khipu-computational-toolkit

Phase 11: Color × Value Interaction

Generated: 2026-03-08
Database: K-CAT SQLite database (built from KFG source data)
Script: scripts/run_phase11_color.py
Inputs: data/kfg/khipu_database.db · Phase 7/8 assignments
Status: ✅ Complete


Research Question

Does cord color associate with value order-of-magnitude, hierarchy level, summation compliance, attachment type, or behavioral cluster?

Five hypotheses tested. All five significant at p < 0.001.


Methods

Color encoding

The KFG cord_colors table stores one or more color codes per cord in sequence_ord order. The primary color (sequence_ord = 0) is the outermost/dominant color. Composite colors (e.g., W%AB, MB-KB) indicate banded or variegated cords.

Only cords with a primary color code occurring ≥ 50 times in the corpus are retained (59 qualifying codes). Tests use non-zero values only (zero-value cords excluded per Phase 10).


Results

H1 — Primary color predicts value order-of-magnitude

Kruskal-Wallis: H = 1,419.0, p = 3.6 × 10⁻²⁶⁷

Primary color n (non-zero) Median value ≥ 100 % ≥ 1,000 %
W%AB (banded) 83 126.0 54.2% 10.8%
DB-CB (banded) 44 125.0 97.7% 0.0%
MB-KB (banded) 37 90.0 48.6% 13.5%
YG 224 80.5 45.5% 4.9%
W%MB (banded) 101 70.0 41.6% 4.0%
NB 801 42.0 32.7% 9.0%
W (white) 14,443 13.0 17.9% 4.5%
B (brown) 2,254 6.0 10.5% 1.4%

Composite/banded colors consistently carry higher median values than monochrome codes. The W%AB, DB-CB, and W%MB medians are ≥ 70, compared to plain W at 13. Banded-color sample sizes are small (37–101 cords).

H2 — Color distribution shifts by hierarchy level

Chi-square (level-0 vs level-1, top-15 colors): χ² = 1,267, df = 14, p = 7 × 10⁻²⁶²

Color Level-0 % Level-1 % Level-2 %
W (white) 37.0 30.3 31.2
YB 8.6 4.8 3.4
KB 2.2 6.3 6.0
AB 15.4 15.8 21.4
B 5.1 4.5 2.3

Color frequencies are not uniform across hierarchy levels. YB decreases from 8.6% at level 0 to 3.4% at level 2. KB increases from 2.2% at level 0 to 6.0% at levels 1–2. AB increases from 15.4% to 21.4% at level 2.

H3 — Color in summation-compliant groups

The 680 compliant parent-child groups show a parent-child same-color match rate of 30.6%, compared to 27.7% in the sub (non-compliant) class. The difference is modest (2.9 percentage points). W, AB, and MB dominate both compliant and non-compliant groups, reflecting corpus-wide prevalence.

H4 — Attachment type modulates the color-value signal

Kruskal-Wallis color × value is significant for all three attachment types:

Attachment H-statistic p-value
U (upward) 1,285.6 2.0 × 10⁻²³⁸
R (recto) 695.0 5.4 × 10⁻¹²⁷
V (verso) 521.1 1.3 × 10⁻⁹¹

V-attached cords record higher median values than U-attached cords for the same color (e.g., AB: V = 14, R = 11, U = 5).

H5 — Color composition differs by behavioral cluster

Chi-square: χ² = 3,482, df = 70, p ≈ 0

Cluster Notable color feature
B1 High LB (13.4% vs corpus 2.0%)
B2 Elevated MB (16.9%) and YB (12.1%)
B3 Elevated B (7.3%)
B5 Highest W (43.8%)
B6 Elevated B (8.1%) and NB (5.1%)

Limitations


Outputs

File Description
data/processed/phase11_color_value.csv Per-cord: primary color, value, hierarchy level, behavioral label
data/processed/phase11_color_stats.csv Per-color: n, median, % ≥ 100, % ≥ 1,000
visualizations/phase11/color_value_boxplot.png H1: color × log-value boxplot
visualizations/phase11/color_by_level.png H2: color distribution at levels 0, 1, 2
visualizations/phase11/color_compliance.png H3: color in compliant vs non-compliant groups
visualizations/phase11/attachment_color.png H4: color × value by attachment type
visualizations/phase11/color_cluster_heatmap.png H5: color composition heatmap across behavioral clusters

Corpus sweep run against K-CAT SQLite database. Re-run with scripts/run_phase11_color.py to refresh.