Door Comparison — what the offer costs, and what the store does

The finding that moved the design

What was verified about the Exchange

The three-level model

The ID.me finding forces a clean separation the first draft did not have. Each fact lives at the level it is actually true at, so it is stored once and cannot disagree with itself.

idmeShop carries a single brandHandoff mode, which is the ID.me finding encoded where it cannot be forgotten. The section below explains why one flag was not enough for the other channels.

GovX is two products, and the app already models them

So a brand appearing "both ways" sells on GOVX.com and gates a discount on its own site behind GOVX ID. In Basefox terms those are already two distinct routes:

What merchant: true/false still gets wrong

One policy shape across the exchanges, with CGX the outlier

The genuine variation is narrow: a 95-cent threshold gap, NEX alone stating its threshold is measured after discounts, and CGX as the real outlier with a band table and a $150 bar. Two exchanges price a tier they will not publish — AAFES's Priority upcharge sits on each product page and CGX's USPS rate is computed at checkout. Neither can ever be stated by Basefox, so the model must be able to say that rather than omit the method.

Waivers, and who can apply them

An honest correction to the first draft

What the member sees

NIKE · BRAND DIRECT                    −30%
  discount     −30% via SheerID
  shipping     free over $50, else $6.95
  ship time    3–5 business days
  sales tax    collected · TX ~8.20%
  price match  no
  adjustment   within 14 days of purchase
  ✓ verified Aug 30 · Aug 30 sweep

NIKE · ID.ME SHOP                      −25%
  discount     −25% via ID.me
  ships to     nike.com — same store as above
  exclusions   sale items
  ✓ verified Aug 30 · Aug 30 sweep

AAFES · EXCHANGE                    varies
  discount     varies by item
  shipping     free — MILITARY STAR card
               (else free $49+, otherwise $4.95)
  ship time    7–10 business days
  sales tax    none collected
  price match  yes · Exchange Price Match
  pickup       on-post, free
  ⓘ also free with an MCSS item in the order
  ✓ policy checked Aug 31 · source
  ⓘ Basefox has no Exchange listing for Nike.
     Check on base or at shopmyexchange.com.

Two details fall out of the ID.me finding and matter:

• A non-merchant door does not repeat the store's rows. The ID.me card says "ships to nike.com — same store as above" rather than reprinting freight and tax. Printing them twice would imply two stores where there is one.
• The exchange card carries its own dated policy stamp (policy checked Aug 31), distinct from a route's verified stamp. A standing policy is a different kind of claim from a swept offer and must not borrow the sweep's seal.
• A waiver that fires says which one fired. free — MILITARY STAR card teaches the member why, and it is the app showing its work: a member without the card sees free $49+, otherwise $4.95 instead, and the two readings never silently swap.

The calculator

Default — one price. Type what the item costs; each door applies its own discount, then its own shipping and tax. Zero friction, and correct for the common case where two doors are the same merchant.

Override — prices differ by site. A toggle opens a per-door field. This is not polish: the Exchange does not apply a percentage, it has its own sticker price. Without per-door entry the AAFES column would assume the Exchange lists the same price as the brand — the exact assumption the comparison exists to remove.

Gate checks

Sequencing

1. Channel policy end to end. FulfillmentMode, ShippingWaiver, Shipping; schema; Swift model; staleness and route-id-uniqueness gate checks. AAFES, NEX and CGX captured; mcx retired and cgx added. All five surviving channels captured. The NEX/CGX tax posture is the one remaining gap.
2. Sales tax and member attributes. salestax.yaml, generator, SalesTax.swift, Profile gaining zip, taxRateIsManual and hasMilitaryStarCard. The ZIP table also resolves apoFpoAddress.
3. Comparison surface, channel-level only. Stacked cards on channel policy plus the tax rate. First visible slice, and it already answers the AAFES question — before any brand-level extraction exists.
4. Merchant policy + extraction. MerchantPolicy on Brand, Shipping, Conditions.excludes, Extraction fields, INTAKE_FIELDS widening, re-mine, uniformity check.
5. Targeted fetches. shippingURL per brand, fifteen dual-door brands first.
6. GovX fulfillment mode. Teach the GovX source to record retail vs. handoff, and let a brand publish both with the suffixed id. Until this lands, GovX cards read fulfillment unknown and state no freight or tax.
7. Calculator. One-price default, per-door override, LandedCost, optional-basket ranking.

Steps 1 and 2 are independent. Step 3 depends on both and is the first shippable slice.

Final review

What this gets right

• It carries facts a percentage cannot express. Ship time, price match, a 14-day price adjustment, on-post pickup — none is expressible as a discount fraction, and each changes a decision.
• Every fact is stored at the level it is true at. The three-level split means a store's freight is written once and cannot disagree with itself across two doors.
• The AAFES comparison ships in step 3, without waiting on exchange routes or brand extraction. The first draft could not do this.
• Two waivers are member attributes the app can actually resolve. A MILITARY STAR holder and a deployed member each get a materially different answer, and the ZIP table loaded for sales tax resolves the second one for free.
• Ship time is carried but never folded into the total, so a 45-business-day APO delivery visibly outweighs a cheaper price instead of hiding inside one.
• Four channels are captures rather than summaries. Rendering the pages produced facts a paraphrase dropped, resolved $49.00 vs $49.95 as two real thresholds on two exchanges, retired a channel that was never a storefront, and surfaced a channel the app was missing entirely.
• MILITARY STAR® waives freight at all three exchanges, across thresholds spanning $49 to $150. One profile toggle changes the answer at three channels at once.
• A hazard was retracted rather than carried. The GovX route-id collision rested on an assumption the capture disproved. GovX needs no new modelling: the existing channel/verify split already separates GOVX.com from GOVX ID.
• Published policy beats compiled policy. A wrong threshold is fixable by a sweep, and it carries a date and a source like every other claim in the app.
• It ships with its own detectors, one per failure mode it introduces.

Where it is weak

• Brand-level merchant policy will be sparse for a while. Shipping and price-adjustment policies are not on the discount pages already cached, so steps 4–5 yield less than the first draft implied. Most brand cards will read not stated until targeted fetches run.
• Channel policy is hand-maintained and will rot. Five rows is small, but nothing re-reads those pages automatically. The staleness check makes rot visible; it does not fix it.
• State-average tax rates are wrong for specific members by design. A member in Chicago pays 10.25% against an Illinois average near 8.9%. The override is one tap away, but the first number shown to them is too low.
• "Does the Exchange carry this brand" is still unanswered. The column is factually solid on policy and silent on stock. A member can be shown a compelling tax-free comparison for a brand AAFES does not carry. The copy says so; that is mitigation, not a fix.
• The capture needs a browser, which the pipeline does not have. curl gets an SPA shell; a rendered session gets the policy. Channel policy therefore cannot be re-read by an unattended sweep the way a brand page can, so the staleness check can only flag rot, never repair it. The strongest argument yet for the deferred scraped-drift option, now with a concrete blocker attached.
• NEX and CGX sales-tax posture is unread. Exchanges are generally tax-free, and "generally" is not a citation. Neither shipping page states it, so neither row asserts it until a tax page is captured.
• GovX publishes no shipping fee publicly. The card can state that a fee exists — returns are refunded minus shipping — without stating what it is. Correct, and less useful than a number would be.
• GovX ship time is per product, not per channel. Brand partners fulfil, so a channel-level estimate would be a fiction. The card says where to look rather than inventing a number.
• MCSS and pickup waivers are stated and never computed. A member whose order qualifies through an MCSS item sees a total higher than they will pay. Understating the Exchange is the safe direction of that error, and it is still an error.
• Tax on shipping is ignored. ~20 states tax delivery charges; on $7 of freight at 8% that is 56 cents, below the noise floor of the rate average. Footnoted.
• The route board is accumulating surfaces. Cards, cost lines, blocker lines, a comparison view, a calculator. Worth a design pass once real data is in it.

Adjustments worth considering

• Ship steps 1–3 and stop to look. That slice delivers the AAFES comparison — the part of the request with no existing answer — without touching extraction. Judge 4–6 on how it feels in the hand.
• Reconsider scraped drift detection early if channel policy moves more than annually. Staleness in this data is the risk the plan is least able to absorb.
• Consider shipping in diff._MATERIAL so a threshold moving $50 → $75 reaches the What's New feed. Held back because each field needs its own headline sentence or the feed emits something wrong.

Confidence