Ink & Ember / Arnow Solutions

Testing/DX deductions are system-level (no suite, no git), not per-finding. Full dimension×severity detail in artifact 12.

1 · Find-the-Flaw: "Next Passage" button is dead

The button calls nextFlawPassage(), which no fragment defines. Clicking it after finishing a passage throws ReferenceError and nothing advances — the player must find the "New Passage" ghost button instead. Harvested live as a pageerror during the walkthrough.

Fix: define nextFlawPassage() as clear-score → newFlawPassage(). 15 minutes, include a regression step in the smoke harness.

2 · arnow.db has zero scheduled backups

/root/backup-db.sh (3 AM cron) covers Postgres only. The single SQLite store holding every lesson pass, capstone draft, and reflection is unprotected against host loss. A manual probe confirmed sqlite3 .backup works on the VPS — the fix is one line in an existing script + one restore drill.

• Day 1: nextFlawPassage fix + regression step in harness — 15 min
• Day 1: arnow.db cron line in backup-db.sh + one restore drill — 30 min
• Day 2: contrast token darken — 20 min
• Day 2: /api/health endpoint — 15 min
• Day 3: CORS origins pinned to the 2 known hosts — 15 min
• Day 3: git init + first commit — 10 min

2-week sprint

Days 1–3: the six quick wins above. Days 4–5: pytest-ify the smoke harness (3 assertions: lesson pass, flaw-next regression, capstone gate). Days 6–9: inline empty-submit feedback + hide admin nav by role. Day 10: print-capstone stylesheet start.

6-month vision (what to add next)

• Anthology export — capstones become typeset, printable "author's anthology" extracts with shareable no-login links
• Daily commitment layer — 20-minute writing streaks; habit beats intensity for a 10K-books/month goal
• Curriculum tier 2 — 6 more deep lessons per author + a true Suspense track; the content pipeline is proven to scale
• Coach intelligence — index the learner's own drafts: "you lean on filter verbs in high-tension scenes"; author-voice practice prompts
• Time-instrument report — "when you actually write" (event timestamps exist already)
• Multi-learner mode — only when the academy grows; single-learner design is the product's strength

New features (ranked by ROI)

1. Anthology/capstone export — the app's signature print artifact; differentiates it from every generic writing app
2. Streak habit layer — cheapest behavior lever, data exists
3. Second curriculum tier — highest long-term value; pipeline validated
4. Draft-pattern intelligence — the "coach that knows your writing" is the differentiator the AI backend enables

UI / design & graphics

1. Reading-position rail on lessons-in-progress (bookish, small)
2. Lesson-modal 150 ms custom-bezier ease (motion polish)
3. Critique results as an "editor's letter" typography treatment (looks like a letter from a mentor)
4. Print stylesheet (manuscript look) for any exercise/capstone
5. Warm-duotone author portraits — the last text-only surface