# v17e postrun diffusion diagnosis

Status: `v17e_within_family_diffusion_without_cross_start_convergence`.

## Input boundary

This is a descriptive aggregation of frozen v17e outputs. It reruns no chain and computes no source spectrum or observed-effect statistic.

- scale-response SHA-256: `e5d7ae295c6332a4fb6932abaca13f95818d30bdc3608bdf3f774028951708d0`
- feature-response SHA-256: `1259a6d05c6ab3918554ed9dca0e038fc107155918dcec75d787a74619410a2e`
- residual-audit SHA-256: `f5e1ca4daf95cd2fe0838e2802e2bd6edf82e275bd37e1f1325846ad2f497959`
- proposal-overlap SHA-256: `d8839ec47d40deb569f3dcdc2a45c3693224a8e4c1054089d242c77385274b31`

## Diagnosis

All `6/6` sources match the disclosed descriptive pattern: cross-start distance stays within the postrun near-flat band while within-start dispersion expands by at least 25 percent. Cross-start scale/baseline ratios span `0.978973-1.005348`, while within-start dispersion ratios span `1.385802-1.470668`. Cross/within separation ratios fall to `0.673283-0.725462` of baseline because the within-family clouds spread, not because absolute cross-start distance materially contracts.

Source-edge and concrete-conflict gaps contract in `12/12` source-feature cells. Candidate-rank gaps contract in `3/6`. Exact residual-profile identity remains in `6/6`. These facts show coarse-coordinate movement and stable residual algebra alongside persistent state-level start memory.

## Decision

The v17e stop rule stands: do not spend more budget scaling the length-2-to-4 kernel. The next effect-blind gate should change the move class while preserving an explicit stationary target, exact reverse accounting, representation checks, and a matched realized-work comparison. Longer alternating cycles or reversible compound-cycle moves are candidates; neither is qualified by this diagnosis.

## Claim limit

The descriptive thresholds in this report were selected after the formal v17e result and are not a preregistered gate. The pattern does not prove disconnected state-space components, convergence, slow mixing, a canonical measure, or any source-spectrum or physics effect.