Job 1: Your whole file (English) → count inbound tokens (what the LLM would receive).
Job 2: Same file after compression → compressed text in the output box → count inbound tokens again.
Product goal: Job 2 should be about 30% fewer inbound tokens than Job 1. This page shows whether you hit that on your file.
File contents go in the input box (full English).
No file chosen yet.
Optional:
Compress into the output box and fill the token table.
Output area = compressed file (Job 2 text).
| Job 1 — full file | Job 2 — compressed file | |
|---|---|---|
| Characters in box | — | — |
| Inbound tokens | — | — |
| Job 2 vs Job 1 | — | |
| ~30% product goal | — | |
Load a file, then click RUN JOB 1 + JOB 2.
Token counts: estimated here (chars÷3.8). Measured Claude counts when server API is connected.
One essay, hand-built notation: 1,661 → 548 inbound tokens (~67%). That is a separate Demonstration test, not what Job 2 produces today.