YEET /SEMANTIC
PLAYGROUND
Astra UltraIndependent build showcase

Less noise.
More possibility.

A playground for clear thinking and creative making. Break an idea open. See what comes out.

EXPERIMENT 002
A little structure. A lot of YEET.
04 working studios12 demo cartridgesBrowser tools · no API keyBuilt to be taken apart

Choose your experiment.

Semantic purifierLive preview
44 words
243 / 18,000
02 / THE SIGNAL4 edits

to finish the accessibility pass, we should move the release to Friday. Now, the error rate may be above 1.5%. Do not launch until the checks pass, because keyboard navigation is still incomplete.

33 words · 25% shorter
44 33Words, before & after
25%Word reduction
01Protected spans
REMOVED OR REPLACED

Okay,in order toAt this point in timedue to the fact that

Inspect exact changes (4)
  • Okay, removed · Opening discourse phrase · characters 06
  • in order to“to” · Shorter opening connective · characters 617
  • At this point in time“Now” · Shorter time expression · characters 88109
  • due to the fact that“because” · Shorter connective · characters 182202
Balanced also shortens wordy connective phrases. Uncertainty, negation, and detected literals are retained. Rules cannot guarantee meaning; review the output.

A dozen ways in.

Pick a cartridge. Every demo opens in a working studio.

12 experiments

Take the tools.
Keep the curiosity.

Small, useful instruments you can inspect, export, and make your own. This playground runs on ordinary browser capabilities.

The build, honestly credited

Original concept and scaffolding: Sol. Reimagined and expanded with Astra Ultra in ChatGPT Work, directed by Nick. Original prism artwork generated with OpenAI image generation. This is an independent creative showcase, not an official OpenAI site or a model benchmark.

What is actually running?

Deterministic text rules, explicit-label brief extraction, a term co-occurrence graph, and a seeded Canvas 2D flow field. The site does not call a language model. Astra’s contribution is the design and implementation of the experience; there is no simulated AI response.

What stays in the browser?

Your entered text is processed in memory and is not sent by these tools to a model or saved by the app. Reloading clears your work. Copy or export what you want to keep. Normal website hosting still handles page requests.

What is in the toolkit?

A dependency-free text engine, the flow-field renderer, and a complete standalone demo page. Includes Markdown instructions, a JSON project-brief schema, twelve engine checks, and an MIT license. Unzip and serve as a static website. No API key or build step required.

Show me the engine checks

Run twelve narrow checks for protected text, uncertainty, negation, line structure, and explicit decisions. Passing these checks does not prove semantic equivalence for arbitrary text.