# v17d next direction

Formal status: `v17d_endpoint_centers_not_stable`.

Do not inspect the source spectrum. Preregister one bounded effect-blind scale-response gate on the same six spaces, both starts and independent seeds. Use substantially longer checkpointed chains and make direct cross-start endpoint distance the primary response. Retain center features, residual profiles and proposal footprints as diagnostics.

Rationale: seed and early/late distance comparisons passed, source-edge/conflict gaps contracted in `12/12` cells, traversal/resource passed `24/24`, and residual partitions were exactly identical within source. Yet direct cross-start distance remained essentially flat and start-family distance agreement failed `0/6`. This justifies one scale test, not indefinite budget growth.

Stop rule: if direct cross-start distance does not contract materially at the longer checkpoint, retire further scaling of the length-2-to-4 kernel and change the move class. Source effects and physics claims remain closed either way until finite start memory is resolved.