# Study 000J Results

## Primary Answer

`State-conditional answer, supported.`

When embedding support and bout length diverged in the directly observed archive, `bout structure` carried more weight in preserving low-burden expression. Inside the later specialized state, `embedding support` remained a strong organizer, but the archive does not support one global winner across all states.

## Direct Divergence Case

- Early long outside expression:
  - distance: `10.2265 miles`
  - running-share context: `28.2459%`
  - treadmill neighbors: `3.0`
  - HR residual burden: `-3.7503 bpm`

- Early short outside probes:
  - mean distance: `2.9819 miles`
  - mean running-share context: `11.3091%`
  - mean treadmill neighbors: `4.0`
  - mean HR residual burden: `19.3947 bpm`

This is the cleanest direct separation in the archive. The early long bout stayed low burden even though embedding was still only low-to-moderate, while the shorter early probes were costly.

## Later Specialized State

- October 2025 short low-support cluster:
  - mean distance: `2.091 miles`
  - mean running-share context: `5.1299%`
  - mean treadmill neighbors: `1.75`
  - mean HR residual burden: `13.8844 bpm`

- April 2026 boundary expression:
  - distance: `5.4541 miles`
  - running-share context: `91.0839%`
  - treadmill neighbors: `32.0`
  - HR residual burden: `0.6887 bpm`

In the later specialized state, embedding and bout length both moved in the favorable direction together, so the archive supports both but does not fully isolate them there.

## Correlation Ranking

- Across all outdoor cases:
  - distance vs burden: `r = -0.5985`
  - running share vs burden: `r = -0.4895`
  - treadmill neighbors vs burden: `r = -0.3636`
  - composite support vs burden: `r = -0.4265`

- Inside the later specialized subset:
  - distance vs burden: `r = -0.8135`
  - running share vs burden: `r = -0.8433`
  - treadmill neighbors vs burden: `r = -0.8418`
  - composite support vs burden: `r = -0.8426`

## Interpretation

`000J` answers a narrower mechanism question than `000H`.

It does not say embedding does not matter.
It says:

- bout length carried the clearer protective weight in the directly observed divergence case
- embedding remained highly relevant inside the later specialized state
- the archive therefore supports a two-factor, state-conditional model rather than one universal factor ranking