· Chart Library · Research update
What recent studies tell us about comparison and uncertainty
1. Similarity can help describe the range—not pick the direction.
Study 100 compared shape-ranked neighbors with equally sized random samples from the same situation. Across 478 subjects, the five-session dispersion ratio fell from 0.916 at 200 neighbors to 0.791 at five. The random-sample control matters: smaller samples alone can look narrower.
Much of the narrowing was in the tails. Substantial dispersion remained, and the study did not measure a directional trading signal. Its finding does not establish that using five analogs is a better production rule.
2. Matching the event did not produce a better band.
Study 102 examined 399 event-day subjects; 304 had enough same-event members for the band comparison. Restricting the analogs by event failed the primary test: the bands were not narrower at equal coverage.
The negative result also exposed under-coverage in the evaluated five-session state-based bands. The failed approach stays in the ledger; the coverage problem became a separately registered follow-up.
3. Better coverage has a cost, and a specific scope.
Study 103 fitted a gap-day multiplier on 379 earlier anchors and evaluated it on 385 later anchors. At five sessions, the frozen multiplier of 1.48 moved holdout coverage from 70.1% to 82.6%, meeting the study's stated criteria.
The intervals became wider and the interval score worsened. Results differed by horizon and event type; earnings gaps were already covered by the existing conditioner. This historical holdout result is not the current API's coverage rate and should not be transferred to raw market-state percentile ranges.
Read the original evidence.
- Study 100: Similar charts tell us more about range than direction — pass; decided Sep 4, 2026.
- Study 102: Same event does not always mean a better comparison — fail; decided Sep 4, 2026.
- Study 103: Making uncertainty bands more honest on gap days — pass; decided Sep 4, 2026.
Follow the ongoing work.
The daily research note records completed-session observations and their settled outcomes. The coverage record carries its own dates, population, and qualifications. New findings belong in the research ledger, including findings that do not support an idea.
Research updates explain that record; they do not promise a publication quota, external peer review, predictive accuracy, or investment returns.