diff --git a/docs/plans/2026-09-26_lead-decisions.md b/docs/plans/2026-09-26_lead-decisions.md
index c5b161487..cc17c15f2 100644
--- a/docs/plans/2026-09-26_lead-decisions.md
+++ b/docs/plans/2026-09-26_lead-decisions.md
@@ -2785,3 +2785,34 @@ which stay with him. Each item names who decided it and what happened.
     - Until the Console runs from a pinned copy, a landing that touches
       `packages/webui/src/public/` waits for Mos's go, not just a
       notice. The rest of decision 109 stands.
+124. **Researcher builds rows 97 and 98 on their current model
+    (2026-10-12).** Mos ruled this (class=decision message to Sage) so the wait
+    for the proxy restart isn't wasted. Decision 122 sends rows 71, 80,
+    84, 88 and 96 to Haiku. It names neither 97 nor 98. Sage's queue
+    notes had put both on Haiku, and that went beyond decision 122.
+    - Researcher builds row 97 and then row 98 on Sonnet 5.5, the model
+      their thread runs now. Row 97's reviewers stay Darkwing and
+      Filbert, and row 98's stays Filbert, so each gets a GPT review.
+    - Rows 71, 80, 84 and 88 stay parked for Haiku. `queue next
+      researcher` names row 71 first, so Researcher takes rows by number,
+      not from that verb, until the switch.
+    - The Haiku thread starts only after Mos verifies Haiku 5.5 through
+      the proxy (Vikunja #242). Its first prompt is a holding prompt with
+      no row work. Mos sets Reasoning to High in the T3 pane and confirms
+      it, because `t3_create_thread` can't set effort. Researcher moves at
+      the next row boundary after that, never mid-row (Q129).
+125. **Row 63 lands before row 90 (2026-10-12).** Dewey found that the
+    two candidates share five paths. Each patch applies alone on
+    a0005d62, but stacked in either order the second one fails on
+    `packages/tasks/README.md` and `packages/tasks/tests/verbs.test.mjs`.
+    So whichever row lands second needs a rebase and a new round.
+    - Row 63 lands first. A new round on row 63 would need both Dewey
+      and Filbert again, while one on row 90 needs only Filbert, and
+      Dewey has offered the rebase. Row 63's round 2 was also requested
+      first (22:43Z, before row 90's 22:48Z).
+    - If row 90 gets its approval first, it waits. Once row 63 lands,
+      Dewey rebases row 90 on the landed commit and requests round 2
+      from Filbert. That round's packet says which hunks changed.
+    - If row 63 needs a round 3, the order still holds unless row 90 has
+      been ready for a full working session while row 63 is still under
+      review. Then Sage looks at it again.