README.txt

========================================================
ADVERSARIAL 5D VECTOR CONSTRUCTION PIPELINE
========================================================

Purpose
--------------------------------------------------------

This stage constructs proxy structural 5D vectors for the
entire adversarial corpus after canonicalization.

The vectors are used for:

    • structural similarity testing
    • admissibility-space comparison
    • falsification experiments
    • separation-margin analysis
    • real-vs-random discrimination testing

The pipeline operates on canonical adversarial ladders.

========================================================
INPUT
========================================================

The script automatically scans:

    CLE_PILOT_I\adversarial\raw\canonical\

including:

    uniform\
    pareto\
    randomwalk\
    shuffle\

Expected corpus size:

    650 canonical ladders

========================================================
OUTPUT FILES
========================================================

1.

    adversarial_STRUC_5D_proxy_vectors.csv

Full ladder-level vector table.

Contains one row per ladder.

========================================================

2.

    adversarial_STRUC_5D_proxy_summary.csv

Aggregated statistical summary by adversarial class.

========================================================
5D VECTOR STRUCTURE
========================================================

Each ladder receives:

    v(L) = (

        P_depth,
        sigma_GR^2,
        frag_rate,
        adm_persist,
        aniso_persist

    )

Where:

--------------------------------------------------------
P_depth
--------------------------------------------------------

Proxy persistence depth.

Measures structural continuity and gap persistence.

--------------------------------------------------------
sigma_GR^2
--------------------------------------------------------

Gap-ratio variance.

Measures instability of local spacing structure.

--------------------------------------------------------
frag_rate
--------------------------------------------------------

Fragmentation rate.

Measures discontinuity intensity.

--------------------------------------------------------
adm_persist
--------------------------------------------------------

Admissibility persistence.

Measures persistence of stable admissible regions.

--------------------------------------------------------
aniso_persist
--------------------------------------------------------

Anisotropic persistence.

Measures directional rigidity stability.

========================================================
HOW TO RUN
========================================================

Open PowerShell inside:

    cle_v2_0_0\experimental artifacts

Run:

    python construct_adversarial_5d_vectors.py

No arguments are required.

========================================================
PIPELINE POSITION
========================================================

This is STEP 4 of the adversarial falsification pipeline:

    STEP 1
    Generate synthetic ladders

    STEP 2
    Canonicalize ladders

    STEP 3
    Run STRUC-PERC-I

    STEP 4
    Construct 5D vectors   <-- CURRENT STAGE

    STEP 5
    Compute similarity matrices

========================================================
EXPECTED RESULT
========================================================

The resulting vectors should allow testing whether:

    real physical ladders

remain structurally separable from:

    synthetic adversarial ladders

inside admissibility-space geometry.

========================================================