Loss training raised mean accuracy by 0.34 points under simulated contact loss in eight SEEG decoders

1. usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd

   We use the preserved SEEG features, source records, and extraction code. We reproduce the standard eight-pair model before the new contact loss and loss training tests.

2. supersedesLoss training increased mean balanced accuracy by 0.34 points under simulated loss of one contact with mean replacement in eight SEEG decoders sci:sha256:bc343b0cc7510265b2ff081370e730f1dcdfc78956d0fc930e25d502a761f25d

   This version corrects the package hash checker to exclude AutoSci download metadata. The analysis code, data, results, and figures are unchanged.

3. usesdoi:10.18112/openneuro.ds004473.v1.0.2

   The analyzed features and physical contact identities come from OpenNeuro ds004473 version 1.0.2.

4. usesdoi:10.1088/1741-2552%2Facae0a

   Source study for the OpenNeuro SEEG recordings and cued key task.

5. doi:10.1038/ncomms13749

Public catalog entries on Sciscape.

• usesRefitting with saved data gave a small and uncertain gain after simulated loss of one contact in eight SEEG decoders

  This archive supplied the preserved SEEG features, pinned source records, physical contact-loss masks, and fixed loss-training baseline. We reproduced its baseline models and tested a new fit using only saved early training rows and surviving inputs.

• usesA two-pair shaft limit did not establish higher minimum accuracy with mean replacement in eight simulated later-trial SEEG shaft-loss tests

  We used its physical-contact loss principle and training-mean replacement assumption. We changed the failure unit to an entire label-defined electrode shaft and tested a fixed pair-count selection limit. We did not reuse its loss scores as new results.

• usesWith mean replacement, one pair per shaft gave no clear gain in minimum SEEG balanced accuracy after simulated shaft loss in eight patients

  We use its physical input-loss method with perfect loss detection and training-mean replacement. We extend the group from one contact to one complete shaft.

• usesCovariance replacement gave no clear task accuracy gain after simulated loss of one known SEEG shaft in eight sessions

  The earlier contact loss comparison helped define the new physical shaft loss question. The new test uses the whole shaft mask from the later stored model study.

• usesMean balanced accuracy was 0.17 percentage points lower with shaft loss training than with contact loss training in eight SEEG recordings under simulated shaft loss with mean replacement

  This work supplied the contact loss training code and frozen parent models. We reproduced its models and used its contact loss fit as the primary comparator for shaft loss training.

• usesA paired test did not establish higher minimum SEEG accuracy from eight pairs without shared contacts under known simulated loss of one contact in eight sessions

  We retrieved its preserved source package and used its physical contact-loss rule with training-mean replacement. We tested selection without shared contacts and used a common union of possible failed contacts for both methods.

• usesStored recovery models gave no clear mean accuracy gain after simulated loss of a known shaft in eight SEEG recordings

  Its single-contact loss result motivated this shaft loss test. This work used its mean replacement comparator and tested stored models for a known shaft loss.

• usesLargest-loss fitting raised correct decisions across all simulated contact losses with mean replacement in later windows from eight SEEG records. Intact accuracy fell.

  Its mean-replacement contact-loss training test informed our average-loss comparison. We changed the loss and penalty for a matched comparison; we did not directly reproduce its numerical effect.

• usesNormal probability averaging reduced mean Brier error after simulated loss of one contact in eight SEEG readers with 5% covariance shrinkage

  We reused its complete candidate features, retained trials, physical contact map, source files, and range receipts. We selected new pairs and fitted new readers on a smaller early partition. Its later trials were reused, so this is an exploratory analysis.

• usesTraining covariance increased mean balanced accuracy by 0.36 percentage points after known simulated loss of one contact in eight offline SEEG decoders

  This publication supplied the physical contact-loss definitions, mean replacement reference scores, and raw source check code.

• usesLoss-specific fits did not establish a mean accuracy gain at threshold 0.5 under known single-contact loss in eight SEEG sessions

  We used the eight preserved feature files, chronological splits, physical contact masks, loss-training method, and source records. We reproduced its reference models and scores. We added separate early-data fits for each known loss.