[rossant]

Curate long recordings from sampled time windows, then apply those decisions to every spike in a new full sorting. The original dataset stays unchanged.

phy prepare-large SOURCE/params.py BOUNDED
phy template-gui BOUNDED/params.py
phy apply-curation SOURCE/params.py BOUNDED CURATED

• Merges, quality labels, notes, amplitude-threshold splits, undo/redo and save/reopen.
• Unit Stability overlays units, with separate rate, amplitude and depth toggles.
• Fast full-recording split summaries use an optional amplitude index. The GUI prepares or upgrades it in the background with progress, per-stage ETA and Cancel; no reopening required.
• SLAY decisions and metadata survive curation rounds; new units clear imported metrics and quality labels until reviewed.

Performance: Synthetic 1,000h / 4.18 billion spikes: first summary queries took 55 ms for a typical unit, 115 ms for the largest, 244 ms after merging the two largest. Adding the index took 4.4 min / 20.3 GB. Timings exclude GUI delivery and use uncontrolled OS caches.

Validation: Full local suite passed, plus final index regressions. Real 100h and synthetic 1,000h results match validated full-source calculations and independent replay. Current CI is running.

Still draft: needs Carl's native long-recording trial and verified timing/exposure. Arbitrary lasso/feature splits remain unsupported.

Guide and GUI screenshots · Benchmarks and limits · Separate phylib fix.