CORE 0.1 · INTERACTION PROTOTYPE

Point at the web.
Make the meaning visible.

This playground proves the first Televizer interaction: calm hover intent, spatial provenance, semantic presentation, and four commands.

QUOTE MODE

A little text to lift

Good presentations do not need to repeat every word on the page. They need to find the sentence that carries the idea, give it room, and let the audience understand why it matters.

Televizer keeps the source visible while a selected phrase moves into focus. Select a few words, release at the end of the swipe, and leave the pointer still for the quote to appear.

MEDIA ZOOM

Images and video get a stage

Pause over either frame to open Televizer's media view.

Serena, a white dog with blue eyes, relaxing on a leather couch — Image · Serena

Video · Flower time-lapse

CHART ZOOM

Gesture through the whole story

Pause over the chart, then move across it to reposition the magnifier.

DeepSWE benchmark score Average pass@1 · Higher is better

ELEMENT MODE

A television layer for ordinary metrics

Activate Televizer, then pause over a card.

↯

Average TTFB

286 ms Latest check per image

✓

Healthy now

99.4% 54 of 55 serving normally

◷

Complete load

681 ms Full image response

Active failures

1 One asset needs attention

TABLE MODES

Benchmark comparison

Hover a value. Try R, C, then 1.

Selected model capability scores
Model	MMLU score	GPQA score	SWE-bench score	Math score
GPT-5	89.3	82.1	74.8	96.2
Claude	87.9	84.8	77.2	94.7
Gemini	88.1	81.7	75.6	95.4
Llama	82.4	73.2	62.1	88.9
Mistral	80.7	70.6	58.9	86.3

DIRECTION-AWARE ORDINALS

Response latency

Here, lower is better. Source order still never changes.

CDN latency by region
CDN	Virginia TTFB	Oregon TTFB	Frankfurt TTFB	Sydney TTFB
Adobe	38 ms	74 ms	112 ms	184 ms
Stripo	67 ms	48 ms	91 ms	152 ms
Cloudflare	18 ms	26 ms	34 ms	61 ms
Fastly	29 ms	31 ms	46 ms	72 ms

THE INVARIANT

Televizer transforms emphasis, not topology.

Rank can change what the audience sees. It never changes where an item belongs.