improvise-lxg0/w1bu: evidence, and command.md describes the fixed state

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdfEGXA7aJ6ftMsCc9PJze

context/code/command.md

@@ -818,15 +818,20 @@ would therefore drop a frame the client keeps, and the host and the client
 would disagree. The first version of this fix made exactly that mistake.
 `colon_view_back_respects_whether_history_exists` caught it, because its
 fixture is such a frame. `reduce_view`'s `stepped` is unchanged, because the
-client holds no `Workbook` to observe against. That leaves the host and the
-client able to disagree wherever a client's stack holds a frame the document
-would refuse. An earlier version of this paragraph said such a frame could not
-exist, because `ForgetView` purges deleted views and a load arrives as a fresh
-`Snapshot`. **That is false.** `project.rs::fresh_view` starts from
-`ViewState::default()` and folds `SwitchView(active)`, so every connected or
-rebased client gets a phantom `{Default, Normal}` back-frame, and `Default`
-need not exist in the document. That is improvise-lxg0, and until it is fixed
-the divergence is live rather than theoretical.
+client holds no `Workbook` to observe against. It is safe only while no client
+stack holds a frame the document would refuse, which rests on three paths:
+`ForgetView` purges a deleted view from every mirror, a load arrives as a fresh
+`Snapshot`, and that snapshot's history starts **empty**. The third was false
+until improvise-lxg0. `project.rs::fresh_view` used to start from
+`ViewState::default()` and fold `SwitchView(active)`, which recorded a visit to
+the literal `Default`. So every connected or rebased client carried a phantom
+`{Default, Normal}` back-frame, whether or not the document had such a view.
+It now builds the slice pointed at the document's active view and folds
+nothing, as `App::with_storage` does. The guard is still one level away from
+the hazard. A new path that seeds or replaces a client's history without
+emptying it reopens the divergence, and nothing but
+`a_fresh_session_has_no_history_to_step_back_through` and its rebase twin
+would notice.
 
 ### `ReturnToNormalUnlessMoved` — observation, not prediction (improvise-toc)
 

context/test-evidence/command.md

@@ -1993,8 +1993,10 @@ property: it is still not a `SetCell` target, because `typed_cell_command`,
 `record_cell_at_cursor` and `render.rs::editable` all match `Record` alone.
 The test seats the cursor on the **second** row, so the first-row and
 first-item cases cannot coincide. It then enumerates every cell of the records
-grid and requires each to carry an id, which is the ticket's claim with no
-case picked by hand. Before the fix it failed at `a drawn coordinate projected
+grid and checks each one's kind against its column's `RecordCol`
+(`names_its_column`), which is the ticket's claim with no case picked by hand.
+The first version of that enumeration checked only that each id was `Some`,
+which was weaker than the test it replaced. See improvise-w1bu below. Before the fix it failed at `a drawn coordinate projected
 as no cell`. Rival injected afterwards: `RecordCol::Category(_) => return None`
 restored. This test failed at the `Some(CellId::Coordinate)` assertion, and
 `a_creating_column_is_named_rather_than_left_as_no_cell` (whose `None`-count
@@ -5313,3 +5315,37 @@ refusal is recorded and the mode is not `Quit`. `import-json` failed first,
 before `refuse_and_stop` was applied to its arms. For `load`, the injected fault
 is removing its abort (`refuse_and_stop` back to `app.refuse`), and the test
 then fails with "`load absent.improv . q` quit having loaded nothing".
+
+### a_fresh_session_has_no_history_to_step_back_through
+
+improvise-lxg0. `project.rs::fresh_view` built a new session's slice by folding
+`SwitchView(active)` into `ViewState::default()`. That default's `active_view`
+is the literal `Default`, so the fold pushed a `{Default, Normal}` back-frame
+for a visit that never happened. The test connects to two documents, and both
+open on `Second`. One **has** a `Default` view: there `<` would take the client
+to a view it never visited, which is improvise-lrrx's symptom seen from the
+client. One **has not**: there the frame names a dead view, and
+`reduce_view::stepped` would step onto it anyway. A fixture that opens on
+`Default` is right by accident, because the phantom frame then names where the
+client already is. Before the fix both cases failed at the back-stack
+assertion with `left: [ViewFrame { view_name: "Default", mode: Normal }]`.
+The rival is that unfixed fold. The second assertion folds `ViewBack` and
+checks that the client stays on `Second`, so a fix that emptied the stack
+while still recording a move elsewhere cannot pass.
+
+### a_rebased_session_has_no_history_to_step_back_through
+
+improvise-lxg0, through the other door. `Sessions::rebase` reseeds through
+`fresh_view` too. The session connects to a document opening on `Default`, then
+`install_workbook` swaps in each of the two documents above. Against the
+unfixed code it failed the same way.
+
+### The records-grid enumeration checks kinds, not presence (improvise-w1bu)
+
+The review of #196 injected `RecordCol::Category(c) if row == 0 => Some(c.clone())`
+into `cell_id`. That projects row 0's coordinate as an editable `Record` with
+`measure: Some("Region")`, and the whole workspace passed, because the
+enumeration asked only "is every cell something". It now asks "is every cell
+the right thing", column by column. Re-injected here, exactly one test in the
+workspace failed, this one, with `a records cell of the wrong kind: [(0, 0,
+Some(Record(RecordCell { .., measure: Some("Region") })))]`.