Codes menu — focus on a code group

A proposal. Introduces a focus cursor for code groups and codes on the codebook lens, mirroring the signal-card focus already shipped on Analysis — which is what gives every Codes command a target and makes the menu live again.

12 Sep 2026 · proposal, not a decision · supersedes the 12 Sep retirement of four of six commands

1 — Focus is not selection

The retirement earlier today argued these commands had no target, citing the 29 Aug pin: “selection is single, it lives in the master list … no second place a thing can be ‘current’.” That reasoning was too strong. The pin governs selection — which codebook the detail pane renders. A focus cursor inside that pane is a different axis, and the codebase already models the two separately in two places.

Precedent 1 — FocusContext, the quote grid. Its own docstring lists them as separate concerns: “Focus (keyboard cursor): at most one quote focused at a time” and “Selection (multi-select): zero or more quotes selected for bulk actions”. The Quotes lens has had both since the React migration, and nobody calls the focused quote a second selection.

Precedent 2 — AnalysisSignalStore.focusedKey, the signal cards. A module-level store holding one nullable key, read by the card (bn-selected → blue wash), the sidebar and the inspector. 40 lines. This proposal is that store again, with two keys instead of one.

Doc obligation. The pin’s wording — “no second place a thing can be ‘current’” — is what invited the over-reading, twice now. If this proposal lands, the pin says selection is single and names focus as a separate axis, or a future session re-retires these commands on the same misreading.

2 — The focus ring

One click on a group card’s background focuses it. Same gesture as a signal card, same bn-selected wash, same clearing rules (click elsewhere, or Escape). Focus is a cursor, not a mode: every existing direct-manipulation affordance keeps working untouched.

Onboarding✕

first runsignup empty state

+ Add code

Trust & safety✕

privacyconsent data sharingaudit

+ Add code

Pricing✕

too expensivetiers

+ Add code

Group focused focusedGroupId = 2 — Rename Code Group, Delete Code Group and New Code are now live.

Focus on a code, and the rename field

Trust & safety✕

privacy consent ✕ data sharingaudit

+ Add code

Trust & safety✕

privacy consent data sharingaudit

+ Add code

Left: code focused — the chip carries the wash and reveals its ✕, exactly as hover does today. Right: rename in progress — the field is open on the group title and on the chip. Delete Code stays enabled here and does what the chip’s ✕ does; it must not be the ⌫ key (§5).

3 — The menu, in four contexts

Menus dim, never hide. Every row below is present in all four; only the enablement changes.

Floor page, group focusedThe everyday case. Everything the lens can do, the menu can do.

Floor page, nothing focusedCreation needs no target, so New Code Group survives. The four that name a thing dim.

Framework detail pageA framework’s structure is read-only, so every mutation dims. The last row swaps its verb to mirror the page button.

Library cards pageNo codebook is open, so the menu cannot name one. Install shows as the resting verb, dimmed.

Why the four rename/delete rows carry no icon. The icon audit already flags pencil ×4 and trash ×3 as the thinning candidates. Giving these rows the obvious glyphs takes the app to pencil ×6 and trash ×5 — the menu bar’s two most repeated symbols, on rows whose labels are already unambiguous. Tier 2 said the payoff of an icon falls as its repeat count rises; this is that rule applied before the fact rather than after.

4 — Enablement

Menu item	Twin on the page	Target	Enabled when
New Code Group	the New Group button	none — creation	floor page open
New Code	a group’s + Add code row	focused group	floor + group focused
Rename Code Group	click the group title	focused group	floor + group focused
Delete Code Group	the ✕ top-right of the card	focused group	floor + group focused
Rename Code	click the chip’s name	focused code	floor + code focused
Delete Code	hover the chip, click ✕	focused code	floor + code focused incl. while renaming
Browse Codebooks	the Browse Library button	none — navigation	always, on the lens
Install / Uninstall Codebook	the page’s own Install ⇄ Uninstall button	the open codebook	detail page open not floor, not Sentiment

The verb swaps, it does not gain a checkmark. Same idiom as Turn On / Turn Off Agent Access in the sidebar (§3.6a). And the exclusions are not new rules — canInstall on both the browse card and the detail page already excludes the floor and Sentiment (D20). The menu mirrors that predicate rather than restating it.

5 — Three things the proposal has to decide

Delete Code must not take ⌫, and probably takes no shortcut at all. A SwiftUI .keyboardShortcut installs an NSMenu key equivalent, which is matched before the responder chain — the same mechanism that forces Edit ▸ Undo to hide itself while editing. So a ⌫ on Delete Code would steal the keystroke from the rename field the proposal explicitly wants live. ⌘⌫ is already spent twice in Project. Recommendation: no shortcuts at all in v1. A shortcut is a speed affordance, so the test is demand for speed — not how important a command is, nor how advanced. Two things earn one: often, in the main workflow (edit, delete, bring to front), or spontaneously needed (Settings, Print — rare, but wanted now). Nothing here passes either; every row is the twin of a control already on screen, and nobody is in a hurry to open the codebook library. And the size of the set is itself the value — a small set stays learnable, so each addition taxes all of them. A draft gave Browse Codebooks ⇧⌘L, which was invented, not derived: unclaimed across every Swift file and the web layer, but free is not earned.
Focus on a framework’s group — allowed, but inert. A cursor is still useful for reading, and refusing focus there means two click behaviours for one control. Let focus land, let the mutations dim. That is the “menus dim, toolbars morph” rule applied to a lens.
Native has to be told. Dimming needs focusedGroupId, focusedTagId and which page is open on the Swift side — a new bridge payload alongside focusedQuoteId. Per the house rule that is a two-file change plus the fixture, and it is most of the work: the web half is one 40-line store and three handlers.

6 — What stays retired

Show/Hide Code Group — unchanged by this proposal. D7 puts the eye in TagSidebar/TagGroupCard on the Quotes lens and says hide “was never a third axis here”. That was never a selection argument, so focus does not revive it. G7/Q11 stay closed.
Merge Codes — focus gives one target; merge needs a source and a destination. Drag-one-chip-onto-another remains the only gesture that can say which two. Returns with multi-select, if ever.

Correction carried by this mockup. Four of the six commands retired earlier today — Rename/Delete Code Group, Rename/Delete Code — were retired on the reasoning that no target could exist. Focus is that target, and it was already shipped twice elsewhere in the app. The retirement of the other two stands on separate arguments.