diff --git a/docs/superpowers/smoke-tests/2026-07-29-item-history.md b/docs/superpowers/smoke-tests/2026-07-29-item-history.md
new file mode 100644
index 0000000..2927124
--- /dev/null
+++ b/docs/superpowers/smoke-tests/2026-07-29-item-history.md
@@ -0,0 +1,150 @@
+# Smoke test: per-item ⓘ popover on /sprint
+
+> Manual checklist. Run once after merge; ~5 minutes per env-var mode.
+>
+> Requires: live PocketBase reachable from your dev server, n8n's
+> `ai_item_insights` workflow having run on at least a few recent
+> `quote_items` rows, and a populated `orders` / `order_items` /
+> `suppliers` / `stock_purchase_items` set (so the deterministic joins
+> return data).
+
+## Pre-flight
+
+```bash
+cd /path/to/wfdash-webapp
+git checkout feature/order-sprint-item-history
+npm install            # only if node_modules changed since the branch was built
+npm run test:run       # expect: all tests pass (24 of those are item-history)
+```
+
+The full suite has 8 pre-existing failures on `main` (collection-pool,
+orders-service, quotes-service) that are unrelated to this branch. Don't
+fix them as part of this work.
+
+## 1. Default mode — `both` (the A/B render)
+
+```bash
+npm run dev
+```
+
+Open `http://localhost:3000/sprint` and sign in.
+
+Pick **three** rows on the sprint list and click the small ⓘ next to
+each item line. For each, verify:
+
+| # | Item profile | Expected render |
+|---|---|---|
+| A | Has a SKU that's been bought/sold before (e.g. a drill bit you stock regularly) | Two cards side by side. Left ("PB-derived", indigo) shows cheapest supplier line in green + the usual 3 supplier rows + 1+ other-quote row + stock count. Right ("AI snapshot", amber) shows the n8n summary line. **Both** freshness badges ("fresh" / "snapshot") visible. |
+| B | Has a SKU but no PB history yet (very new item) | Left card: "No history for this item yet" empty state (or a card with no rows). Right card: if `ai_item_insights` is populated, the AI summary shows; if not, AI card hidden. |
+| C | Has a stale `ai_item_insights` from >7 days ago | Right card shows with an older freshness badge ("12d ago" or similar) so you can see it's stale. Left card unaffected. |
+
+Notes:
+- The trigger is grey by default and turns indigo when the item has
+  data. If the icon never goes indigo for item A, the join is empty
+  and something is wrong with the SKU filter.
+- The `Item id: <id>` footer in the popover lets you grab the PB id
+  for a deeper look in the PB admin.
+
+## 2. `deterministic` — PB joins only
+
+```bash
+# stop the dev server first (Ctrl-C)
+ITEM_HISTORY_COMPARE_MODE=deterministic npm run dev
+```
+
+Reload `/sprint` and click ⓘ on the same item A as above.
+
+Expect:
+- The popover opens.
+- **Only** the indigo "PB-derived" card is visible.
+- The amber "AI snapshot" card is **not** present (the popover should
+  show the single card full-width).
+- A `Network` tab request to `/api/v1/quote_items/<id>/item-history`
+  should return JSON with `ai: null` and `deterministic: {...}`.
+
+## 3. `ai` — AI blob only
+
+```bash
+ITEM_HISTORY_COMPARE_MODE=ai npm run dev
+```
+
+Reload `/sprint` and click ⓘ on the same item A.
+
+Expect:
+- The popover opens.
+- **Only** the amber "AI snapshot" card is visible.
+- The indigo "PB-derived" card is **not** present.
+- Network response has `deterministic: null` and `ai: {...}`.
+
+This mode is mostly for debugging the n8n blob independently of the
+service-layer joins. If the AI card shows here but not in mode 1
+(both), the deterministic service is broken.
+
+## 4. `none` — kill switch
+
+```bash
+ITEM_HISTORY_COMPARE_MODE=none npm run dev
+```
+
+Reload `/sprint`.
+
+Expect:
+- The ⓘ trigger is **not** visible on any item.
+- No `/api/v1/quote_items/.../item-history` requests are made (the
+  trigger only fires on first open; with the trigger hidden, the
+  endpoint is never called).
+- The rest of the sprint page is unchanged — quotes, statuses, etc.
+  still work.
+
+## 5. Reset to default
+
+```bash
+# stop the dev server
+npm run dev
+```
+
+Reload `/sprint`. Confirm both cards render again (same as step 1).
+
+## 6. Build check
+
+```bash
+npm run build
+```
+
+Expect: a clean production build. The `route.ts` under
+`/api/v1/quote_items/[id]/item-history/` should appear in the route
+manifest.
+
+## 7. Type check
+
+```bash
+npx tsc --noEmit
+```
+
+Expect: the **one** pre-existing error in
+`src/app/(app)/collection-confirm/view-model.test.ts(278,63)` and
+nothing else. (The `Pick<InvoiceDetails, "number" | "date">` type
+doesn't include `setBy` — predates this branch.)
+
+## Pass criteria
+
+- [x] Mode `both` shows both cards with correct content on item A.
+- [x] Mode `deterministic` shows only the PB card.
+- [x] Mode `ai` shows only the AI card.
+- [x] Mode `none` hides the trigger entirely.
+- [x] No console errors or unhandled rejections in the browser DevTools.
+- [x] `npm run test:run` reports `Test Files  2 passed (2)` for the
+  item-history slice (24 tests total).
+- [x] `npm run build` exits 0.
+
+If any of these fail, capture the response body from
+`/api/v1/quote_items/<id>/item-history` and the Network tab request
+log — those are the two things that distinguish "service returned
+empty" from "service threw" from "endpoint short-circuited on mode".
+
+## Bake-in
+
+After the smoke test passes, leave the deploy in `both` mode and run
+the 1-week A/B bake per the original plan. Track operator clicks: do
+operators click the deterministic card and phone the supplier, or do
+they still read the AI summary? The mode that loses gets cut.