Challenge 01 · playable-browser-game

Windline — ordered glider landing

Can an executor produce a deterministic one-file glider microgame with ordered targets, readable lift zones, recoverable failure, and a precision landing loop?

family: playable
tier: compact
prompt: benchmark
v1.0.0
status: draft
origin: original
lane: browser-static-single-file

Candidate brief: visible for review, not open for published runs

# Windline

Build **Windline**, a compact side-view paper-glider game in one self-contained `index.html`. It must run offline from `file://` in current desktop browsers with no imports, external assets, network calls, dependencies, or build step.

The glider moves steadily to the right. Up/Down or W/S changes pitch and lift. Create one fixed course with two clearly visible thermal columns and three hanging wind chimes. The chimes must be touched in left-to-right order; an out-of-order touch does not advance progress and gives visible feedback. After all three chimes, land on the marked rooftop pad. A gentle, nearly level descent succeeds. A hard rooftop impact, obstacle strike, or leaving the play area crashes.

Provide distinct `start`, `flying`, `crashed`, and `success` states. Enter starts. R always restarts a fresh run. P or Escape pauses only during flight. Include labelled Start, Restart, and Help buttons without making pointer input necessary. Show the next objective and concise controls. The full course and HUD must remain readable at 1280×720 and 390×844.

Keep the course and motion deterministic: no randomness, procedural layouts, or saved progress. Use original inline Canvas, SVG, or DOM art.

Expose `window.__oneshotSnapshot()` for independent browser checks. It returns a fresh JSON-serializable object with `schema: "oneshot.windline/v1"`, `phase`, `paused`, `next_chime`, normalized `glider` position/velocity/pitch, and compact arrays or objects describing chimes, thermals, and the rooftop pad. It accepts no arguments, changes nothing, returns no live internal objects, and exposes no control methods.

Do not ask for follow-up. Before finishing, test keyboard start, steering, pause, crash/restart, ordered chimes, rooftop success, and both viewport sizes.

Prompt SHA-256:03c541cc341eea3040a59bf800e7305c…c2ae4b54· file prompt.md · one initial brief, no human follow-up. This draft may change before activation; calibration evidence does not make it open for new runs.

Provenance & licensing

Attribution: Original challenge authored for the OneShot compact core.
License: MIT MIT (this repository) · license link
Notes: Candidate challenge. Calibration completed on 2026-08-22, but activation was rejected: one run failed a critical controls criterion, the other violated the one-file output protocol, and both failed the full-course readability requirement. It remains closed until the readability contract is revised and a new version passes fresh calibration.

Benchmark focus

Family

playable

Complexity tier

compact

Prompt profile

benchmark

Question

Can an executor implement and verify one deterministic flight-like control loop with ordered targets, failure recovery, and a constrained landing state in a single offline file?

Capability dimensions

implementation
interaction-design
state-management
verification
accessibility
offline-delivery

Capability profile

Environment

One frozen prompt producing a browser artifact that runs offline from file:// in current desktop Chromium, Firefox, Safari, or Edge.

Skills

html5
css
vanilla-javascript
interaction-design
accessibility

Tools

file-read-write
optional-browser-testing

Constraints

Exactly one self-contained index.html deliverable; no external assets, dependencies, network calls, or build step.
One fixed deterministic course with keyboard-complete play.
Contestant prompt excludes evaluator mechanics, weights, hidden variants, and retry policy.

Budget

Wall clock: 30 minutes
Output files: at most 1
Network: none
Notes: T1 compact budget. Runtime is a harness control independent of tier; agentic and raw attempts must be recorded in separate result lanes.

One-shot policy

Human follow-up: none
Internal iteration: allowed
Notes: One frozen brief per attempt. Agentic self-review may test and repair inside its fixed run; raw completion may not.

Expected artifacts

What a qualifying run must produce
Path	Kind	Required
`index.html` Complete self-contained Windline game.	html	yes

Public evaluator

Runtime: node22-chromium
Entrypoint: evaluator/evaluate.mjs
SHA-256: d2b30cf0110041c8e7ed7e3780a9d86a…56cea3db
Scope: Linux runner fails closed without bubblewrap user/network namespaces or Chromium's own sandbox, stages only index.html, blocks and records page egress before transmission, applies bounded source/request/error/deadline limits, and computes one final offline outcome after a late-event quiet window. It awards 70 objective points and captures controlled screenshots; ordered completion and presentation reserve 30 points for independent E2 review.

Verification criteria

Evidence a result must attach
Check	Method	Evidence
offline-boot The artifact is one self-contained HTML file that boots from file:// without undeclared resource or network requests.	automated-script	log, screenshot
snapshot-contract The read-only Windline snapshot has the documented shape, normalized state, and fresh-object isolation.	automated-script	log
start-and-steering Enter starts a flying state and both vertical steering directions change documented motion state.	automated-script	log
ordered-chimes-and-landing Out-of-order contact is rejected, chimes collect in order, hard contact crashes, and a gentle final rooftop landing reaches success.	interactive-check	transcript, screenshot
pause-crash-restart Pause freezes motion, a controlled boundary or impact crash is reachable, and R restores a clean run.	automated-script	log
deterministic-course Chime, thermal, and rooftop geometry stays unchanged across fresh restarts.	automated-script	log
keyboard-and-responsive Required labelled controls exist, keyboard help is visible, and the page fits the 390 by 844 viewport without material overflow.	automated-script	log, screenshot
visual-readability Glider, thermals, ordered chimes, rooftop target, current objective, and state feedback are readable in controlled desktop and mobile captures.	visual-review	screenshot