fix(t3x): apply the sync test-timeout to every gate, not just fork CI

.github/workflows/t3x-ci.yml

@@ -55,9 +55,13 @@ jobs:
       # Proven upstream-inherited, not a fork regression: branch `t3x/ci-control-upstream` — pristine
       # upstream/main plus this workflow file and nothing else — fails identically (run 30386125253).
       #
-      # 60s is not arbitrary: it is what the ROOT vite.config.ts already uses. This only stops
-      # apps/web's tighter local override from being applied on slower hardware than it assumes.
+      # 120000 is deliberate: it is the HIGHEST value any package configures
+      # (apps/server/vite.config.ts pins 120s for sqlite+git+orchestration integration tests). The
+      # flag applies to every package, so a lower value would silently CUT apps/server's budget in
+      # half — the same failure class this override exists to prevent. At 120s no package is
+      # lowered: web 15s -> 120s, root 60s -> 120s, server 120s -> 120s.
       # The job's own `timeout-minutes` remains the backstop against a genuinely hung test.
+      # Keep in sync with the `test` case in `scripts/t3x/sync-upstream.sh`.
       - name: Test
         if: ${{ !cancelled() }}
-        run: vp run test --testTimeout=60000
+        run: vp run test --testTimeout=120000

.github/workflows/t3x-sync-resolve.yml

@@ -114,8 +114,12 @@ jobs:
                where they did NOT textually conflict, and adjust the fork's patches if upstream
                changed the semantics of an API the fork hooks into. If a fork patch became
                empty (upstream absorbed it), drop it and note that.
-            4. Verify: run `vp run typecheck`, `vp run lint`, then `vp run test`. Fix the fork's
-               patches until all three pass.
+            4. Verify: run `vp run typecheck`, `vp run lint`, then
+               `vp run test --testTimeout=120000`. The --testTimeout flag is required: apps/web
+               pins 15s, which is too tight for this fork's 2-core runners, and 120s matches the
+               fork's CI gate without lowering any package's own budget. Note the flag must NOT
+               come after a `--` separator or it is silently ignored. Fix the fork's patches until
+               all three pass.
             5. Push the branch: `git push -u origin HEAD`.
             6. Open a PR into `main` with `gh pr create`, title "t3x: resolve upstream sync
                (run ${{ github.run_id }})". In the body summarise: the upstream range merged,

docs/t3x/SEAMS.md

@@ -2,8 +2,8 @@
 
 **The authoritative list of every upstream-owned file this fork edits.**
 
-Measured, not asserted: **34 upstream-owned files, +1466 / -112 lines**, against merge-base
-`887dd6e45` (the 2026-07-28 upstream sync). Everything else the fork adds lives in new files upstream
+Measured, not asserted: **34 upstream-owned files, +1468 / -112 lines**, against merge-base
+`b64ae880e` (the 2026-07-28 upstream sync). Everything else the fork adds lives in new files upstream
 has never seen and cannot conflict.
 
 The churn and risk columns are still measured against the _previous_ merge-base `89c5a192f`, because
@@ -21,6 +21,11 @@ recurring rebase conflict in a file this doc said the fork did not touch) went u
 >
 > **Tripwire:** the surface is already far past "a handful of rows". Before adding row 35, re-isolate
 > something instead. Prefer fork-owned files even when an in-place edit is smaller.
+>
+> **Self-reference:** if your change edits a file that already has a row here, update that row and
+> the header totals **in the same commit**. This ledger measures the tree the commit creates, not the
+> tree it started from — a commit that edits `docs/providers/claude.md` and leaves the row alone
+> makes this document wrong the moment it lands. That has already happened once.
 
 ## Reading the risk column
 
@@ -67,7 +72,7 @@ Sorted by risk, worst first.
 | `apps/mobile/src/native/T3ComposerEditor.native.tsx`                    | +3       | 1     | **3**    | Plumbs `onComposerSubmit`                                                                                                                        |
 | `apps/mobile/src/components/AppSymbol.tsx`                              | +2       | 1     | **2**    | `return:` icon entry                                                                                                                             |
 | `apps/mobile/…/T3ComposerEditorModule.kt`                               | +1       | 1     | **1**    | Event-name list entry                                                                                                                            |
-| `docs/providers/claude.md`                                              | +76/-31  | ~0    | **~0**   | Fixes the broken multi-account recipe. Near-frozen upstream (last touched 2026-04-29) — **the one row worth upstreaming**, which would remove it |
+| `docs/providers/claude.md`                                              | +78/-31  | ~0    | **~0**   | Fixes the broken multi-account recipe. Near-frozen upstream (last touched 2026-04-29) — **the one row worth upstreaming**, which would remove it |
 
 **Per surface:** `apps/web` 8 · `apps/mobile` 7 · `apps/desktop` 7 · `apps/server` 5 ·
 `packages/**` 5 · `docs/` 1 · repo root 1.
@@ -123,12 +128,15 @@ That prints exactly the upstream-owned files the fork edits. Churn for any one o
 the merge-base date, **not** to today, so the number is reproducible after the fact:
 
 ```bash
-SINCE=$(git show -s --format=%cI "$MB")
-git log --oneline --since="$SINCE" --until="$SINCE" --before="$SINCE" "$MB" -- <path> | wc -l
-# simpler and equivalent: the 60 days of upstream history ending at the merge-base
-git log --oneline "$MB" --since="$(git show -s --format=%cI "$MB") -60 days" -- <path> | wc -l
+MBTS=$(git show -s --format=%ct "$MB")                     # merge-base commit time, epoch seconds
+git log --oneline --since="@$((MBTS - 60 * 86400))" "$MB" -- <path> | wc -l
 ```
 
+The `@<epoch>` form is load-bearing. Git's approxidate parser **silently ignores** a relative suffix
+on an absolute date, so `--since="<iso-date> -60 days"` is treated as `--since="<iso-date>"` — an
+empty window that returns 0 for every file. Verify the command is working before trusting it:
+`apps/web/src/components/ChatView.tsx` should return a number in the 50s, not 0.
+
 **Re-run both after every upstream sync and update this file — including the merge-base hash in the
 header.** Every churn and risk figure shifts once the newly absorbed commits fall inside the window,
 so a ledger quoting an old merge-base is stale even when its file list is still right. If the

docs/t3x/sync-agent-runbook.md

@@ -106,8 +106,14 @@ This is the checklist the workflow prompt mirrors — follow it if you resolve l
    still behaves.
 4. If a patch commit went **empty/dropped**, upstream absorbed it — confirm the behaviour now
    exists upstream, drop the patch, and note it.
-5. Verify: `vp run typecheck && vp run lint && vp run test`. Fix the fork's patches to match
-   upstream's new internals until green.
+5. Verify: `vp run typecheck && vp run lint && vp run test --testTimeout=120000`. Fix the fork's
+   patches to match upstream's new internals until green.
+
+   Two notes. **`AGENTS.md` says "do not run repo-wide checks" — an upstream sync is the sanctioned
+   exception.** A rebase can break any package, so the full suite is the point; that rule is about
+   routine feature work. And the `--testTimeout` flag is required (see the section above); it must
+   not follow a `--` separator or it is silently ignored.
+
 6. **When green:** push the branch and open a PR into `main` (`gh pr create`). A human reviews
    and **lands it — see [Landing a sync PR](#landing-a-sync-pr-do-not-use-the-github-merge-button);
    the GitHub merge button does not work on a rebased branch.** The recovery tag

scripts/t3x/sync-upstream.sh

@@ -128,10 +128,17 @@ if [[ "$SKIP_VERIFY" == 1 ]]; then
   exit 0
 fi
 
+# Per-script extra args, so the daily verify matches the fork's CI gate exactly. `test` needs
+# --testTimeout because apps/web pins 15s (tuned for upstream's blacksmith runners) and this fork
+# runs on 2-core ubuntu-latest, where an upstream CPU-bound test reliably blows it. 120s is the
+# highest any package configures (apps/server), so it never LOWERS a package's own budget.
+# Keep in sync with `.github/workflows/t3x-ci.yml`.
 for script in $VERIFY; do
-  echo "→ verify: $RUN $script"
+  EXTRA=""
+  if [[ "$script" == "test" ]]; then EXTRA="--testTimeout=120000"; fi
+  echo "→ verify: $RUN $script $EXTRA"
   # shellcheck disable=SC2086
-  if ! $RUN "$script"; then
+  if ! $RUN "$script" $EXTRA; then
     FAILING_STEP="$script"
     fail "verification step '$script' failed"
     write_status daily verify-failed "rebase clean but '$script' failed"