Sessions lens — native popover navigation

Rows are real Fishkeeping data — genuine names and durations, genuine multi-participant sessions (#6, #8, #10) where two or three people share one session and the second and third have no name, because the pipeline only named the first speaker in each. Codes stay globally sequential the way the pipeline assigns them, so a three-person session pushes every later code along — #11 is p16, not p11. Timestamps are substituted (Fishkeeping is a transcript-only import; see the note under the swatches). Twelve sessions is the default because that's the top of the common range — switch to 6 or 34 above to see the short case and the scrolling case.

• A compresses the shared prefix — p15, 16, 17 in one badge, names comma-joined in the title. Uniform 42pt row height, which is what a 20-row scrolling list wants. Watch it truncate at 308pt: “Session 16 · Nina Petrović, Participan…”. That truncation is real, not a mockup artefact — it is what 308pt buys you.
• B stacks — participant badges in the margin, one name per line, and no #N session badge. The p-codes are the nomenclature researchers already put on slide decks; #N is a CLI habit that reads as an escapee in a Mac list. Nothing truncates and every code is legible; rows go 42 → 90pt.
• Badges align to the names (chosen), not to the top of the margin. This is not tidiness: at top, the first badge sits level with the title, so p6 appears to label Session 6 — reintroducing exactly the session-number/participant-number conflation that dropping #N was meant to remove — and the last name is left with no badge. Aligning costs 16pt of row height. It needs no table-in-a-table: two columns sharing a row pitch is one CSS grid, or one NSGridView in AppKit. Flip the control to see the rejected shape.
• A's cost, visible once it's on screen: the wider compressed badge pushes that row's title right, so titles no longer align down the list — Session 6 and Session 10 start further in than their neighbours. It reads as ragged rather than broken, and it only happens on the rare row. Worth deciding deliberately: accept the ragged edge, or pin the badge column to a fixed width and let a multi-party badge overflow it.
• Badge is sized for pN and pNN — the 99.9% case. Three digits widens that one badge and shifts its title; it doesn't break, and it isn't a design target before beta.

Why a popover and not an NSMenu. The thing being replaced is already a dropdown, so this isn't a sidebar being demoted into a menu — it's a dropdown moving to the titlebar and gaining the density it always wanted. NSMenu can't do a two-line row with a muted subtitle, and the duration/date line is the whole reason this beats the old selector. The list is sized for what actually happens: 5–12 sessions very commonly, occasionally 30+. At 12 nothing scrolls; past that it does, and the “full fat” answer was always Sessions index rather than a longer menu.

Decided: the real AppKit source-list capsule — which decides the implementation, not just the styling. The capsule colour is internal to NSTableView, so having it means using the control: the popover stays a popover (SwiftUI .popover, same as the share menu) and its content becomes an NSTableView with style = .sourceList. The other three below are kept switchable as the record of what was weighed. Note the specimens are SwiftUI-style approximations — only the shipped control draws the true capsule.

The popover family has no precedent for a selected row — ExportPopoverRow has hover only (Color.primary.opacity(0.06)), because actions have no “current one”. This adds a state the family doesn't have, so it has to borrow from somewhere.

• Lozenge — the colour authored in Assets.xcassets for, in its own words, “any future SwiftUI selection surface”. That is this. Same 6pt rounded rect as the hover, so hover and selection are one shape at two weights.
• Accent only — matches the web .session-entry.active text treatment but drops its wash. Quietest; also the easiest to lose when the row is also hovered.
• Wash + check — the strict NSMenu idiom. It costs a third column, and the check pushes the badges out of alignment with every other row.
• AppKit capsule — reference only. The real source-list capsule colour is internal to NSTableView and matches no public UI-element colour (verified by sampling all of them). The chip here is an imitation; a SwiftUI row cannot read the real value.

Flip the palette above and watch these three converge and diverge. Under Edo the web nav-selection and the Lozenge are byte-identical (#e6ddc9) — they were authored as one decision. Under Default they are not even the same hue: the web sidebar washes rows neutral grey #efefef while the Lozenge is pale blue #eef4fc. AppKit's real capsule is a third value under both, and unreadable under both.

So “match the project sidebar” cannot be satisfied literally on any palette. The question the side-by-side actually answers is which mismatch reads as deliberate — a popover that agrees with the web list it replaces, or one that agrees with the native list above it.

On the date: rows show duration · formatFinderDate(...) — exact parity with the Sessions grid's Start column. Sessions ingested from media carry a real clock time (st_birthtime, s01_ingest.py:23). Sessions imported from transcript text with a bare YYYY-MM-DD header do not, and render 11 Feb 2026, 00:00 — the last row of each popover is pinned to that state so the failure mode is visible rather than described.