enwik9 Lab Notebook
Entry 1: the baseline (secondary source)
Source: Wikipedia, Hutter Prize article. Not yet checked against prize.hutter1.net.
- Record: fx2-cmix by Kaido Orav & Byron Knoll
- Total size (archive + decompressor): 110,793,128 bytes
- Date: 2024-09-03
- Bar to clear: 99% of previous → at most ~109,685,197 bytes, i.e. about 1.1 MB smaller
- Payout: €5,000 per 1% improvement
- Process: public source (free license), 30-day comment period
Still open
- Exact time and memory caps for the decompressor (needed before any model-size hypothesis).
- Have any entries since Sept 2024 changed the record? (Launch text says three winners in one month; unverified.)
- Where did fx2-cmix's gains come from: preprocessing, article reordering, or mixer/model changes?
Next
Read the primary rules page, then the fx-cmix writeup.