---

name: preprocessing-epochs-evoked description: "Guide MNE-Python event extraction, sensor-space preprocessing, Epochs, Evoked, artifact correction, covariance, and rank workflows." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

MNE-Python preprocessing, Epochs, and Evoked router

Use this sub-skill when a task needs to turn continuous MNE data into events, epochs, evoked responses, or cleaned sensor-space data before downstream analysis.

Load order

1. For signatures, object contracts, and parameter caveats, read references/api-reference.md.
2. For event, epoch, evoked, covariance/rank, fNIRS, and eye-tracking recipes, read references/workflows.md.
3. For ICA, SSP, EOG/ECG, muscle, reference, interpolation, fNIRS, and blink correction decisions, read references/artifact-correction.md.
4. If the user reports an error, warning, dropped trials, missing events, bad shapes, filter issues, or fitting failures, read references/troubleshooting.md.
5. For a deterministic local smoke check, run or adapt scripts/preprocessing_smoke.py with a Python environment that has MNE-Python and NumPy installed.

Own these requests

• Event arrays from stim channels, annotations, or fixed windows: find_events, events_from_annotations, make_fixed_length_events, pick_events, merge_events, event dictionaries, duplicate-event policy, and BAD* annotation behavior.
• Epoch construction and arrays: Epochs, EpochsArray, metadata, selection, baseline, reject, flat, proj, detrend, decim, reject_by_annotation, on_missing, on_outside, and event_repeated.
• Evoked construction and arrays: epochs.average(), Evoked, EvokedArray, read_evokeds, write_evokeds, combine_evoked, grand_average, cropping, filtering, resampling, baseline, channel picking, and save/load round trips.
• Sensor-space cleaning before epochs or averaging: filtering, notch filtering, resampling, bad-channel marking, interpolation, EEG reference changes, projectors, ICA, SSP, EOG/ECG/muscle artifact helpers, fNIRS preprocessing, and eye-tracking blink handling.
• Noise covariance and rank estimates that depend on preprocessing choices: compute_covariance and compute_rank.

Route elsewhere

• File loading, vendor-specific readers, RawArray basics, Info creation, channel metadata ingestion, raw concatenation, and file export belong to the sibling io-raw-data sub-skill.
• Plot layout, interactive browsers, reports, topomaps, mne.Report, and headless rendering belong to visualization-reporting.
• Time-frequency, spectra beyond filtering, statistics, decoding, CSP, simulation, and source-level statistics belong to timefreq-stats-decoding-simulation.
• Source spaces, BEM/forward/inverse operators, beamformers, and source estimates belong to source-modeling-inverse.

Operating rules

• Prefer copy-before-mutate for user data. Many MNE methods mutate in place and return self (filter, resample, set_eeg_reference, interpolate_bads, apply_proj, ICA.apply).
• Preserve event timing deliberately. If downsampling after extracting events, use Raw.resample(..., events=events) or recompute events from the resampled raw object; do not silently reuse stale event sample numbers.
• Keep artifact-correction plans evidence-based: mark bad spans/channels first, document whether ICA/SSP/filtering/referencing changes rank or baseline, and inspect drop_log, info['bads'], projector state, epochs.selection, and evoked.nave after each stage.
• Never require the original repository checkout at runtime. This skill is self-contained; source paths mentioned in reference files are provenance only.