Almost every performance number you encounter has been produced by someone with an interest in how it looks. That does not make them dishonest. It means the number needs structure around it before it can carry weight.
The first requirement is that the rule was fixed in advance. If a strategy can be adjusted while it runs, then a good result may just be a series of timely corrections, and no amount of history distinguishes skill from steering.
The second is that the original claim survives unedited. Prices get restated as dividends and splits are applied, so re-deriving an old backtest today quietly moves the target. A claim stored on the day it was made cannot drift toward whatever happened next.
The third is the one most often missing: the failures have to stay visible. A list that only shows what worked is not a track record — it is a survivor list, and it will always look excellent no matter what the underlying process is.
A restaurant that deletes its bad reviews still has five stars. The rating has not improved; it has just stopped measuring anything.
There is a fourth issue that is subtler and catches careful people too: how the history itself was assembled. If you test a strategy on the companies that exist today, you have quietly excluded everyone who went bust, and the past will look far kinder than it was.
The honest alternative is to treat each holding as having a lifespan — it enters the test on the date it genuinely became investable and leaves if it stopped trading. That is called point-in-time, or cohort, simulation, and it is what lets a rule be tested over decades even when its youngest holding is only a few years old.
- Fixed in advance — the rule cannot be edited once the clock starts.
- Claim preserved — the original figures are stored, not recomputed later.
- Failures retained — retired strategies stay listed, with the reason.
- Point-in-time data — holdings join when they actually existed, so the dead are not excluded.
- Long enough to read — a few months of live data is noise, not a record.
Notice that none of these is about being clever. They are all about removing the ways a number can flatter itself after the fact. That is the whole job.
Finisdom’s Fund Foundry enforces all four structurally rather than by good intentions: publishing fingerprints the rule and freezes it permanently, the backtested claim is stored at publication and never recomputed, published funds can be retired but never deleted, and the simulator treats every member as a lifespan. The public leaderboard ranks funds on the share of their claim they kept, not on their returns.
What is survivorship bias?
Judging the past using only the things that made it to the present. If you test a stock strategy on companies that exist today, every company that went bankrupt is silently excluded, and the results look far better than the strategy would actually have done at the time.
Why rank funds on kept promises instead of returns?
Ranking on returns mostly surfaces whoever took the most risk in the luckiest window. Ranking on retention — how much of the backtested claim actually arrived after publication — asks the one question a backtest cannot answer about itself.
Why can a published rule never be edited?
Because the record belongs to the rule that earned it. If the rule could change, the history would describe a strategy that never actually ran. Changing your mind should start a new record from zero rather than inherit the old one.