Neural Timing and Feedback Perturbation Experiments
This program asks when a registered perturbation is detected, when a measured muscle signal changes, when motion or load changes, and when a task outcome is corrected during golf-like motion. Those are four different observables. The protocol freezes phase, perturbation, sham, clock, calibration, response-window, hypothesis, uncertainty, and adverse-result rules before any model is fit.
This protocol supplies no unique neural-pathway attribution, no muscle-force identification, and no inference that an EMG onset caused a mechanical or task correction. Timing layers can overlap because sensory and motor processes use parallel pathways; their latencies must not be added into one serial round-trip budget. The current numerical records are manufactured synthetic-fixture evidence, with no coaching or clinical authority.
The checked-in state contains no participant data and cannot authorize participant collection. A future study requires externally governed ethics, privacy, consent, data-license, device-risk, calibration, stopping-rule, and independent-release records. The participant-held-out analysis plan, exclusion rules, power inputs, and multiplicity families must be registered before unblinding. Passing software or synthetic-fixture tests is not a human release decision.
Evidence Map
General Upper-Limb Evidence
| Primary Experiment | Bounded Support | Does Not Authorize |
|---|---|---|
| Kurtzer, Pruszynski, and Scott (2009) (Kurtzer et al. 2009) | Mechanical perturbations with separately analyzed early and long-latency upper-limb EMG responses during reaching | Golf transport, a universal latency band, or unique cortical-pathway attribution |
| Saunders and Knill (2003) (Saunders and Knill 2003) | Online hand-trajectory correction after virtual-hand position perturbations during reaching | A universal visual delay or golf transfer |
| Boyer et al. (2020) (Boyer et al. 2020) | Movement change after an unannounced perturbation of a movement-to-sound mapping | Auditory-pathway identity, correction latency, or golf transfer |
| Akbaş et al. (2024) (Akbaş et al. 2024) | Anticipatory kinematic and EMG changes under expected and partly predictable arm loads | A universal preparation strategy, full-swing transfer, or a causal neural mechanism |
Golf-Specific Evidence
Causer et al. (2017) used visual occlusion during novice golf putting at two distances and reported condition-dependent putting and quiet-eye outcomes (Causer et al. 2017). This is task-specific visual evidence. It does not establish mechanical or auditory feedback timing, elite-golfer behavior, full-swing transfer, or an individual neural pathway. The literature map therefore does not fill unsupported golf cells with reaching results.
Preregistered Questions and Hierarchy
Protocol affinedrift.neural-timing-feedback/v1 declares three families before analysis:
| Hierarchy | Perturbation and Phase | Frozen Contrast | Decision Boundary |
|---|---|---|---|
| Primary | Unexpected mechanical pulse at transition | Club-frame state divergence versus its paired sham | Holm-adjusted complete interval must exclude the minimum relevant effect |
| Secondary | Direction-unknown visual cursor jump at transition | Target-frame correction versus its paired sham | Holm-adjusted complete interval must exclude the minimum relevant effect |
| Exploratory | Predictable auditory mapping shift before impact | Club-frame speed change versus its paired sham | Report the complete interval without confirmatory promotion |
Each confirmatory family uses family alpha 0.05, target power 0.80, a frozen minimum relevant effect in its declared units, and Holm step-down multiplicity control. Participant count is unavailable until an external pilot variance and device reliability record are entered into the registered simulation plan. Recruitment cannot begin with an invented variance or an after-the-fact effect size. The final count, attrition allowance, exclusions, and stopping rule must be frozen before unblinding.
Perturbation, Sham, and Blinding Contract
| Modality | Active Component | Matched Sham | Physical-Onset Witness |
|---|---|---|---|
| Mechanical | Club-frame transverse load pulse | Same trigger, device sound, and duration without commanded load | Force/load onset plus command loopback |
| Visual | Target-frame cursor displacement | Same refresh and visibility without displacement | Photodiode on the rendered transition |
| Auditory | Movement-to-sound mapping-slope change | Same level, duration, and carrier without mapping change | Microphone loopback |
Condition codes are generated outside the analysis module. Participants, operators where practicable, onset reviewers, and primary analysts remain blinded to the active/sham mapping until frozen exclusions, calibration checks, and primary results exist. A sham is not assumed inert: expectancy, device sound, display refresh, and carrier sound remain measured covariates.
All active magnitudes, directions, durations, onset events, calibration revisions, and stopping rules are frozen. Saturation, trigger disagreement, limit crossing, unexpected contact, discomfort, or loss of retention stops the trial. The manufactured magnitudes are normalized software fixtures, not safe human operating limits.
Four Observed Timing Layers
| Layer | Operational Observation | Prohibited Shortcut |
|---|---|---|
| Perturbation Detection | First persistent physical-onset witness on the shared clock | Treating the software command time as delivered stimulus time |
| Muscle Response | Preregistered EMG-envelope change relative to its baseline and sham | Naming a unique reflex, voluntary source, intent, force, or stiffness |
| Mechanical Effect | Kinematic or kinetic divergence in a declared frame and uncertainty interval | Assuming the effect was generated by the measured muscle response |
| Task Correction | Target- or outcome-frame change versus sham | Calling any movement difference successful correction or learning |
These layers are analyzed on one event clock but are not required to form a serial chain. A direct load can create a Mechanical Effect before a measured Muscle Response. A detected EMG onset can occur without a task-relevant change. A task change can also reflect passive mechanics, pre-existing control, expectation, or an unmeasured pathway.
Synchronization, Calibration, and Onset Detection
The executable contract requires a hardware-shared clock for command, photodiode, microphone, motion, load, EMG, and task channels. Every channel declares units, frame, sample rate, and calibration revision. The manufactured clock declares 0.02 ms timestamp resolution, at most 0.10 ms channel skew, and 0.50 ms onset uncertainty. These are fixture gates, not achieved device performance.
Physical device qualification must measure command-to-load latency and bandwidth, photodiode transition latency, audio loopback latency, force and motion calibration, EMG delay and filtering, clock drift, dropped samples, and trigger jitter across the complete operating envelope. The onset algorithm uses a frozen threshold and persistence count. Alternative thresholds, filters, window edges, and clock corrections enter the complete sensitivity interval.
Phase, Expectation, Learning, and History
The manufactured phases are transition (top-of-motion to mid-downswing) and pre-impact (mid-downswing to nominal-impact). A future study must replace these labels with reproducible event definitions and quantify event uncertainty. Phase is not inferred from wall-clock percentage alone.
The analysis model includes phase, perturbation magnitude and direction, expectation condition, trial index, block, preceding perturbation, time since the preceding perturbation, and cumulative exposure. Unexpected, direction-unknown, and fully predictable conditions are not pooled. Acute within-trial response, anticipatory preparation, across-trial adaptation, retention, and transfer are separate estimands. Order is counterbalanced, sham frequency is fixed, and washout sensitivity is reported; a late-block change cannot be relabeled as immediate feedback.
Deterministic Synthetic Fixtures
The code fixture supplies one manufactured observation for each modality and layer:
| Modality | Detection | Muscle Response | Mechanical Effect | Task Correction |
|---|---|---|---|---|
| Mechanical | 2 ms | 52 ms | 16 ms | 138 ms |
| Visual | 8 ms | 132 ms | 148 ms | 176 ms |
| Auditory | 5 ms | 104 ms | 120 ms | 168 ms |
The mechanical fixture deliberately places the mechanical effect before the EMG response. It is a regression against the false serial-latency story, not a participant finding. A separate onset trace has a persistent threshold crossing at exactly 12 ms. Holm decisions preserve original hypothesis order. The interval ledger retains one supported, one negative, one null, and one unavailable record, each with an exact evidence origin, record ID, revision, and synthetic marker.
Outcomes, Uncertainty, and Falsifiers
Primary outputs are onset latency, response amplitude, club- or hand-frame mechanical divergence, and target-frame task correction, each versus its paired sham. Kinematic, kinetic, EMG, and outcome variables retain their own units and frames. An association between layers is secondary; mediation, pathway, and causal-source claims are unavailable without an independently identified intervention and model.
Complete uncertainty spans calibration, clock skew, onset threshold, persistence, filter, baseline, response-window edge, event timing, perturbation magnitude and direction, expectation, history, sham activity, missingness, model family, and multiplicity. The primary claim is falsified when its complete interval lies inside the preregistered equivalence region. An interval that straddles the minimum relevant effect is null/inconclusive, not supportive.
Promotion and Generalization Limits
The current result tier is analytic and deterministic synthetic feasibility. Promotion to measured human evidence requires the external records in the human boundary, qualified hardware, frozen recruitment and power, independent release, and held-out analysis. Even then, conclusions remain limited to the tested modality, phase, magnitude, direction, expectation, history, task, population, and device envelope. Reaching and novice-putting evidence cannot be generalized to elite full swings or to a universal theory of feedback control.