The days it stayed out: what share of the window each robot traded

2026-09-04. Every figure quoted here is a field in a published audit JSON, and the listed audits are linked in place.

A backtest that covers 23 years does not trade for 23 years. Every audit we publish counts the days on which at least one trade closed and sets that count against the weekdays available in the same window. Of our 26 default audits, 22 carry the resulting share and the answers run from 6.27 percent to 91.08 percent. The two ends are not better and worse, they are different machines, and the number is worth reading before the profit line because it says how much of the record is actually record.

How the number is built

Three fields, all in the concentration block of the audit file.

The active count. Our engine groups every paired trade by the date on which it closed and counts the distinct dates. A day is active when something closed on it.
The denominator. We count every weekday between the first and the last day of the tested window. That count has no holiday model in it.
The share. Active days divided by weekdays in the window, rounded to four places. Nothing about size, direction or result enters the calculation.

The busy end of the catalog

Prop Firm Gold EA closed a trade on 5,533 of 6,075 weekdays, a share of 91.08 percent, across 9,538 trades. Smart Gold Hunter is second at 5,329 of 6,076 days, 87.71 percent, across 10,416 trades. Gold Atlas follows at 4,636 of 6,075 days, 76.31 percent, across 10,999 trades. Eight of the twenty two closed something on more than half of the available weekdays.

The quiet end

Logan is the quietest record we hold. It closed a trade on 381 of 6,075 weekdays, a share of 6.27 percent, across 1,372 trades in a window running from 2003.05.05 to 2026.08.16. The Gold Reaper is next at 775 of 6,076 days, 12.76 percent, across 1,742 trades. Ten of the twenty two closed a trade on fewer than one weekday in four.

Activity is not a verdict

The busiest records in this catalog are not the best ones. Smart Gold Hunter closed trades on 87.71 percent of the available weekdays and still ends at minus 1,418.55 USD after costs. Logan closed on 6.27 percent and ends at minus 24,334.65 USD after costs. Gold House closed on 3,192 days and CryonX EA on 3,191, one day apart on the same length of window, and their results after costs differ by 7,598.31 USD. The share tells you how often a machine finished a trade and nothing at all about what the trade was worth.

Why it changes the concentration line

Our concentration number asks how few days carried most of the profit, and its denominator is the active days rather than the window. See the concentration article for how that count is built. The two numbers stack, and reading the second without the first flatters a rare record. The same best day count means one thing against 381 active days and another against 5,533.

Four audits cannot answer

Market Anomalies EA, Pulse Engine, Quantum Emperor and Waka Waka carry the active count, at 203, 229, 131 and 170 days, but no denominator beside it. Their audit files were built by engine version 0.2.0, before the window count existed, and we hold all four as unlisted so they are not rebuilt on the way past. No share can be quoted for them, and inventing one from the window text would be a calculation of ours rather than a field of theirs.

Run these checks yourself

Read the three fields, not the sentence. They sit in the concentration block of the audit JSON as trading days, window trading days and active share.
Watch which field you grabbed. A plain text search for trading days also finds a field in the end of day block that counts something else entirely.
Check the holding time before you call a robot quiet. A position that stays open for a week counts on its closing day only.

Every count above is public in the audit files linked in place. Holding your own tester report? The browser check is free.

The honest limits. A day counts as active when a trade closed on it, so a robot that holds positions for weeks looks quieter here than it was. The denominator counts every weekday in the window and knows nothing about market holidays, so each share above is slightly too low rather than slightly too high. Windows differ in length, with OilVector X measuring 2,325 weekdays against 6,075 or 6,076 for the records that start in 2003, so these shares compare as ratios and never as day counts. Quantum Queen and Quantum Queen X are the same measured run published under two slugs and both record 820 of 6,076, which makes 22 audits but 21 distinct runs. And the share counts days rather than trades, so a day with one trade and a day with forty count the same.