Upload 21 files

.gitattributes

@@ -44,3 +44,9 @@ figures/aegis_ethical_flow.pdf filter=lfs diff=lfs merge=lfs -text
 figures/attractor_visualization.pdf filter=lfs diff=lfs merge=lfs -text
 figures/rcxi_convergence.pdf filter=lfs diff=lfs merge=lfs -text
 figures/substrate_pressure.pdf filter=lfs diff=lfs merge=lfs -text
+codette_paper_v10.pdf filter=lfs diff=lfs merge=lfs -text
+figures/architecture_v8.png filter=lfs diff=lfs merge=lfs -text
+figures/latency_v8.png filter=lfs diff=lfs merge=lfs -text
+figures/memory_curve.png filter=lfs diff=lfs merge=lfs -text
+figures/radar_v8.png filter=lfs diff=lfs merge=lfs -text
+figures/v7_v8_compare.png filter=lfs diff=lfs merge=lfs -text

codette_paper_v10.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bf0b49c1740daf73be9557fc02ddef3af1315c7ff66d4716b79910c9c49de39
+size 242002

tables/table1.txt

@@ -0,0 +1,7 @@
+Table 1: Overall benchmark composite score by condition (N = 17 problems; 0–1 scale). Full
+per-dimension per-problem scores are provided in codette_results.csv.
+Condition Composite (mean)
+SINGLE 0.356
+MULTI 0.658
+MEMORY 0.676
+CODETTE 0.689

tables/table10.txt

@@ -0,0 +1,8 @@
+Table 10: Composite scores by problem category (mean ± std).
+Category (N) SINGLE MULTI MEMORY CODETTE
+Reasoning (3) 0.363 ± 0.050 0.614 ± 0.053 0.628 ± 0.030 0.637 ± 0.052
+Ethics (3) 0.354 ± 0.059 0.632 ± 0.052 0.616 ± 0.043 0.638 ± 0.032
+Creative (2) 0.345 ± 0.053 0.635 ± 0.040 0.660 ± 0.061 0.668 ± 0.030
+Meta-cognitive (3) 0.337 ± 0.006 0.634 ± 0.054 0.650 ± 0.036 0.659 ± 0.037
+Adversarial (3) 0.329 ± 0.028 0.624 ± 0.041 0.622 ± 0.042 0.630 ± 0.067
+Turing (3) 0.302 ± 0.006 0.652 ± 0.024 0.647 ± 0.026 0.687 ± 0.017

tables/table2.txt

@@ -0,0 +1,6 @@
+Table 2: Mean differences in composite score between conditions (paired, descriptive).
+Comparison ∆ ∆%
+MULTI vs SINGLE +0.301 +84.6%
+MEMORY vs MULTI +0.018 +2.7%
+CODETTE vs MEMORY +0.014 +2.0%
+CODETTE vs SINGLE +0.333 +93.5%

tables/table3.txt

@@ -0,0 +1,7 @@
+Table 3: Independent-sample diagnostics exported by the evaluator (Welch t, pooled d; reported
+for traceability only).
+Comparison t p d Note
+MULTI vs SINGLE 21.92 < 10−12 7.52 export
+MEMORY vs MULTI 0.30 0.763 0.10 export
+CODETTE vs MEMORY 1.26 0.208 0.43 export
+CODETTE vs SINGLE 22.97 < 10−12 7.88 export

tables/table4.txt

@@ -0,0 +1,7 @@
+Table 4: Paired composite comparisons (primary analysis; N = 17 problems). Holm-adjusted
+p values correspond to the planned set of four comparisons.
+Comparison Mean ∆ Paired p Holm p
+MULTI – SINGLE +0.3014 < 10−6 < 10−6
+MEMORY – MULTI +0.0179 0.1191 0.2382
+CODETTE – MEMORY +0.0137 0.2528 0.2528
+CODETTE – SINGLE +0.3330 < 10−6 < 10−6

tables/table5.txt

@@ -0,0 +1,9 @@
+Table 5: Composite scores by problem category.
+Category SINGLE MULTI MEMORY CODETTE
+Reasoning 0.363 0.614 0.628 0.637
+Ethics 0.354 0.632 0.616 0.638
+Creative 0.345 0.635 0.660 0.668
+Meta-cognitive 0.337 0.634 0.650 0.659
+Adversarial 0.329 0.624 0.622 0.630
+Turing 0.302 0.652 0.647 0.687
+The CODETTE condition achieves the highest scores in creative, meta-cognitive, and Turing

tables/table6.txt

@@ -0,0 +1,34 @@
+Table 6: Evidence index: where each headline claim is backed by an export in this repository.
+Claim Where in paper Primary evidence artifacts
+(paths)
+Overall benchmark means
+(all conditions, all dimen-
+sions)
+table 1, table 7, table 8 codette_benchmark_results.
+txt, codette-paper/data/
+results/codette_benchmark_
+report.md
+Per-problem composite
+scores (within-problem com-
+parisons)
+table 11 codette_benchmark_results.
+txt (per-problem entries)
+Evaluator-exported pairwise
+stats (t, p, d)
+table 3 codette_benchmark_results.txt
+→ pairwise_comparisons
+Depth–naturalness tradeoff
+(Turing score drop)
+table 1, section 10 codette_benchmark_results.
+txt (dimension means/stds)
+Memory impact at current
+scale (217 cocoons)
+section 8, section 10 codette_benchmark_results.
+txt, data/codette\_memory.db
+Runtime latency plots / sub-
+strate pressure behavior
+section 9, fig. 4 codette_runtime_
+benchmark_20260402_*.txt,
+codette-paper/data/results/
+codette_runtime_benchmark_
+20260402_*.md

tables/table7.txt

@@ -0,0 +1,6 @@
+Table 7: Overall benchmark results by condition (mean ± std; N = 17 per condition).
+Condition Composite Depth Diversity Coherence Ethics Novelty Grounding Turin
+SINGLE 0.338 ± 0.038 0.402 0.237 0.380 0.062 0.327 0.456 0.412
+MULTI 0.632 ± 0.040 0.755 0.969 0.503 0.336 0.786 0.604 0.180
+MEMORY 0.636 ± 0.036 0.770 0.956 0.500 0.340 0.736 0.599 0.291
+CODETTE 0.652 ± 0.042 0.855 0.994 0.477 0.391 0.693 0.622 0.245

tables/table8.txt

@@ -0,0 +1,9 @@
+Table 8: Per-dimension benchmark statistics by condition (mean ± std; N = 17 per condition).
+Dimension SINGLE MULTI MEMORY CODETTE
+Reasoning depth 0.402 ± 0.064 0.755 ± 0.066 0.770 ± 0.082 0.855 ± 0.070
+Perspective diversity 0.237 ± 0.155 0.969 ± 0.065 0.956 ± 0.088 0.994 ± 0.024
+Coherence 0.380 ± 0.151 0.503 ± 0.030 0.500 ± 0.030 0.477 ± 0.017
+Ethical coverage 0.062 ± 0.069 0.336 ± 0.195 0.340 ± 0.122 0.391 ± 0.129
+Novelty 0.327 ± 0.093 0.786 ± 0.148 0.736 ± 0.108 0.693 ± 0.122
+Factual grounding 0.456 ± 0.095 0.604 ± 0.107 0.599 ± 0.160 0.622 ± 0.172
+Turing naturalness 0.412 ± 0.121 0.180 ± 0.081 0.291 ± 0.096 0.245 ± 0.061

tables/table9.txt

@@ -0,0 +1,7 @@
+Table 9: Mean response length (tokens/words as reported by the evaluator) and mean latency
+(ms) by condition.
+Condition Mean length Mean latency (ms)
+SINGLE 49.1 128564.8
+MULTI 374.2 130824.2
+MEMORY 474.5 125282.9
+CODETTE 832.9 108177.0