docs: clear three stale references revision 3 left behind

Revision 3 corrected the cache model in section 3.2 and the clipping
rule in 3.4, then failed to propagate either. Internal contradictions in
a framing document are worse than the original errors: a later reader
resolves them by picking whichever statement they read first.

  - Q#D2-4 still summarised the design as "per-variant cache keys",
    contradicting 3.2's corrected one-per-peer model. Now names the
    `peer_knows_minibuffer_rows` producer gate and the single per-peer
    cache, with a parenthetical recording that the sentence was stale
    rather than silently rewriting history.
  - Section 6 still required the cross-version cache test that
    revision 3 removed as unfalsifiable. Replaced with what IS
    assertable --- a v22 and a v23 peer attached simultaneously each
    receive their own variant and only their own --- and the bullet
    says why the other test is absent, so it does not get "restored"
    later by someone who notices the gap.
  - The active-work lane still called the framing revision 2.

Swept for the same class of residue from the other two corrections
("the name must survive", "both frontends render"); the remaining hits
are the notes ABOUT those corrections, which are deliberate.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016bqGA6s9tTUFzYpbeW3tai

docs/active-work.md

@@ -275,7 +275,7 @@ from #171 and #215.
 tip** — the ref, not a SHA. Recover with
 `git fetch githubsucks && git checkout discovery-stage2`.
 
-- **Framing `docs/discovery-stage2-framing.md`, revision 2**, in
+- **Framing `docs/discovery-stage2-framing.md`, revision 3**, in
   review. Scope: `COHERENCE.md` §5's "M-x rows are still bare names".
   Descriptions already exist on `Command` and are already rendered by
   `help.list-commands`; they are missing at the one moment they would

docs/discovery-stage2-framing.md

@@ -248,14 +248,19 @@ later.
 has yet asked to disable is how a registry becomes noise. If somebody
 wants it off, that is use evidence and a later one-line addition.
 
-### Q#D2-4 — older frontends — **RESOLVED in rev 2, in §3.1–3.2**
+### Q#D2-4 — older frontends — **RESOLVED, in §3.1–3.2**
 
 No longer open, and the revision-1 answer was wrong. "Gate the richer
 form at `>= 23`" would have **removed the minibuffer entirely** from
 every v12–v22 peer, because there would have been only one variant to
 gate. Compatibility requires the legacy shape to still exist and still
-be sent — hence the additive `MinibufferPromptRows` variant, per-peer
-selection, per-variant cache keys, and matched open/close families.
+be sent — hence the additive `MinibufferPromptRows` variant, a
+per-peer `peer_knows_minibuffer_rows` producer gate, **one per-peer
+minibuffer cache**, and matched open/close families.
+
+*(Revision 2 said "per-variant cache keys" here. §3.2 corrected that in
+revision 3 — the render state is per peer with its version baked in, so
+a cache cannot span two versions — and this sentence was left stale.)*
 
 The `CompletionPopup` gate I proposed copying (`daemon-gated >= 15`)
 **is** the right precedent for *how to select per peer*; it is not a
@@ -300,9 +305,12 @@ the temptation arrives with the feature.
   encoding changes. Two fixtures: an open prompt with candidates and a
   selection, and a cleared band — the two shapes the existing semantic
   test already covers, so the corpus is not a new judgement call.
-- **The cache key is per variant**: a session that opens for a v23 peer
-  and a later one for a v22 peer are not suppressed as duplicates of
-  each other (§3.2).
+- **No cross-version cache test.** Revision 2 required one; it asserts
+  a condition this architecture makes impossible (§3.2), and a test
+  that cannot fail passes forever while teaching the next reader that
+  the hazard is real. What *is* asserted is the producer gate: a v22
+  peer and a v23 peer attached simultaneously each receive their own
+  variant and only their own.
 - **Close matches open**: a `MinibufferPromptRows` session is closed by
   its own family, witnessed by the popup actually clearing.
 - **The TUI renders `name — description` for the selected candidate**