The decision at the end of an experiment — kept or discarded — is almost the least useful thing to remember about it. It's a single bit of information. What's actually worth carrying forward is the specific reason behind that bit, and getting that reason out of a finished result takes a deliberate step of its own.
The decision isn't the knowledge
"It didn't work" is a decision, not knowledge. It tells you nothing about the next idea, even a closely related one. "It didn't survive realistic slippage at this holding period" is knowledge — specific enough to change how the next similar idea gets approached, because it names the actual mechanism that broke this one, not just the outcome.
The difference matters because the vague version is what's left if nobody deliberately extracts the specific one. A result that gets closed with "well, that one didn't pan out" and nothing more precise has quietly thrown away almost everything worth keeping from the time spent on it.
What's actually in a finished result, worth looking for
A .reamer file already holds the reproducible detail this requires — the closed trades, the execution config that was actually used, the specific parameters tried. None of that automatically becomes an insight just by existing. Going back through it looking for a specific, nameable reason — a slippage assumption the strategy couldn't absorb, a regime the pattern only appeared in, a cost that ate the whole apparent edge — is a step someone still has to take. The record makes that possible to do accurately. It doesn't do it by itself.
A useful test for whether knowledge was actually captured
If the lesson from a discarded idea can be stated as one specific sentence — naming the mechanism, not just the outcome — it's usable on the next idea. If the only thing left is a vague sense that "something like this didn't work before," it isn't knowledge yet, no matter how much time went into the experiment that produced it. That vague version doesn't transfer; it just becomes a source of unfounded hesitation the next time something similar comes up.
Why this is worth treating as its own step
It would be convenient if knowledge were just a natural byproduct of running enough experiments. It isn't, quite — a finished result contains the material for an insight, but the insight itself has to be pulled out on purpose, in specific enough language to actually be useful later. Skipping that last step is how a research process can run for a long time and still not feel like it's accumulating anything.
Full reference: docs · The workflow this maps onto: reamerlabs.com