A fixed test did not confirm smaller high-load response steps during delay in 12 SEEG patients and 16 fMRI adults

We tested whether measured working-memory response steps are smaller from medium to high load than from low to medium load. Separate public studies supplied SEEG from 12 epilepsy patients and fMRI from 16 adults. The electrical loads were 4, 6 and 8 letters; MRI loads were 3, 5 and 7 items. One data block set feature scales. Two other blocks supplied response-step products. Spatial means were removed. Sessions and domains were averaged within people. A local code lock preceded reserved-person scores for this new endpoint. Mean compression was +0.00481 in SEEG and +0.04423 in fMRI. The respective 95% person Student limits were [-0.02524,+0.03486] and [-0.02978,+0.11824]. The fixed shared rule failed, with gate p=0.73123. All fourteen fixed controls also had compression intervals through zero. Added history and repeat-variation checks remain descriptive. An independent calculation matched all 67 primary development and test records. A second decoder matched every MRI source row. Two fresh raw SEEG rebuilds matched stored features. A clean replay matched all 107 numerical result files. Model checks show that unequal noise can bias ordinary step comparisons and that shared block errors can also defeat the separate-block method. This is a checked conditional assay and a complete negative benchmark. It does not establish equal steps, absent compression, a memory capacity or a common neural cause. Code, permitted compact inputs, source versions and terms, plans, figures and repeat instructions are supplied. Exact MRI rows remain external.

working memory · SEEG · fMRI · response geometry · load compression · negative result · noise bias · reproducible methods