docs(fork): 1.3 provenance exit check PASS; 2.4 review-size counts corrected to measured

The 1.3 grep found one nonconforming phrase — the plan's own closing line
said 'and credit'; corrected to the Part-1.2 framing. The draft's review-size
claim (~873 lines) was stale and understated the branch: measured
~1,432 source insertions across 5 files plus ~2,202 test lines
(git diff --shortstat upstream-mirror...fix/dom-oom-virtualization);
odysseus-dev#4661's vendored trim is 142 lines. Matrix row and prose updated.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: James

docs/fork/plans/dom-oom-virtualization-upstream-plan.md

@@ -35,6 +35,10 @@ message window and scroll fixes"): authored 2026-06-11 20:47 UTC. #4661 opened
 1.3 **Exit check:** before filing, grep the draft and all fork docs for `4661`; every
 reference must match 1.2 verbatim in substance. Nothing may claim adaptation from #4661 and
 nothing may argue about influence.
+**Run 2026-07-18: PASS** (draft, decision record, pr-status, eviction draft all conform;
+the one nonconforming phrase found — this plan's own closing line said "and credit" — was
+corrected to the 1.2 framing). The streaming-throttle branch's genuine adaptation credits
+are a different branch and are out of scope. Re-run once more at file time.
 
 1.4 **Exit check:** a reviewer comparing the two diffs finds every claim verifiable and
 nothing argued about provenance beyond the 1.2 facts.
@@ -88,6 +92,12 @@ and verified.
 2.4 **Honest tradeoff, stated up front.** Ours is ~873 lines vs #4661's ~145. Cite the
 maintainability cost honestly; do not bury it. The pitch is "more complete and measured,"
 not "theirs is bad."
+**Verified 2026-07-18 — the old counts were stale and understated our size.** Measured:
+#4661's trim is 142 lines (the vendored file); the branch's source diff is ~1,432
+insertions across 5 files, plus ~2,202 test lines
+(`git diff --shortstat upstream-mirror...fix/dom-oom-virtualization`). Draft matrix row
+and prose corrected to the measured numbers. Re-measure after the Part-5 rebuild (the
+fold will grow the diff further).
 
 2.5 **Exit check:** every row of the matrix has a number and a verified citation, and the
 tradeoff is stated.
@@ -283,4 +293,4 @@ first):
   claims, no fork-internal leaks).
 
 Until then the branch stays unfiled. #4661 being open is not a blocker; it is the reference
-we measure against and credit.
+we measure against, acknowledged as parallel work per Part 1.2.

docs/fork/upstream/pr-drafts/fix-dom-oom-virtualization.md

@@ -110,7 +110,7 @@ commit `27f35e1c` as `tests/bench/vendor/trimChatHistory_4661.js`, provenance-gu
 | Reaching old history | scroll (all live) | 97 "show older" clicks, full restore, scroll position lost | scroll-up pages in place, window stays bounded |
 | Append layout cost, 25-msg stream (ms) | 101.3 | 4.7 | 3.2 |
 | Scroll smoothness (mean frame ms) | 16.7 | 16.7 | 16.7 |
-| Review size (lines) | — | ~145 | ~873 |
+| Review size (source lines changed) | — | ~142 | ~1,430 (5 files; plus ~2,200 test lines) |
 
 The two rows that decide it: #4661's cap genuinely bounds the steady state, but the moment
 a user reads old history the bound is gone — the DOM and USS return to the unpatched
@@ -134,7 +134,7 @@ relationship is precise, and stated honestly:
   releases hljs-defer observer references via `hljsDeferForgetNode`. Those uncleaned
   observers and renderers were confirmed leak sources in the fork's OOM investigation
   (`docs/fork/memory-explosion-research.md`).
-- **The windowing is more complete.** #4661 is a focused top-only trim (~145 lines): a
+- **The windowing is more complete.** #4661 is a focused top-only trim (~142 lines): a
   150-node cap, removal from the top, and a "load older" bar backed by server pagination.
   This adds bidirectional windowing so a user can scroll back up and reload evicted
   messages in place.