State what is being tested
Define the selection rule and what “stable” means before looking at the held-out evidence.
Examples / evidence stories
Follow the research question all the way to its limitations.
Examples connect isolated concepts into complete reasoning chains. Public examples use synthetic or explicitly publishable data and emphasize interpretation rather than headline results.
Example outputs are illustrative. They are not recommendations, forecasts, or demonstrations of future performance.
Evidence story / never end at the winner
01Questiondefine before results
02Selectionshow the family
03Holdoutreserve evidence
04Validationstress the claim
05Interpretwhat changed?
06Limitationswhat remains?
Example 001 · synthetic
Does a selected candidate remain stable outside the window that selected it?
This first complete public case study follows a synthetic selected candidate from development through reserved evidence, time-aware checks, uncertainty analysis, limitations, and the next research question.
Separate development from evaluation
Use synthetic candidate data and preserve a holdout boundary so the example can teach the difference clearly.
Show the population
Inspect the selected candidate in the context of the broader family instead of presenting one result in isolation.
Apply time-aware checks
Compare multiple held-out windows and sequence-sensitive diagnostics rather than treating one split as definitive.
Explain what changed
Describe where the evidence agrees or disagrees and avoid converting a diagnostic into a certainty claim.
Name what the test cannot prove
Document selection effects, assumptions, data limitations, and the next question that remains unanswered.
Read Example 001 end to end
Follow the complete chain instead of stopping at the setup skeleton.
Open the synthetic case study →Example 002 · synthetic / interactive
Can a replay chart accidentally know the future?
Move a synthetic historical cutoff and compare an information-safe chart with an explicitly invalid view that leaves future bars visible.
Hide what was not knowable yet
See how the admissible chart shrinks when the replay cutoff moves backward.
Make leakage visible
Future bars stay faintly visible only in the comparison panel so the error is obvious.
Explore Example 002
Use the slider or keyboard, then follow the Method and How-To links.
Open the replay-boundary example →Why examples need their own area
Docs answers “what does this do?” and Methods answers “what does this mean?” Examples answer “how do these ideas fit together in a complete research argument?” That is a different job and deserves a dedicated format.