Capstone — Recording report

Brief

For one raw file from each spine dataset (ds-eegbci, ds-lemon, ds-erpcore), produce a one-page recording report: montage, sampling, duration, event table, an artifact timeline by type, a first-look PSD, and a go/no-go recommendation with reasons. This is the Level 0 skill made concrete: you look at a file you have never seen and say, in writing, whether it is fit for analysis and why.

Use first_look_report() from notebooks/_shared/helpers.py (written in L0.6) as the skeleton; the artifact timeline is the L0.5 atlas applied to a whole file rather than 5-s items.

Inputs

• One raw file per spine dataset. helpers.load_spine fetches only that file; it excludes the documented defective ds-eegbci subjects (S088, S089, S092, S100) by default and says so.
• Prior lessons L0.1–L0.6, in particular the first-look checklist (L0.6) and the artifact atlas (L0.5).
• ds-erpcore access is TODO(confirm) (§13 item 22); if it cannot be fetched, substitute a ds-brain-invaders file and say so in the report.

Subset rule and runtime

Capstone notebooks run on a documented subset within the §11 limit of ten minutes on Colab’s free tier, and scale to the full cohort locally:

• The documented subset for C0 is the three named files (one per spine dataset; the ds-brain-invaders substitute if ERP CORE is unavailable). The notebook states the subject and run of each file in its first cells.
• A FULL_COHORT switch at the top of the notebook (default False) extends the run to every subject the notebook can enumerate for each dataset, for local execution only; the report function is the same, so the output is one report per file.
• Per-subject products that are expensive to compute (for example the artifact timelines for a whole cohort) may be precomputed by data/scripts/ and shipped as assets rather than computed in the notebook.
• The notebook downloads only the files it needs, uses _shared/helpers.py, contains no absolute paths, and prints in its final cell the numbers the rubric asks for (event counts, artifact counts by type, alpha peak frequency, line-noise frequency).

Deliverables

1. Notebook nb-c0-recording-report that runs top to bottom on the documented subset.
2. Rendered HTML of each report.
3. Each report contains: montage and reference; sampling rate and hardware filters as recorded in the header and as verified from the data; duration; event table (counts per event type, first and last onset, any timestamp anomaly); artifact timeline with type, channels and timestamps; first-look PSD with the alpha peak and any line-noise line identified in words; a go/no-go recommendation with reasons.

Rubric

• All metadata correct (sampling rate, channel count, reference, duration, event counts), with header values verified against the data.
• At least 5 artifact types located with timestamps.
• PSD interpreted in words (alpha peak, line noise, hardware fingerprint).
• Go/no-go recommendation is justified from the evidence in the report, not from a rule of thumb.

Example report structure

1. File identity (dataset, subject, run, format, license, checksum).
2. Acquisition (montage, reference, sampling rate, hardware filters, duration) — header values and verified values side by side.
3. Events (table; anomalies such as overlapping timestamps noted).
4. Artifact timeline (type · channels · start · end · confidence).
5. First-look PSD (figure with units; alpha peak frequency; mains line and harmonics; hardware roll-off).
6. Recommendation (go / go with caveats / no-go) and the reasons.

Estimated time: TODO(confirm) — the minutes value in the frontmatter is a site-design estimate, not from the spec.