# v16t stability interpretation and next direction

Date: 2026-07-15

## What v16t changed

V16t tested the event-footprint null without computing the source DAG interval
spectrum or any observed/null effect statistic. The frozen matrix used six v16s
source DAGs, four protocols, and sixteen independent nulls per protocol and
source. All `384/384` perturbations passed the declared structural and
uniqueness checks.

All `12/12` nominal direct-length comparisons passed the frozen ratio gate, and
the largest center-shift ratio was `0.237211`. Direct `0.200` versus staged
`0.100 + 0.100` also passed in all `6/6` source DAGs, with maximum ratio
`1.463604` against the preregistered limit `2.0`.

The realized-effort audit changes the semantic reading. Short, reference and
long direct protocols averaged `993.145833`, `1005.020833`, and `998.343750`
accepted swaps: the common `10%` changed-edge stopping condition erased the
intended direct chain-length separation. Staged averaged `2023.343750`, about
twice direct-long, so path segmentation is confounded with extra effort. See
`v16t_realized_effort_interpretation_audit.md`.

## Current evidential position

The leading live result remains the v16s finite event-DAG full-spectrum
contrast under the qualified event-footprint null. V16t adds effect-blind
evidence that null centers stayed inside the frozen region across independent
seeds and roughly one versus two thousand accepted swaps. It does not isolate
the intended chain-length and path-segmentation effects.

V16t does not re-evaluate the v16s effect. It does not establish Markov-chain
irreducibility, mixing time, convergence, stationarity, independence,
representativeness, or uniform sampling. It also does not establish concrete
resource independence, dimension, manifoldlikeness, Lorentz symmetry,
spacetime, particles, entanglement, an invariant, or physical law.

## Recommended next gate: v16u

Run a corrected effect-blind realized-effort stability gate before constructing
a new null family. Use the current `10%` edge-change rule only as burn-in. From
each burn-in state, create branches with exact accepted-swap increments `+K`
and `+2K`, then compare direct `+2K` with segmented `+K + K` at exactly matched
effort. Keep the existing center-shift threshold `2.0`, require every source
DAG to pass, and compute no source spectrum or observed-effect metric.

If corrected realized-length and matched-effort path comparisons pass, the next
gate after v16u should be an effect-blind feasibility audit for a globally
reconstructed or matched event-DAG null that does not use the v16q local switch
path. If v16u fails, the footprint null remains procedure-conditional and the
project should not proceed to effect or scale claims.

## Why this is the highest-value move

The repair is necessary because v16t's formal pass did not instantiate its
intended direct-length contrast and did not match effort in its path contrast.
Moving directly to an independent null would leave a known, cheap design gap in
the existing sampler evidence. Correcting that gap is narrower and more
informative than scale-up, effect reruns, or dimension fitting.