docs(diff-ux): add §3.13 — two blind independent redesigns, convergence, and adoptions

Two designers were given only the task list and research evidence (the
existing Fresh UI and this doc's workspace design withheld) and asked to
design the review TUI from scratch. Both independently converged with the
workspace on the same skeleton — comparison-as-identity, section tree,
selection-drives-preview, relabeling verbs, hash-keyed viewed marks,
resume-leads-with-delta, undo over confirmation, runnable help — which
de-risks the §3 core.

The section presents each alternative's distinctive ideas with one
representative frame and a verdict table, keeps the genuine differences
explicit (chips vs hint bar, cycling rail, in-workspace presets), and
ends with the adoption list: uppercase-escalation keymap rule, C conclude
verb, runnable-help-as-palette, persisted undo journal via git objects,
waiting-on-you session ordering, narrative default order for agent
scopes, live-update queueing, symbol-graph reject guard, last-review
default target, interdiff pseudo-commit, noise out of the progress
denominator, command log, and the sub-100-col zoom model.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01756Rh35n65zC4xHQcQAqBL

Author: claude

docs/internal/unified-diff-experience-design.md

@@ -616,6 +616,127 @@ column (→ notes rail + inline boxes), the two-line hint bar (→ scope bar +
 context line + chips), the separate picker dialog (→ scope-choosing state),
 and the separate History tab (→ history preset + commits lens).
 
+### 3.13 Independent redesigns: two blind alternatives, and what to take from them
+
+To pressure-test §3.1–3.12, two designers were given the same brief — the task
+list (§ Tasks) and the research evidence (§2) only, with the existing Fresh UI
+and this document's workspace design deliberately withheld — and asked to
+design a TUI review experience from a blank slate. Their full documents are
+preserved with the session research notes; this section presents each design's
+distinctive ideas and the resulting decisions.
+
+**The headline is the convergence.** Working blind, both designs and the
+workspace (§3.2–3.9) independently arrived at the same skeleton:
+
+- the comparison (`base ⟶ target`) as an always-visible, clickable, editable
+  control that *is* the review's identity — every source a preset of it;
+- a magit-style typed section tree with persistent folds and depth presets;
+- a navigator whose selection reactively drives a read-only diff projection
+  (`Enter` = portal to the live buffer; never a second writable copy);
+- verbs that relabel per source on unchanged keys (stage/discard ⇄
+  keep/reject), shown in-place so nothing is hidden;
+- viewed marks keyed by content hash, mark-and-advance, resume that leads
+  with "what changed since I last looked";
+- undo instead of confirmation prompts, with the real git command echoed;
+- runnable `?` help; noise quarantined by default; one-column marker
+  scrollbar as the review's map.
+
+Three independent derivations from the same evidence landing on the same core
+is the strongest de-risking signal this document has: **the §3 skeleton is not
+one designer's taste.** The interesting material is where they differ.
+
+#### Alternative A — "Ledger" (outline-first, conclude-verb design)
+
+One frame (agent session; note the narrative ordering and the inline guard):
+
+```
+◆ Review: session fix-flaky-tests   base session-start ▾ ⟶ head worktree (live ◉)
+   agent: working…  ·  kept 9 · rejected 2 · open 7  ·  viewed 11/18  ·  Flat ▾
+├──────────────────────────────┬────────────────────────────────────────────────────┤
+│ narrative order (agent's) ▾  │ ⌁ fn wait_for_port() · tests/util.rs               │
+│  ✓ tests/util.rs     kept    │   88│ 88▎  let deadline = Instant::now() + timeout;│
+│  ✗ tests/flaky.rs    rejected│   89│   ▁  «sleep(500)»                            │
+│ ▸· tests/net.rs     ↺ changed│     │ 89▎ «backoff.retry(|| probe(port))»          │
+│  · src/backoff.rs  ● new     │ ⚠ tests/net.rs:14 (kept) calls probe(port) —       │
+│  ▸ 12 more (unreviewed)      │   rejecting this hunk removes its definition.      │
+├──────────────────────────────┴────────────────────────────────────────────────────┤
+│ hunk: s keep · x reject · v viewed · c note · N next open · live: files update    │
+```
+
+Distinctive ideas, with verdicts:
+
+| Idea | Verdict for the workspace |
+|---|---|
+| **Uppercase escalates lowercase** (`v` viewed → `V` unmark; `c` comment → `C` *conclude*; one rule makes the map guessable) | **Adopt the rule** as a keymap design principle; `C` = conclude (commit / submit / finish) fills a gap — the workspace had completion *exits* but no single conclude verb |
+| **`?` help = runnable, filterable command list** (help and palette are one thing) | **Adopt** (both designs had it; §3.7 said "runnable" — make it the palette itself, scoped) |
+| **Undo journal persisted across restarts**, discarded hunks saved as patch files | **Adopt**; upgrade §3.1 principle 9 from "session undo" to journal-backed |
+| **Entry rows sorted "waiting on you" first** with per-session progress sparklines | **Adopt** in the scope presets |
+| **Narrative order (agent's edit sequence) as the default file order in agent scopes** | **Adopt as the default** there (deterministic — it's the edit log, not an AI summary; path order one toggle away) |
+| Comments have **no panel at review width** — inline pills only; a comments rail exists only ≥160 cols | Partial: keep the notes rail as a cycled projection (a severity-sorted index of a 50-file review is worth a rail), but adopt the **wide-screen anchored rail** for ≥160 cols |
+| `v` = viewed (because `Space` = extend selection) | Keep `Space` = advance (§3.6); selection stays on `v`/drag — the burn-down key deserves the biggest key |
+
+#### Alternative B — the "lens" design (take/untake/drop, live-queue design)
+
+One frame (the live-update queue and the symbol-graph guard — its two best
+ideas — visible together):
+
+```
+┌ Review ▸ session lexer-rewrite  start ⇄ now  ● agent working   ▰▰▰▱ 11/18 · [s 1/3]┐
+├──────────────────────────────┬─────────────────────────────────────────────────────┤
+│ ▾ To review (7)              │ src/parser/lex.rs › fn next_token()                 │
+│ ▸ ▾ src/parser/lex.rs   ↻2   │  150       ─   let c = self.peek();                │
+│     ▾ fn next_token()        │       150  +   let Some(c) = self.peek() else {    │
+│       hunk @201  NEW         │ ┌─────────────────────────────────────────────────┐│
+│ ▾ Rejected (2)               │ │ ⚠ rejecting this hunk conflicts with 2 KEPT     ││
+│  Sessions needing review:    │ │   hunks that call `peek()`'s new signature:     ││
+│  ▸1 lexer-rewrite ● 7 left   │ │   mod.rs @44 · ast.rs @210                      ││
+│   2 fix-ci        ✓ 3 left   │ │   x reject those too · Enter reject anyway      ││
+├──────────────────────────────┴─────────────────────────────────────────────────────┤
+│ s keep · x reject · u unkeep · N jump to 2 new changes · ]s next session · ? help  │
+```
+
+Distinctive ideas, with verdicts:
+
+| Idea | Verdict for the workspace |
+|---|---|
+| **Live-update queueing**: files you're inside queue agent edits behind a `↻N` badge (`N` applies and jumps); other files refresh freely — *content never moves under your cursor* | **Adopt verbatim** — the best answer found to watch-mode vs. rug-pull, stronger than plain watch |
+| **Symbol-graph context-blindness guard**: rejecting a hunk checks identifiers it removes against identifiers *kept* hunks reference; inline warning lists the linked hunks with one-key "reject those too" | **Adopt** — replaces §3.9's cruder file-default heuristic (keep file-level default too; they compose) |
+| **"Since my last review" as the *default* target when reopening a previously-reviewed lens** (full range one keypress away) | **Adopt** — §3.6's resume banner asked; B's default answers |
+| **`⇅ interdiff` pseudo-commit** in commits lens after a force-push (v3→v5 as just another section) | **Adopt** in the commits lens |
+| **Noise excluded from the progress denominator** (lockfiles/generated/fmt-only don't count against 19/34) | **Adopt** — progress must measure judgment, not scrolling |
+| **Undo ring backed by git objects** (`refs/…/undo/*`), destructive verbs recoverable across restarts | **Adopt** as the implementation of A's journal |
+| **`` ` `` VCS command log** toggle (every echoed command, reviewable) | **Adopt** — cheap, completes the transparency story |
+| **<100-col zoom model**: one pane at a time (Esc shows the tree full-screen, Enter dives back); burn-down loop works without ever visiting the navigator | **Adopt** — cleaner than shrinking panes side-by-side at 80 cols |
+| **`h`/`l` widen/narrow granularity** (file ⇄ hunk ⇄ line as a ladder you climb) | Consider (conflicts with pane-focus muscle memory; prototype both) |
+| Verb triad named **take/untake/drop** (`s`/`u`/`x`) relabeling per lens | Naming kept as-is (stage/keep are the users' words), but the *triad symmetry* (every take has an untake) is adopted as a rule |
+
+#### What stays deliberately different in the workspace design
+
+- **Chips over a global hint bar.** Both alternatives keep a persistent
+  bottom hint bar listing current verbs; the workspace puts the verb hints
+  *on the focused section* (chip row) with a leaner context line. Chips keep
+  eyes on the code and scale better to the web renderer; the hint bar
+  duplicates what chips already say. (If chips test poorly for
+  discoverability, B's hint bar is the fallback — it's the same data.)
+- **The rail cycles (files/commits/notes); the alternatives fix it to
+  files+commits.** The notes projection earns its slot in marathon reviews;
+  cycling keeps the one-rail budget.
+- **Scope presets open in-workspace** (§3.3's choosing state) rather than as
+  a modal overlay picker — one less floating surface, same content.
+
+#### Adoption summary (folded into the next revision of §3)
+
+From A: uppercase-escalation rule; `C` conclude verb; runnable-help-as-palette;
+persisted undo journal; waiting-on-you session ordering; narrative default
+order in agent scopes; ≥160-col anchored comments rail.
+From B: live-update queue (`↻N`/`N`); symbol-graph reject guard; last-review
+default target; `⇅` interdiff pseudo-commit; noise out of the denominator;
+git-object undo ring; `` ` `` command log; sub-100-col zoom model.
+Under prototype: `h`/`l` granularity ladder vs. pane focus; chips vs. hint bar
+(instrumented, not argued).
+
+---
+
 ## 4. Part II — talking to agents and outside systems (lower priority)
 
 Everything here is **orthogonal to Part I by construction**: it consumes the