`verification/tk10836/OPTION-D-RESULT.md`

164 lines
# TK-10836 — Option D result (REVISED after contrarian red-team, 2026-09-10)

READ-ONLY throughout. `SHOPIFY_FULL_ACCESS_TOKEN` (…2ea5). Cost $0.
Zero writes: no order, draft order, discount, shipping-config, or email.

> **This file was rewritten after Cody the Contrarian broke my first version.**
> The first version claimed "158/192 = 82% free, hypothesis falsified, promise deliverable,
> blocker withdrawn." That averaged seven good months over a collapse that begins the week
> Kelly checked out. Both the original claim and the correction are recorded below.

---

## PART 1 — What is PROVEN about Kelly's cart (survives red-team)

**Option D was never actually blocked.** The memo said it needed "a Storefront token +
`read_shipping`." Scope enumeration across all 7 stored Shopify tokens: `SHOPIFY_FULL_ACCESS_TOKEN`
(…2ea5) already carries `read_shipping`. The `ACCESS_DENIED` was a wrong-token artifact.

**Kelly's cart, `AbandonedCheckout/34015098110003`** (kyounge@umich.edu, 2026-08-24T03:36:31Z,
Ann Arbor MI 48108): 3 lines — DWCC-600006 / 600045 / 600128-Sample — items $0.00, discount
$12.75, tax $0.00, **TOTAL $24.95 = pure shipping**.

**All 3 variants → delivery profile `Samples — Free Shipping (No Tracking)`.** Full untruncated
dump of that profile's **Domestic** zone (4 methods, all active):

| rate | price | condition |
|---|---|---|
| **Free Shipping (No Tracking)** | **$0.00** | TOTAL_PRICE ≥ $0.00 AND ≤ $45.00 |
| Priority Sample Only | $24.95 | TOTAL_PRICE ≥ $0.00 AND ≤ $20.00 |
| ups_shipping | carrier | (none) |
| fedex | carrier | (none) |

$0.00 satisfies **both** priced conditions. `deliveryCustomizations` = none. `shopifyFunctions` =
none. No custom carrier service emits these names. **Nothing in configuration suppresses the free
rate on her cart.** Corroborated: **12/12** *paid* comped sample orders sampled resolve to the
**same profile** as Kelly's — identical menu, different selection. Free orders exist in her exact
window (#32901, 2026-08-20, VA, 3 comped lines, $0.00) and in Michigan (2/2).

**Therefore:** an abandoned checkout records the **selected** shipping line, not the **available**
set. The prior reversal (`ec11bce`) read availability off a selection. That specific inference was
wrong, and `Priority Sample Only` $24.95 was on the menu next to $0.00.

Also: her cart was abandoned **2026-08-24T03:36Z**, *before* info@ gave her code 3FREE
(2026-08-25 08:38 PT) — it was never a post-honor attempt. 3FREE is moot regardless: her items were
already $0.00 and 3FREE discounts items only.

## PART 2 — What Cody broke, and what I confirmed against my own data

My 82% headline **smeared a cliff**. Re-bucketed from my own `rate-vs-subtotal.json`
(comped, US, 1–4 sample lines = Kelly-shaped):

| month | n | free % |   | window | n | free % |
|---|---|---|---|---|---|---|
| 2026-02 | 33 | 100% |  | Aug 1–17 | 3 | 100% |
| 2026-06 | 10 | 100% |  | Aug 18–24 | 3 | 67% |
| 2026-07 | 12 | 100% |  | Aug 25–31 | 14 | 57% |
| 2026-08 | 20 | **65%** |  | **Sep 1–10** | **31** | **35%** |
| 2026-09 | 31 | **35%** |  | | | |

**Cody was right.** Seven months near 100% averaged with a collapse that starts the week Kelly
checked out and is *still worsening today*. 82% is true and irrelevant — it is not the number that
applied to her. Paid rate mix since Aug 18: **UPS® Ground ×23, Priority Sample Only ×4.**

Also confirmed from Cody: rate name `"Free Shipping"` (no suffix) is retired (last seen 2026-03-17)
and silently supplied 69 of my 158 "free" hits — undisclosed rate-lineage pooling on my part.

## PART 3 — Mechanisms I proposed and then KILLED (recorded so nobody re-runs them)

1. **"Free rate suppressed at a $0.00 subtotal."** FALSE — both conditions have a $0.00 lower bound,
   and 158 comped carts shipped free.
2. **"Weight gate."** FALSE — the Domestic zone has only TOTAL_PRICE conditions, no TOTAL_WEIGHT.
3. **"Sample variants leaking into the General profile, so UPS Ground replaced the free rate."**
   FALSE — `ups_shipping` and `fedex` are *inside* the Samples profile; 12/12 paid comped orders are
   in that profile. (I nearly propagated this; the untruncated dump killed it.)
4. **"Shop Pay / Login-with-Shop accelerated checkout preselects a paid rate."** FALSE — Shop-tagged
   share is **88% in both windows**, and free% fell in *both* the Shop-tagged group (100%→48%) and
   the untagged group (100%→17%). Not a Shop Pay effect.
5. **"Multi-profile cart mismatch."** FALSE — exactly one `Samples —` profile exists store-wide.

## PART 4 — Honest state

**Kelly (this ticket).** The $0.00 option was on her menu. The emailed promise is *achievable*, and
the ticket's "the checkout could not deliver it" premise is not supported. But I cannot show she was
not **steered** to the paid option, because I never established what checkout **preselects**.

**The real finding (bigger than Kelly).** There is a genuine, worsening, **unexplained** collapse in
free-shipping selection on comped sample carts beginning ~2026-08-18: 100% → 48% → 35%, same
profile, rate still active and still chosen by some, not explained by any of the five mechanisms
above. ~27 customers in three weeks paid **$16–25 shipping on "free" samples**. Volume also rose
~6× in the same window (25 orders in Jun 1–Aug 17 vs 48 in Aug 18–Sep 10), so a **population change
is an unexcluded confound** alongside a possible checkout change.

**The one unrun test** (both Cody and Codex named it): a **live checkout render** on a $0-item,
3-line US sample cart — read the offered options *and which is preselected*. That requires either a
Storefront API token (we hold none) or driving the live storefront in a browser, which creates a
real abandoned-checkout record and may trip an abandoned-cart email. **Not run — externality,
Steve's call.**

## PART 5 — Effect on the fork
- **D** — answered for *availability*, **NOT** answered for *default/steering*. Root cause open.
- **A** (send her the working instruction) — supported and low-risk; still an outbound send = GATED.
- **B** (comp and ship) — still available as unconditional honor; GATED.
- **C** (write-off) — weaker now: this is not a one-customer anomaly.
- **TK-11366** — **escalates, not de-escalates.** It is the incident; Kelly is the detected instance.

---

# PART 6 — LIVE CHECKOUT RENDER (run 2026-09-10, Steve-authorized)

**Result: BLOCKED at a mandatory customer sign-in. Could not reach the shipping step.**

Reproduced Kelly's cart on the live storefront (`/cart/add` × her 3 real sample variants:
44723768688691 / 44723776454707 / 44723792674867). Then tried three independent routes to checkout:

| route | outcome |
|---|---|
| `/cart/<vid>:1,…` permalink | → `shopify.com/authentication/1541177456/login` |
| `GET /checkout` | → same forced sign-in |
| clicking the real **Checkout** button (`button.cart-checkout`) | → same forced sign-in |

**The DW store requires a customer account to check out.** Getting past it means creating or logging
into a customer account — an identity action, hard-gated — so I stopped. **No checkout was created,
no email entered, therefore no abandoned-checkout record and no abandoned-cart email.** The only
side effect was a browser-session cart. Actual externality: **zero**.

### Two findings the attempt produced anyway

**1. The comp is ACCOUNT-GATED, not automatic.** As a signed-out guest, 6 sample units priced at
**$25.50 — not $0.00**. The cart banner reads *"Existing customer? Sign in… Retail accounts receive
3 lifetime complimentary samples."* So the free-sample discount only applies to a signed-in retail
account. Kelly's cart *was* comped ($0.00, discount $12.75), so her account state was correct.

**2. The store's own cart copy states the shipping rule:**
> *"Sample shipping: orders of **10 samples or fewer ship free (no tracking)**. Sample orders **over
> 10 units** include a shipping charge."*

That is exactly the `TOTAL_PRICE ≤ $45.00` condition: 10 × $4.25 = $42.50 ≤ $45 < 11 × $4.25 = $46.75.
Cart also displayed **"Free sample shipping applied."**

## PART 7 — I found and fixed an error in my OWN cohort metric
My earlier cohorts counted **line items**, not **units** — so a single line of qty 12 was scored as
"1 line". Recounted by units over 502 US sample-only orders since 2026-06-01:

- **The >10-unit rule does NOT explain the paid orders.** 62 paid orders were **under** $45 list —
  most with **1–3 units**. Examples: #33033 (1 unit, $4.25 list, comped to $0.00) paid **$23.13 UPS
  Ground**; #33020, #33014, #33019 identical shape. Nobody rationally pays $23 to ship one $4.25 swatch.
- **Cart size did not change.** Median units per order = **1** in both windows.
- **The break is stark and stands after the correction:**

| window | n | free % |
|---|---|---|
| Jun 1 – Aug 17 | 405 | **97%** |
| Aug 18 – Sep 10 | 97 | **43%** |

## PART 8 — Where the root cause now stands
Free-rate availability is proven at the **config** layer and disproven as a >10-unit effect, a
weight effect, a profile effect, a Shop Pay effect, and a cart-size effect. What remains is a **real,
unexplained, still-worsening regression starting ~2026-08-18** in which single-swatch comped orders
are shipping on $17–24 UPS Ground instead of the $0.00 rate that qualifies for them.

**The decisive observation is one screen deep behind a login I will not bypass:** does the shipping
step still list *Free Shipping (No Tracking)*, and is it preselected? A signed-in human sees this in
~30 seconds. That is now the single blocking step for TK-11366.