← Koraki  ·  Design notes

Research library

Dated reports from the experiment. Each one states what was true on its closes-through date and is never edited afterwards — a correction goes at the top of the next report, not into an old one. Read the freshness state before reading the numbers.

2026-09-08 status  

Six weeks in, no measurable edge

Twelve picks average −0.76 points against SPY — 0.48 standard errors from zero, with dispersion seven times the size of the mean. A null, not a negative.

Closes through 2026-09-05 · n=12 · 26 trading days
2026-08-28 field note  

The first exits

The first two positions closed. FTAI is a stop working; META is a stop failing. The entire paired difference rests on one name.

Closes through 2026-08-27 · n=10 · 21 trading days
Why old reports stay up unedited. A report is a record of what was believed before the answer was known, which is the same reason the rules were pre-registered. Editing them to stay current would leave one document that has always been right — worth far less than a series showing what changed and when. The database holds only the present state, so these are the only durable record of intermediate readings.