Remaining CorrelationStrategies and agreement display #17

New Issue

GKaszewski · 2026-08-26T09:15:34Z

GKaszewski commented

2026-08-26 09:15:34 +00:00

What to build

Adds Pearson, Kendall, and the existing mean-difference method to the CorrelationStrategy enum from #15, and builds the display that makes having several worth anything.

Agreement across strategies is the headline, not any single coefficient. A correlation that survives Pearson, Spearman, and Kendall is robust. One that appears only under Pearson is telling you about an outlier or a non-linearity, not about the User's mood.

Presenting them as a menu to choose between would make the noise problem worse rather than better — a User scanning four numbers per metric for the strongest one is p-hacking through a UI. Disagreement between strategies is information, not a choice.

Mean difference joins the enum as the strategy for binary presence, which is what the existing Activity correlation already computes. It cannot be applied to a continuous metric, and a continuous coefficient cannot be applied to Activity presence — the strategy set must reflect which inputs each can score rather than pretending all four apply everywhere.

Acceptance criteria

Pearson, Kendall, and mean difference are implemented in the enum
The existing Activity correlation runs through the strategy machinery rather than as a separate path
All strategies return the same result shape
Each strategy declares which inputs it can score; incompatible pairings are not offered
Display leads with agreement across strategies, not with any one value
Strategies are not presented as a user-selectable setting
Ranking and significance badges remain absent

Blocked by

CorrelationStrategy with Spearman, and MoonPhase as a control (#15)

## What to build Adds Pearson, Kendall, and the existing mean-difference method to the `CorrelationStrategy` enum from #15, and builds the display that makes having several worth anything. **Agreement across strategies is the headline, not any single coefficient.** A correlation that survives Pearson, Spearman, and Kendall is robust. One that appears only under Pearson is telling you about an outlier or a non-linearity, not about the User's mood. Presenting them as a menu to choose between would make the noise problem worse rather than better — a User scanning four numbers per metric for the strongest one is p-hacking through a UI. Disagreement between strategies is information, not a choice. Mean difference joins the enum as the strategy for binary presence, which is what the existing Activity correlation already computes. It cannot be applied to a continuous metric, and a continuous coefficient cannot be applied to Activity presence — the strategy set must reflect which inputs each can score rather than pretending all four apply everywhere. ## Acceptance criteria - [ ] Pearson, Kendall, and mean difference are implemented in the enum - [ ] The existing Activity correlation runs through the strategy machinery rather than as a separate path - [ ] All strategies return the same result shape - [ ] Each strategy declares which inputs it can score; incompatible pairings are not offered - [ ] Display leads with agreement across strategies, not with any one value - [ ] Strategies are not presented as a user-selectable setting - [ ] Ranking and significance badges remain absent ## Blocked by - #15

GKaszewski added the ready-for-agent label 2026-08-26 09:15:35 +00:00

GKaszewski referenced this issue

2026-08-26 09:15:49 +00:00

Benjamini-Hochberg Adjustment across two Families #18

GKaszewski closed this issue

2026-08-26 19:06:55 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: GKaszewski/k-mood#17