# October 28, 2026 — Evening: a scout-written field heard from the outside

**Source:** [Evening](https://github.com/anttiluode/Evening) · [live experiment](https://anttiluode.github.io/Evening/) · [pre-registration 1](https://github.com/anttiluode/Evening/blob/main/PREREG.md) · [pre-registration 2](https://github.com/anttiluode/Evening/blob/main/PREREG_INHERIT.md) · [README / result ledger](https://github.com/anttiluode/Evening/blob/main/README.md)

## An actual seam between two research branches

[BlackBoxLab](https://github.com/anttiluode/BlackBoxLab)/Cell 0 had local scouts that changed a slow, conserved-budget material field. An August history experiment read the final map directly. [WallMemoryLaw](https://github.com/anttiluode/WallMemoryLaw) had a constrained-wave readout and finite-speed scattering-tour logic, but its boundary memories were assigned externally. Evening copies **Cell 0's writer**, removes the scouts, freezes the bulk mass term `V = 8M`, and lets a wave pulse interrogate it without access to the map. This is a cross-project seam, **not** evidence that WallMemoryLaw's wall calibration theorem transfers to distributed material or that the wave paper proves inheritance.

## What survived: the order reader

The pre-registered order comparison uses 24 fresh seek→avoid and 24 avoid→seek worlds with matched total material. A 16-pair broadband ping reader reports **1.000 leave-one-out accuracy** (randomization p floor 0.0005); eight pairs also reach 1.000, one pair 0.833. Mean material reaches **0.625**. A finite-speed ellipse test confirms that changing material outside the causal footprint leaves the echo essentially unchanged. These are **source-reported simulation receipts**, not independent reruns in this atlas review.

The key correction is *what* the receiver identifies: **a distribution of material values**, not uniquely a spatial arrangement. Phase-scrambling and even pixel-shuffling leave the order score at 1.000. Full-map quantiles alone reach 1.000. The receiver demonstrates a blind external readout; not a unique wave-specific or geometry-sensitive encoding advantage.

## What failed: recency and inheritance

The separate 32-world recency comparison fails its ping criterion: **0.625, p=0.18**, against **0.812** from the trivial mean-material attacker. This reverses the older 12-world suggestion that mean amount could not detect recency.

Part 2 asks if the only intergenerational channel, **material → local ping → frozen receptor → child schedule**, can reproduce two types. Eight all-avoid and eight all-seek lineages are tested for eight generations under two receptors. Individual sites retain **0/8** all-avoid vs **8/8** all-seek; designed quorum voting retains **2/8** all-avoid vs **8/8** all-seek. The mixture map shows **5% seek-first parents → 37.9% seek-first children**. The current loop has a winner, **not two robust inherited states**. The quorum is an intervention, not an emergent evolutionary property.

## The open next rung

Evening's latest README **corrects a former overclaim**: a sharper receptor could in principle make the avoid-first basin stable. Its Gaussian noise fit suggests nine independent analog pings averaged **before thresholding** might create a low-p basin; spatial correlations could erase that gain. This is *derived*, **not tested**. An alternative is two positive writing channels with discriminable ping responses. The right new experiment must beat threshold-vote and direct-local-material controls while keeping equal readout, writing, and calibration budgets. A two-presence success would be a synthetic heredity mechanism, not a biological or general learning theorem.

## Genealogy interpretation

- **Code inheritance:** BlackBoxLab → Evening (Cell 0 writer).
- **Method convergence:** WallMemoryLaw → Evening (bounded wave interrogation and causal footprint); not a code fork.
- **Keeps:** a reproducible blind observer for a locally written history.
- **Corrects:** arrangement-only history and scalar-blind recency.
- **Kills for now:** two-attractor heredity under the tested one-material/receptor loop.

Links to raw code, protocols and tabulated measurements remain in Evening's README. Neither this curation nor the new pass independently reran the upstream numerical experiments.