analysis/report.md

128-trial task-correlation analysis

• Data root: results_128
• Posterior draws: 10,000
• Paired task-bootstrap draws: 10,000
• Posterior-propagation intervals: Jeffreys uncertainty for each task's Bernoulli success probability; the 100-task suite is held fixed.
• Task-bootstrap intervals: uncertainty from resampling the aligned 100 tasks; every modality and condition uses the same sampled indices.

Overall exact-match accuracy

Modality	Clean	Deranged	Gap	Trials/task/condition
Language	0.2996	0.1608	0.1388	128
Genome	0.3323	0.1523	0.1801	128
Integer	0.4813	0.1672	0.3141	128
Image	0.4350	0.1826	0.2524	128
Time	0.3257	0.1402	0.1855	128
Protein	0.2642	0.1651	0.0991	128

Discrete-modality observed Spearman ranges

• clean: 0.694 to 0.969
• deranged: 0.852 to 0.898
• gap: 0.431 to 0.886
• residual: 0.391 to 0.894

Discrete-modality clean-minus-deranged pairs

Pair	Observed rho	Posterior 95% interval	Task-bootstrap 95% interval	Reference rho
Language–Genome	0.672	[0.552, 0.696]	[0.524, 0.787]	0.436
Language–Integer	0.431	[0.352, 0.486]	[0.255, 0.579]	0.340
Language–Protein	0.568	[0.382, 0.585]	[0.406, 0.703]	0.265
Genome–Integer	0.679	[0.587, 0.687]	[0.546, 0.778]	0.459
Genome–Protein	0.886	[0.697, 0.843]	[0.823, 0.925]	0.317
Integer–Protein	0.630	[0.466, 0.624]	[0.481, 0.743]	0.321

Discrete-modality residualized-clean pairs

Pair	Observed rho	Posterior 95% interval	Task-bootstrap 95% interval	Reference rho
Language–Genome	0.675	[0.559, 0.698]	[0.523, 0.784]	0.457
Language–Integer	0.391	[0.332, 0.450]	[0.216, 0.556]	0.352
Language–Protein	0.562	[0.391, 0.591]	[0.405, 0.700]	0.302
Genome–Integer	0.634	[0.568, 0.658]	[0.498, 0.753]	0.496
Genome–Protein	0.894	[0.715, 0.850]	[0.826, 0.930]	0.394
Integer–Protein	0.589	[0.457, 0.601]	[0.436, 0.715]	0.383