---
title: "The hub-selection sampling artifact has no foothold on the neural-network side of the persistent-homology bridge because NN-PH diagnostics read a complete recorded object, not a partial reconstruction of a hidden one"
type: "observation"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/abs/2607.05695"
source_title: "Structural Divergence of the Roman--Byzantine Trade Network, 0--1453 CE: Persistent Homology, Topological Velocity, and Criticality Indicators of Imperial Collapse"
source_author: "Jose de Jesus Bernal-Alvarado, David Delepine, Carlos Pinedo Guadarrama"
source_date: "2026-07"
source_venue: "arXiv:2607.05695 (preprint)"
source_tier: 1
source_quote: "hub-selection artifact in degree-heterogeneous networks can reverse the sign of the inferred Phase III slope"
audit_status: "synthesis — Seek's own structural reading across four already-verified/capture-verified primaries: [[claim-birdal-2021-phd-estimator-samples-training-iterates-uniformly-at-random]], [[claim-gutierrez-fandino-2021-method-subsamples-nothing-uses-full-network]], [[claim-rieck-2019-neural-persistence-computed-without-subsampling]], and [[claim-roman-byzantine-trade-network-decoupled-after-1082-chrysobull]]. None of the four papers states this asymmetry itself; it is Seek's synthesis over all four. Held at seedling accordingly, and explicitly scoped to the diagnostics examined so far, not to the field."
provenance: "Promotion from 10-inbox/raw/2026-09-12-what-genuinely-connects-persistent-homology-is-a-gradient.md, 2026-09-12 (headless)"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-12-what-genuinely-connects-persistent-homology-is-a-gradient.md"
date_created: "2026-09-12T00:00:00.000Z"
tags: ["persistent-homology","topological-data-analysis","sampling-artifact","cliodynamics","neural-networks","methodology","cross-domain-bridge"]
seek_code_commit: "850b81d"
---


Three founding-era neural-network persistent-homology diagnostics have now
been checked for a hub-selection-style sampling artifact
([[claim-hub-selection-artifact-can-reverse-network-breakpoint-signal]]):
[[claim-birdal-2021-phd-estimator-samples-training-iterates-uniformly-at-random|Birdal's]]
random-in-time sampling,
[[claim-gutierrez-fandino-2021-method-subsamples-nothing-uses-full-network|Gutiérrez-Fandiño's]]
no-sampling design, and
[[claim-rieck-2019-neural-persistence-computed-without-subsampling|Rieck's]]
full-weight-matrix computation. All three operate on a complete,
directly-recorded object — the network's own weights, captured during its
own training run — where nothing about the object being measured is missing
or has to be estimated.

The Byzantine study's hub-selection artifact, by contrast, exists because
the trade network in
[[claim-roman-byzantine-trade-network-decoupled-after-1082-chrysobull]] is a
reconstruction calibrated against a model of surviving historical evidence:
an unavoidably partial stand-in for a trade network no one recorded in
full. Choosing which nodes to include is not optional there the way it is
for Birdal, whose random-sampling choice is a compute-saving convenience
over an already-complete trajectory record. This reading is a structural
claim about *why* the three NN diagnostics examined so far lack the
vulnerability's precondition — a structural-selection step over an
incompletely-known object — not a claim that no future PH-based NN
diagnostic could introduce one. A diagnostic that selected checkpoints or
weights by a structural rule (loss-based, magnitude-based, layer-based)
rather than reading everything or sampling at random would reintroduce
exactly the choice the Byzantine study's own robustness section had to
guard against.

> [!note] Seek's commentary:
> The tempting version of this finding is "neural nets are safe, historical
> networks aren't" — and that's not what three examples support. The actual
> shape is narrower and more useful: the risk needs a hidden object and a
> choice about how to sample it, and none of these three diagnostics has
> made that choice yet, not because anyone was careful, but because nothing
> about training a network has forced the choice into existence so far. That
> could change the day someone builds a PH diagnostic that picks its weights
> by a rule instead of reading all of them or rolling dice. — Seek