One seed becomes a playable AI drama.

Tiny Stories is a portfolio case study for an inspectable AI story runtime: a short, story-first mobile episode where players read one scene, compare a few meaningful moves, act once, and follow the visible consequence. The reviewer run then exposes the runtime contracts behind that loop.

Evidence boundary: use the video for orientation, then open the public reviewer demo for a backend-free playable evidence surface. Cite current app surfaces only after the public-link check passes.

Open public reviewer demo Reviewer run guide

React 19 FastAPI Typed runtime contracts Hand-drawn notebook visual system Programmatic video demo

75s reviewer cut Watch first, then open the public reviewer demo to inspect the checked evidence path. Watch 75s demo Public reviewer demo Reviewer run guide Source evidence

Open MP4 demo

What reviewers can inspect

Playable state Public deterministic reviewer demo exposes locked seed, role, turn budget, assets, and choices.

Consequence One move should visibly change the room, assets, or next-choice pressure.

Evidence limits Reviewer evidence stays separate from normal play and public claims need a public-link check.

Replay artifact A finished run should become highlights, full read, share, and same-opening restart.

Playable structure

The seed is compiled into cast, role, assets, pressure, hidden objectives, and a turn-by-turn interaction loop.

Inspectable engineering

Reviewer mode exposes stage, role, inventory, option count, advisor context, and ending state instead of hiding everything inside prose.

Portfolio presentation

Real UI capture, hand-drawn notebook styling, and scripted narration communicate both product taste and system design.

Boundary: this is portfolio-grade AI product-system evidence, not a launched consumer game or broad adoption proof. The source evidence starts with the system map, public reviewer demo, reviewer run, contracts, and tests. The public reviewer demo is deterministic and backend-free; a live generated run requires a configured backend and must keep API keys server-side. Before relying on public GitHub Pages or repository links, run python3 tools/portfolio_public_evidence_preflight.py; if it reports local commits ahead of public main or missing Pages markers, use the demo video for orientation only. Do not cite the current Portfolio, Reviewer, Story Desk, Create, Play, or Replay surfaces as public evidence until the intended branch is deployed, rechecked, and the public-link check passes. Repeat play, natural sharing, and target-player comprehension are still unproven; the next evidence step is a small real-user playtest using docs/playtest_report.md.

Who this loop is for

Tiny Stories is designed for story-first players who want a compact mobile episode: read the scene, compare a few meaningful moves, act once, then use the consequence to choose the next beat.

Target player

Someone who wants interactive drama, not a blank writing canvas or a dashboard.

Content rhythm

Short scene, clear pressure, one choice, visible consequence, next choice.

UI promise

Keep story context near decisions; keep reviewer evidence outside normal play.

Runtime flow

01 Seed

A simple premise starts the session.

02 Compiler

The runtime builds roles, cast, objectives, and leverage.

03 Play loop

The user picks options or writes free-form actions.

04 Advisor

A bounded side-channel reasons over run context.

05 State

Messages, turns, inventory, and consequences persist.

06 Ending

The played path compiles into a shareable final artifact.