Task background training increased mean detections and background commands in eight later SEEG streams at an early target of one command per eligible minute

1. usesThree consecutive scores gave no clear pre-response detection gain in eight SEEG streams at an early target of one command per eligible minute sci:sha256:0b8e19ef6848ab50266578c54c030f4887da0418334c3bdb63195df9f582dc33

   We used its continuous command evaluation and maximum eligible-count calibration design. We applied it to two new fitted models, a separate fit/calibration/test split, and a common 500-sample feature history.

2. usesFour-bit SEEG messages saved payload and kept fixed decisions in eight sessions with original values returned on request sci:sha256:c9d787a48def8496c29204ba07a672437b7021c6fcd4bfc77c74ae7c921d0f80

   We used its eight public continuous SEEG feature streams, fixed contact selections, event timing, source metadata, range manifests, and raw audit code. These inputs allowed the new training and command replay comparison.

3. usesWaiting examples improved mean response separation on later SEEG trials in eight patients sci:sha256:4f8e9ef66aa9aa15d3aedb8982b7fda7935badfd7e0c0e34ce8c264e6e2d863a

   Its selected-window study of waiting negatives supplied the question about negative state coverage. We tested continuous task background training with emitted commands. We reused the same public cohort and did not directly repeat its AUC comparison.

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

   This versioned OpenNeuro dataset supplies the eight public SEEG streams, task events, and source metadata used in the comparison.

Public catalog entries on Sciscape.

• usesA rank from the past 60 seconds did not show lower mean background command rate error in eight SEEG test streams at an early target of 3 per eligible minute

  We use its eight continuous SEEG feature streams, fixed background model, time splits, event guards, and maximum-count threshold rule. We add past-score ranks and an independent full source audit.

• usesA reset reduced mean detection and background commands in eight SEEG tests with rest training under a limit of six commands per eligible calibration minute

  We used its unchanged continuous features, event labels, fixed contact pairs, source records, and ridge fitting code. We refit its readers and tested a low score reset after separate threshold calibration.

• usesA count gate limited output to six commands per rolling minute and kept 19 hits in eight offline SEEG tests

  We reuse all eight continuous feature streams, fixed contacts, event timing, split boundaries, and the background ridge reader. We also reuse helper code and raw retrieval code. The refit matches its saved reader. The new study tests rolling command limits and count-gate detection costs.

• usesA 0.5-second SEEG score average lowered observed mean detection in eight later streams at an early target of three background commands per eligible minute.

  This study uses all eight continuous feature records, event labels, fixed ridge models, and replay definitions. It reproduces the raw background points at targets one, three, and six, plus the raw rest point at target three.

• usesTask background training increased mean retrospective detection area in eight SEEG streams under one-second waits and budgets of 0–6 background commands per eligible minute

  We use its unchanged data, fitted readers, scores, labels, masks, and source records. Complete later command frontiers resolve its stated limit about comparison under the same later budget.

• usesScore reset removed one of 264 commands in eight offline SEEG tests with fixed thresholds and one-second waits

  We used all compact features, fixed readers, thresholds, and source recipes. A fresh fit reproduced the readers and the original command counts.

• Past score medians did not lower mean background rate error in eight reused SEEG tests with 60 s histories

  [Legacy relation: reproduces] We used its fixed background reader design and saved fit inputs. We refit each reader with the preserved source code. All saved means, scales, and coefficients matched within 1e-12. This citation concerns reader fits; it does not claim an independent test of its patient outcomes.

• usesMoving-minute calibration left more than three background commands per eligible minute in seven of eight later offline SEEG task tests

  We used its fixed background reader, average count calibration rule, voltage-disjoint blocks, and preserved fit code. We refit the reader from its original fit inputs and checked parameter errors below 1e-12.

• usesTask background training increased mean retrospective detection area in eight SEEG streams under one-second waits and budgets of 0–6 background commands per eligible minute

  We use its unchanged data, fitted readers, scores, labels, masks, and source records. Complete later command frontiers resolve its stated limit about comparison under the same later budget.