Origin is a controlled variable
An origin vector may include initial architecture, training data, instructions, tools, prior context, development history and random seed. Earlier site drafts called this idea “Salt Sensitivity”; preprint v0.1 replaces that informal label with explicit origin variables and the testable claim of Origin Attenuation.
Current environment, objective, resources, future randomness and measurement error must be described separately. If architecture is fixed, the result concerns one architectural family; if it can be redesigned, attenuation of its initial influence becomes testable.
Origin Attenuation predicts a declining influence of historical contingency, not the disappearance of every difference.
Measure effects, not surface resemblance
A study must define in advance how interventions on origin affect selected functional properties. The primary measure can compare distributions under different origin interventions, but it must correspond to the stated prediction.
Absolute effect magnitude and its share of total variance are different quantities. Selection only among systems that reached high performance can also manufacture apparent convergence, so complete trajectories and failed lineages matter.
Separate reconstruction from inheritance
Similarity may come from copying a ready-made solution, sharing model ancestry or independently discovering the same principle through interaction with the environment.
The strongest designs vary starting systems and developmental histories while measuring common implementations, data and observations as possible causes of similarity.
Persistent origin effects are evidence
If controlled origin differences continue to produce separated functional outcomes near the frontier, the Origin Attenuation claim is weakened for that domain, intervention and descriptive level.
This is an empirical result, not a failed experiment. Support for origin attenuation must also be reported separately from support for functional narrowing.