diff --git a/bundle-budget.json b/bundle-budget.json
index 0aa30622d5..f1beb4046c 100644
--- a/bundle-budget.json
+++ b/bundle-budget.json
@@ -2,7 +2,7 @@
   "$comment": "Client JS bundle-size budgets captured from a known-good production build. `production` guards aggregate user-facing chunks, `routes` guards the five Lighthouse journeys against route-local growth, and `mockups` is a looser design-scratch hygiene ceiling. Refresh intentionally with `npm run check:bundle-budget -- --update`.",
   "enforce": true,
   "production": {
-    "gzipBytes": 1405202,
+    "gzipBytes": 1518033,
     "tolerancePct": 10
   },
   "mockups": {
@@ -33,6 +33,6 @@
   },
   "totalGzipBytes": 1708296,
   "tolerancePct": 10,
-  "updatedAt": "2026-08-13T14:08:49.522Z",
-  "baselineSource": "ca788d41e1e6b64dc6b9bc63609074d92470d7e7"
+  "updatedAt": "2026-08-18T00:00:00.000Z",
+  "baselineSource": "9d832452dc317844349d178683ba15d185eceb5d"
 }
diff --git a/docs/branch-review-records/eb993f9206e1027b9919ddfcf00607246f6490249b5ce4f864db107ad619319d.record.md b/docs/branch-review-records/eb993f9206e1027b9919ddfcf00607246f6490249b5ce4f864db107ad619319d.record.md
new file mode 100644
index 0000000000..220f71905b
--- /dev/null
+++ b/docs/branch-review-records/eb993f9206e1027b9919ddfcf00607246f6490249b5ce4f864db107ad619319d.record.md
@@ -0,0 +1 @@
+| 2026-08-18 | claude/bundle-budget-production-baseline | f2cb4ec2ce11bb29444c9aca3c0572c603d48e5e | bundle-budget.json production baseline refresh + measurement evidence doc + outstanding-issues request | Approved — investigated before refreshing: main's +8.03% is diffuse feature growth across 52 of 101 routes with no regression signature and an unchanged lockfile; production baseline moved to the measured 1518033 bytes, route/mockups budgets and tolerances left untouched | two full builds (main 9d832452d and baseline ca788d41) with .next removed each time; baseline reproduced to within 32 bytes of the recorded value; check:bundle-budget passes at the new baseline with all five route budgets and mockups within tolerance; tests/bundle-budget.test.ts 33 passed; outstanding-issues + ledger-write-discipline + docs link guards passed; format clean. Full unit suite, lint and typecheck not run: the diff is one JSON baseline value plus documentation and an inbox request, with no source change and no plausible compile or behaviour path |
diff --git a/docs/evidence/bundle-budget-production-rebaseline-2026-08-18.md b/docs/evidence/bundle-budget-production-rebaseline-2026-08-18.md
new file mode 100644
index 0000000000..9bc20082f5
--- /dev/null
+++ b/docs/evidence/bundle-budget-production-rebaseline-2026-08-18.md
@@ -0,0 +1,93 @@
+# Production bundle-budget re-baseline — 2026-08-18
+
+`check:bundle-budget` compares the client JavaScript a non-mockup route can reach against a
+baseline captured from a known-good build. The `production` baseline in `bundle-budget.json` was
+captured on 2026-08-13 at `ca788d41` and never refreshed. By 2026-08-18 `main` measured **+8.03%**
+against it — inside the 10% tolerance, but with only ~2 points of headroom left, so the next
+feature to land would fail the gate for reasons that were not its own. PR #2096 (Dictionary) hit
+exactly that: +2.7 points of its own weight pushed the total to +10.5% and turned `Build` red.
+
+The rule in `AGENTS.md` is that a production failure means _find the regression_, not move the
+ceiling. This document is that investigation. It concludes the growth is distributed feature work
+with no regression signature, and refreshes the `production` baseline only.
+
+## Method
+
+Two full production builds from the same worktree and the same `node_modules`, with `.next`
+removed before each so no cached output could be measured:
+
+1. `origin/main` at `9d832452d`
+2. the recorded baseline commit `ca788d41`
+
+Client chunks were attributed to routes through the per-route `*_client-reference-manifest.js`
+files under `.next/server/app` — the same source `scripts/check-bundle-budget.mjs` uses — and gzip
+sizes were measured per chunk. Chunk filenames are content-hashed and therefore not comparable
+across builds, so the comparison is per route.
+
+`package-lock.json` is **byte-identical** between the two commits, so no dependency change
+contributes to the delta.
+
+## Result
+
+| Measurement       | Baseline `ca788d41` | Main `9d832452d` | Delta               |
+| ----------------- | ------------------- | ---------------- | ------------------- |
+| Production gzip   | 1372.2 KiB          | 1482.5 KiB       | +110.2 KiB (+8.03%) |
+| Production chunks | 239                 | 238              | −1                  |
+
+The reproduction is exact enough to trust: measuring `ca788d41` independently produced
+**1,405,170 bytes** against the **1,405,202 bytes** recorded in `bundle-budget.json` — a 32-byte,
+0.002% difference, which confirms both the method and that the recorded baseline really was
+captured at that commit.
+
+## Why this is not a regression
+
+Across the 101 non-mockup routes present in both builds:
+
+- **52 grew, 48 were unchanged, 1 shrank.** Median growth +8.2 KiB, maximum +24.5 KiB.
+- No route doubled, no route gained a large isolated chunk, and no new heavy route appeared. The
+  only route added since the baseline is `/api/documents/signed-urls`, which ships no client
+  JavaScript.
+- The shape is uniform growth in shared chunks — what 374 commits of feature work looks like, not
+  what an accidentally-bundled dependency looks like. A regression of that kind concentrates: one
+  chunk or one route jumps while the rest hold still.
+
+Largest per-route growth:
+
+| Route                           | Baseline KiB | Main KiB | Delta |
+| ------------------------------- | ------------ | -------- | ----- |
+| `/therapy-compass/[slug]`       | 215.7        | 240.1    | +24.5 |
+| `/therapy-compass/[slug]/brief` | 215.7        | 240.1    | +24.5 |
+| `/therapy-compass/[slug]/sheet` | 215.7        | 240.1    | +24.5 |
+| `/calculators`                  | 230.5        | 251.4    | +21.0 |
+| `/specifiers/map`               | 246.9        | 264.4    | +17.5 |
+| `/therapy-compass/search`       | 228.6        | 245.5    | +16.8 |
+| `/therapy-compass/compare`      | 219.0        | 234.7    | +15.7 |
+| `/therapy-compass/recommend`    | 218.6        | 232.8    | +14.2 |
+| `/therapy-compass/pathways`     | 218.1        | 232.2    | +14.1 |
+| `/therapy-compass/review`       | 217.0        | 231.1    | +14.1 |
+
+Those routes match where the code actually landed. Between the two commits `src/` changed by
+**+16,559 / −6,988 lines across 254 files with 42 new files**, and the largest client-side diffs are
+`diagnosis-map-panel.tsx`, `calculators/search-page.tsx`, `favourites-command-library-page.tsx`,
+`differentials-home.tsx`, `document-search-results.tsx`, `ClinicalSidebar.tsx`,
+`services-navigator-page.tsx` and `globals.css`.
+
+## What changed here
+
+Only `production.gzipBytes`, plus `updatedAt` and `baselineSource` so the next investigation knows
+which commit to reproduce from.
+
+Deliberately **not** changed:
+
+- **The five `routes` budgets.** Each is within its own 10% tolerance right now
+  (`/` 199.1 → 210.6, `/therapy-compass` 206.1 → 218.9, `/documents/search` 201.7 → 211.3,
+  `/dsm` 201.7 → 211.2, `/forms` 224.7 → 234.4). Leaving them tighter keeps a second, stricter
+  guard on the Lighthouse journeys, which is the point of having per-route budgets at all.
+- **The `mockups` budget**, which is within tolerance on `main` and is a separate hygiene ceiling.
+- **`tolerancePct`.** The tolerance is not the problem; the staleness was.
+
+## Follow-up
+
+Nothing schedules a baseline refresh, so the same squeeze will recur — the gate quietly converts
+accumulated growth into a failure for whichever unrelated PR lands last. Queued as an
+outstanding-issues request alongside this change.
diff --git a/docs/outstanding-issues-inbox/5d626edd-d62a-42b8-ac47-2e9717c99d33.json b/docs/outstanding-issues-inbox/5d626edd-d62a-42b8-ac47-2e9717c99d33.json
new file mode 100644
index 0000000000..71d4fc32bb
--- /dev/null
+++ b/docs/outstanding-issues-inbox/5d626edd-d62a-42b8-ac47-2e9717c99d33.json
@@ -0,0 +1,14 @@
+{
+  "version": 2,
+  "id": "5d626edd-d62a-42b8-ac47-2e9717c99d33",
+  "createdOn": "2026-08-18",
+  "action": "add",
+  "payload": {
+    "pri": "P2",
+    "type": "rec",
+    "summary": "Nothing schedules a bundle-budget baseline refresh, so accumulated growth fails whichever unrelated PR lands last",
+    "detail": "The production baseline sat at ca788d41 (2026-08-13) untouched while main grew +8.03% by 2026-08-18, leaving ~2 points of headroom. PR #2096 (Dictionary) then failed Build at +10.5% for 2.7 points of its own weight. Re-baselined once in docs/evidence/bundle-budget-production-rebaseline-2026-08-18.md, but the same squeeze recurs unless a refresh has an owner or a trigger: options are a scheduled job that re-measures and opens a PR, a drift warning threshold below the failure threshold, or recording the baseline commit distance in the check output so staleness is visible before it blocks someone.",
+    "source": "PR review of #2095/#2096, 2026-08-18",
+    "issueUlid": "01M09SPZ4QQSHHGK3QG194ARKD"
+  }
+}