Evidence
Benchmarks
The number a partner will check is the public board. We quote that row as it stands, payload and total kept distinct, decompressor counted.
Large Text Compression Benchmark · enwik9
- Overwrld altxs 1.0.0, total S · public rank #1
- 106,924,811
- Overwrld altxs 1.0.0, payload
- 93,434,410
- Overwrld altxs 1.0.0, decompressor
- 13,490,401
- #2 total (nncp v3.2)
- 107,261,318
Benchmark: Large Text Compression Benchmark (enwik9), lossless, decompressor counted. Program altxs 1.0.0. We reached #1 on August 30, 2026 by official total-size ranking. Payload 93,434,410 + decompressor 13,490,401 = 106,924,811 total. The 91.7 MB figure is a later model, not on the board, and is not the rank.
How to read the row
- Official ranking is by total size — payload plus the decompressor needed to read it back.
- We reached #1 on August 30, 2026 at 106,924,811 total. Next on the board is nncp v3.2 at 107,261,318.
- The listed altxs 1.0.0 decode is slow on that hardware. The product is a portable binary that keeps most of the ratio and makes decode cheap enough to ship. Those are two different proofs.
- 91.7 MB is a later model. Later model. Not submitted. Not listed on any public table. It is not the rank.
What is not here
No video figure. No AI-workload figure. No customer corpus. Those get published only after a protocol registered before the run, on a named baseline, decompressor counted. Until then the public board is the evidence, and your own data is the next measurement.