-5.5 vs gpt-4 composite −1.00 and tension −7.50; gpt-5.5 is rejected.

Live refine ON vs OFF (gpt-4, n=16, seed 1688): composite **+4.06** (11 up / 4 down / 1 tie), tension_arc **+6.88** (7 / 2 / 7). Time 438s ON vs 276s OFF. Ship bar **PASS**. Flags remain off.

Activation funnel: server-side PostHog events (`signup`, `first_generate_started`, `narrative_created`, `narrative_published`, `returned_d1`) with `distinct_id` bound to the user. Each canonical event fires once per action (server is the only emitter). DarkBrief uses `narrative_published_dark`. Homepage generate no longer client-captures `narrative_created` (now `create_generate_succeeded`). Session ping accepts `surface: mobile` for D1.

Signup persists a non-null acquisition `source` (UTM / referrer host, fallback `direct`) and sets it as a PostHog person property. No narrative body or PII in analytics.

Added `npm run verify:funnel-readiness` to fail if any funnel event is missing in the last 7 days.

First-narrative activation (web): after signup, land on a pre-filled Daily Provocation with Generate above the fold (`?start=guided`, also the default for users with zero narratives). Publish is one tap in the post-generate editor. Wizard, templates, and lens picker remain available as power-user paths.

Discovery Engine intake (flagged first cut): paste a landing-page URL or product notes and get a four-beat arc (Prior Belief, Inciting Truth, Reversal, New Trajectory) as a draft you can Refine or Publish. Framed as **the story your tech is hiding**. Reuses auth, generation rate limits, and guardrails; privacy-safe `discovery_engine_intake` analytics only.

Script brief: duration chips (Free 10/15/30/60s; Pro also 2–30 min) and optional 10s Grok Imagine segments; Pro durations gated on subscription

Video kit: extract stages/sections/summary into timed 10s scenes when the model omits a scene list

Talking-head: Pro camera cap 120s / 80MB

Copy: **Narrlies** (many) · **My Narrlies** · **From your Narrlies** · 3/15 Narrlies per day — never Narrlys

Expo push: `POST/DELETE /api/v1/me/push-token` persists `expoPushTokens` on `userProfiles`; `plan_ready` sends OS push via Expo Push API (no plan body) in addition to Activity.

Loops opt-in auto-publish (Pro): schedule field `auto_publish_on_plan_approve` (default false); after plan approval + passing verification on a schedule-triggered run, may continue to Narrly publish/delivery. Failures leave the run recoverable and never fail approval/verify.

Script brief: `POST /api/v1/narratives/[narrativeId]/script-brief` returns `{ title, script }` (LLM with local fallback; hard length caps; privacy-safe `script_brief_generated` analytics).

Home interests: `GET/PATCH /api/v1/me/interests` persists up to 5 fixed tags on `userProfiles`; trending / for-you ranking applies a capped heuristic soft-boost when narrative mode/type/tags overlap.

Automated Loops: progressive disclosure — hide schedule UI until first completed Manual Loop; hide schedule chrome during Manual Loop intake/run; Manual Loop copy; unlock strip after approve.

Talking-head Narrlies: Pro owners can upload ≤120s vertical video (`POST /api/narratives/talking-head`); status `pending` → `ready` | `failed`; mirrors to `publicNarratives` for Narrlies; DarkBrief blocked; soft fail leaves covers intact.

Scheduled Loops: Activity notify on `plan_ready` (in-app; Expo push when token present).

Admin: sync-only cover backfill `POST /api/admin/cover-sync-backfill` (mirror private→public; dryRun default).

Cover regenerate: Free plan unlocked (still rate-limited); Pro unchanged.

Web Tusk coach panel on owned narrative detail (asks `POST /api/v1/tusk/ask`; no voice).

Tusk ask: cross-session memory — last 8 Q/A turns per `(userId, narrativeId)` in Firestore `tuskSessions`, bounded into the model prompt; Q/A never sent to analytics.

Switched screenshot-intake analysis from Fireworks vision to OpenAI `gpt-4.1-mini`, with auth, per-user rate limiting, request size bounds, and a hard request timeout.

Fixed screenshot-intake analysis: Fireworks serverless no longer serves `qwen2p5-vl-32b-instruct` (404). Switched default vision model to `kimi-k3`, raised route `maxDuration` to 90s, and hardened JSON parsing.

Added server-side Narrative Guidance support for Mobile generation, including profile-default fallback and a bounded `USER NARRATIVE GUIDANCE` prompt section that cannot override safety or required structure.

Added default-safe persistence for `narrativeGuidance` on generated and saved Mobile narratives.

Added Alias publishing support for standard Mobile narratives with frozen `publicAuthorType` and `publicAuthorNameSnapshot` fields.

Hardened public feed/detail serialization so alias-published narratives do not get re-enriched with the account profile name or avatar.

Returned `isOwnedByMe` for DarkBrief detail so Mobile owners can edit their anonymous public signal without revealing identity.

Added an authenticated Fireworks-powered screenshot-intake analysis API for Mobile so a pasted or selected screenshot can become a bounded narrative prompt without persisting raw screenshots.

Reconciled Platform UX with the Mobile-led v1.6.2 activation model: `/workspace/narrly` now offers the same structured Daily Provocation prompts and starts Create with the matching lens.

Updated Platform narrative tools to match Mobile's clearer language: visual Counter claim labels, compact/collapsible Turning Points, source-material Arc copy, and supervised-first Loop schedule gating.

Fixed the scheduled Loop Firestore index configuration needed by the production Loop scheduler.

Added iOS Universal Links for Create at `https://narrly.ai/open/create`, with an Apple App Site Association document and a browser fallback into the guided first-narrative flow.

Added a Smart App Banner and homepage **Get the iPhone app** CTA for the live App Store listing.

Rewrote customer-facing product emails around narrative power instead of four-lenses taxonomy.

Rebuilt lifecycle emails on Narrly’s product light brand: teal wordmark, “Shape the narrative,” white card on `#F9FAFB`, and `#0dceda` CTAs.

Rewrote the narrly.ai homepage hero and three pillars around narrative power. Capability labels stay visual proof.

Verified focused messaging, homepage, Universal Link, and welcome-email tests before production deploy.

Production alias `www.narrly.ai` updated with the v1.6 messaging and Create Universal Link surfaces.

Added Free and Narrly Pro entitlement helpers used by Narrative detail tooling and pricing.

Added a body-first Narrative tools strip on owned Narrative detail with Counter, Turning Points, Arc, Mutations, and Loop access.

Added bounded Counter soft-limit handling with an upgrade path for Free users.

Updated customer-facing pricing to the Free and Narrly Pro packaging model.

Redesigned the verified-user welcome email for first-narrative activation with a stable subject, safe greeting fallback, one guided CTA, and no four-choice lens list.

Hardened Counter Narrative and Turning Points API failures with bounded, user-safe error codes instead of raw provider errors.

Locked Mutations behind Narrly Pro while keeping Counter, Turning Points, Arc, and Loop visible to Free users.

Fixed the Turning Points analyze route so file-parser imports no longer crash JSON narrative analysis at production route bootstrap.

Improved Turning Points provider parsing to normalize safe response variations before schema validation.

Verified focused Counter Narrative, Turning Points, Narrative Loop, and entitlement tests before production deployment.

Verified runtime behavior on production revision `c08c0a65`; subsequent v1.5.1 commits in this line are release-record only.

Added Narrative Loops on owned Narrative detail so users can refresh a Narrative with new context (`What changed?` → Create Loop Plan → Approve plan) without using Narrative Arcs as a primary nav product.

Added `narrative_loop_started` analytics alongside existing Arc rerun events (bounded IDs/statuses only).

Added supportive, hideable Trust checks panel (“What Narrly checked”) on owned Narrative detail.

Added private Counter Narrative generation on Narrative detail with claim labels and no public debate surface.

Added owned Counter **Use to revise** that opens the Narrative editor with private Counter guidance (never auto-published).

Added Explicit Preference Memory on Narrative Loops: save only after confirmation, review/disable/delete, and Loop plan disclosure of enabled preferences.

Improved Narrative Loops so plan approval can generate and verify in place on Narrative detail, with Accept/Reject and optional preference save.

Added in-place Publish & confirm delivery on Narrative detail Loops, plus `second_narrative_loop_completed` and `outcome_delivered` analytics.

Added Outcome Analytics v2 events: `narrative_loop_completed`, trust hide/show, and typed Counter lifecycle captures.

Reframed the Narrative detail Arc panel as owned Narrative pillars and demoted standalone Narrative Arcs from primary navigation; Loop is the primary return path on detail.

Added supervised Narrative Arcs that turn an authorized Radar signal or manual research input into an editable plan, canonical Narrative, Verification Receipt, confirmed delivery, and repeatable Run.

Added Narrative Arc activity to owned Narratives and My Narrlys so users can open the connected Arc, review its delivery history, and continue work without leaving their Narrative library.

Added explicit Arc controls for plan approval and revision, pause and resume, cancellation, outcome acceptance or rejection, bounded retry, and confirmed prior-version restoration.

Added Saved Arc reruns that require fresh source input, preserve earlier Runs, and carry forward only corrections the user explicitly chooses.

Added Google Docs export with secure Google connection and reconnection, recognizable provider branding, clear failure guidance, an explicit Open Google Doc action, and a source link back to published Narrlys.

Improved Narrative Arc verification with a four-section Receipt covering checks, evidence basis, limitations, permissions, and destination readiness before delivery.

Improved public Narrative destination links so legacy and canonical paths open without duplicated route prefixes.

Improved Narrative Arc accessibility and responsive behavior across desktop and narrow mobile-width Platform layouts.

Added private, narrative-native Mutations analysis across Platform and Mobile. The Mutation Halo maps 3–8 potential or evidence-backed observed variants around the canonical Narrative ID, with server-computed gravity, accessible detail views, public-remix lineage, and stress-test refresh.

Added narrative-native Turning Points analysis across Platform and Mobile: generated and existing Narrlys are analyzed by canonical Narrative ID, results live inside narrative preview/detail, and approved sharper turns update the owned narrative. External text, DOCX, and PDF analysis remains available as a secondary draft tool.

Added Tribe-native Constellations beta for explainable, steward-approved discovery across three or more autonomous public Tribes, surfaced inside Tribes with difference-preserving explanations and leave, mute, correction, and report controls.

Added canonical narrative lifecycle analytics IDs across generation, publishing, viewing, sharing, and remix flows.

Added collision-safe friendly public narrative URLs, narrative-specific social metadata, and a dry-run-first legacy slug backfill.

Added an automated public Release Notes page sourced from this changelog.

Improved production dependency security by clearing the critical advisories that were present in the prior stack.

Fixed the signed-out public narrative viewer crash and friendly-route live subscription lookup.

Fixed web publication failing with Firestore permission errors by moving slug reservation and public-mirror writes behind the authenticated server publisher.

Fixed homepage, canceled-checkout, and email-verification prerendering on the patched Next.js release.

Added Master Plots API support for creating narrative arcs and resuming them with chapters.

Added public-safe Discover User Search support for profile lookup by name.

Added Pro-gated Custom Integrations API support for HTTPS webhook connections.