Prop Research Bias: A Stop Chosen After the Outcome Needs a New Test
A stop chosen after seeing a price path can make historical research look better without demonstrating a usable decision rule. Preserve the original parameter and selection time. Treat the improved retrospective version as a new hypothesis requiring its own evidence.
Explore this site’s existing research resource
Freeze the original test definition
Record the dataset cutoff, entry rule, stop rule, cost assumptions and parameter-selection process. A comparison should show which information was available before each decision. Selecting a favourable level after the path is known changes the experiment.
This site’s existing look-ahead audit is the companion. A retrospective chart can help generate ideas, but its later information should not be inserted into the original record as if it were known. Keep a versioned explanation of why the new rule was chosen.
Compare a fictional parameter-selection case
Suppose an original plan uses a two-point stop. After reviewing the completed path, a researcher chooses 2.7 points because that particular path would have survived it. The second result is a conditional replay of known data, not the original plan’s observed outcome.
A future test needs the 2.7 rule fixed before its outcomes and complete cost and execution assumptions. If many alternatives were inspected, record that search as well. Reporting only the winning alternative hides the selection process and does not establish a population advantage.
| Stage | Parameter | Evidence scope |
|---|---|---|
| Original plan | 2-point stop | Chosen before supplied outcome |
| Completed path reviewed | Outcome now known | Later information available |
| Retrospective alternative | 2.7-point stop | New hypothesis from known path |
| Future fixed-rule test | Not supplied | No out-of-sample result established |
Do not equate a replay with an evaluation pass
An actual programme has account-wide conditions and an observed equity path. HashHedge’s current rules need their own review; an improved single-trade replay cannot demonstrate that every applicable requirement would be satisfied.
The optional sponsored link offers programme information. No stop is recommended or changed and no pass probability is estimated. All parameters are fictional teaching inputs. The useful result is an honest distinction between original evidence and a later research idea.
Sources and rule dates
- HashHedge: challenge types
Checked 2026-10-23. Selected published description only. Worked numbers and event records are fictional; no actual provider account, search result, live price, eligibility or payment is established. - HashHedge: terms and conditions
Checked 2026-10-23. Selected published description only. Worked numbers and event records are fictional; no actual provider account, search result, live price, eligibility or payment is established.