07Methods and sources · retrieved 2026-09-28
The point of the labels is that you can check the work instead of believing it. Every claim in this case study traces to a row below, and every gap is written as a gap rather than filled in with something plausible.
In the order I did it. Each step feeds the next, and each one is on a page of this case study you can open and check.
Step 01
One retrieval, on 2026-09-28
I read only what any customer can see: the home page, the specials index, each deal page, the store locator, the sitemaps and the public FAQ. No login, no form, no order, no admin. Every page I relied on is one of the 17 entries in the source register below.
Step 02
30 deal cards → 28 unique offers
Each card was copied as it appears — headline, price, terms, disclaimer, link — before any judgement. Offers shown on both the home page and the specials index were merged, so each one is counted once.
Step 03
28 offers × 14 checks
Does it say which products, which store, until when, and on what terms? Does the link land on something you can buy? Every offer gets the same fourteen checks and one verdict. 2 of them block a customer from knowing what they will pay.
Step 04
Five labels, from observed to unknown
Anything I saw is marked as observed. Anything the company says about itself is marked as its own wording. Anything I could not see is either modelled — and says so wherever it appears — or written down as needing internal data. Nothing is filled in to look complete.
Step 05
7 confirmed observations, proposals kept separate
Each confirmed finding becomes a ticket a web team can pick up: the evidence, the current and desired state, acceptance criteria and how to verify it. Ideas that are mine rather than observed are labelled as proposals and never mixed in.
Step 06
5 metrics, each with its missing source
Every metric has a formula, a denominator, a time window and the internal source it would need. Where that source is not available from outside, the page says so instead of showing a number.
Read left to right, the ground gets softer. Every claim on every page carries one of these, so you always know which kind of floor you are standing on.
Read off a public page on the access date. Reproducible by anyone, and it can change tomorrow.
The company's own published wording. Quoted, not interpreted.
A handful of manual checks on one day. It is not a sample of customers and not an analytics report.
Synthetic and deterministic, built so the workflow can be demonstrated without inventing company performance.
Not knowable from outside. Listed so the gap is visible rather than guessed at.
The "cannot support" column is the useful one. It is where a case study usually overreaches, so it is stated by the author rather than left for a reviewer to find.
https://www.shoppuff.com/ · retrieved 2026-09-28
Supports
The home page leads with deal cards; 18 deal cards were present on the access date. Location names, headline categories and the public FAQ answers.
Cannot support
It cannot say which deals are live in which store today, or how often the card set rotates.
https://www.shoppuff.com/specials/ · retrieved 2026-09-28
Supports
12 deal cards with titles, exclusion sentences and slugs. Category filter chips: Flower, Vapes, Pre-Rolls, Edibles, Concentrates, All Items, Clearance.
Cannot support
There is no location filter on this page, no dates, and no list of which products a deal applies to.
https://www.shoppuff.com/specials/2-25-disposables/ · retrieved 2026-09-28
Supports
The company publishes, verbatim: "*Website cart may not reflect the bundled price but it will be adjusted at time of payment." The page also carries a title and an exclusion sentence and nothing else.
Cannot support
It cannot say how often that adjustment is needed, at which stores, or what it costs the business.
https://www.shoppuff.com/menu/ · retrieved 2026-09-28
Supports
The page is titled with product-for-sale language but the delivered HTML carries a deal grid and a store-selector element rather than product records — the product grid is rendered client-side.
Cannot support
Nothing about what the rendered client-side menu looks like, and nothing about stock. Only the server-delivered markup was read.
https://www.shoppuff.com/wp-json/joint-api/v1 · retrieved 2026-09-28
Supports
WordPress 7.1.2 with Elementor 4.2.4 (generator meta), and a public `joint-api/v1` REST namespace. This is a native first-party storefront on the dispensary's own domain — the product grid is not a third-party embed.
Cannot support
Which inventory source feeds it, or when a given product last synced. Only the existence and shape of the namespace was read.
https://www.shoppuff.com/products/mitten-extracts-disposable-2-in-1-flip-strawberry-gelato-mango-tango-for-sale-monroe-mi/ · retrieved 2026-09-28
Supports
This is the finding that shapes the whole case study. The page is scoped to one store in its URL, and its schema.org Product markup carries brand (Mitten Extracts), size ("2g"), price ("25.00" USD) and availability (InStock), with an image served from the ordering platform. **Price and stock are published per store, per product.**
Cannot support
One page. It says nothing about how many products carry complete markup, how often the price moves, or what any other store shows.
https://www.shoppuff.com/sitemap_index.xml · retrieved 2026-09-28
Supports
29 product sitemaps at 1,000 URLs each — roughly 29,000 store-scoped product URLs — plus separate recreational, medical, delivery and stores sitemaps. The product sitemaps carried a last-modified stamp of 2026-09-27, the day before this retrieval, consistent with a daily inventory sync.
Cannot support
It does not say which of those URLs are unique products versus the same product repeated per store.
https://www.shoppuff.com/categories/flower/ · retrieved 2026-09-28
Supports
A working sub-category taxonomy — Bulk Flower, Infused Bud, Infused Buds, Pre Ground, Premium, Shake Trim, Small Buds, Whole Flower — plus long-form category copy.
Cannot support
The page holds no product records and no prices of its own.
https://www.shoppuff.com/effects/ · retrieved 2026-09-28
Supports
Ten consumer-facing effect labels already exist: Happy, Relaxed, Calm, Energetic, Focused, Creative, Inspired, Sleepy, Clear Mind, Uplifted. Each links onward to a filtered shop view.
Cannot support
Nothing here states how a product is assigned to an effect, so the assignment rule is not observable.
https://www.shoppuff.com/stores/ · retrieved 2026-09-28
Supports
Eleven Michigan locations are listed: Bay City, Center Line, Kalamazoo, Madison Heights, Menominee (marked coming soon), Monroe, New Buffalo, Oscoda, River Rouge, Sturgis, Traverse City — plus Utica in the site navigation.
Cannot support
It does not state which deals each store honours.
https://www.shoppuff.com/faqs/ · retrieved 2026-09-28
Supports
The company's own summary of Michigan purchase limits: "In Michigan, an adult may possess up to 2.5 ounces of marijuana; up to 15 grams of marijuana may be marijuana concentrate." Online pickup ordering is confirmed.
Cannot support
It is consumer guidance, not a regulatory determination. It does not cover edible equivalency or medical-patient limits.
https://www.shoppuff.com/brands/jeeter/ · retrieved 2026-09-28
Supports
The site publishes per-brand pages. Brand names used in this case study's product sample were taken from these public pages and from the deal titles themselves.
Cannot support
It says nothing about which products of that brand are stocked at a given store on a given day.
https://www.shoppuff.com/ · retrieved 2026-09-28
Supports
A single retrieval compared the home page's deal cards with the specials index. 18 cards were present on the home page; 12 on the index; the two sets shared 2 titles. Four home-page cards carried a store prefix ("BC |", "Puff |").
Cannot support
One retrieval on one day is not a rotation study. Both surfaces may rotate, so this is a snapshot of drift, not a rate.
No external URL — generated by this project · 2026-09-28
Supports
A fictional 63-record product sample assembled from publicly named brands, categories and forms so the reconciliation workflow has something realistic to run on.
Cannot support
These are not Puff products. No real SKU, price, potency, stock level or image is claimed. Nothing here describes Puff's actual inventory.
No external URL — generated by this project · 2026-09-28
Supports
A deterministic, seeded set of 4,500 sessions used only to make the measurement module's formulas inspectable.
Cannot support
It is not Puff traffic. It is not a forecast. No uplift shown anywhere is an achieved result.
No external URL — a gap, not a source · 2026-09-28
Supports
Nothing. Listed to make the missing evidence explicit.
Cannot support
Zero-result searches, cart-price adjustments, order completion, basket size and store-level demand all live behind company-owned systems.
No external URL — a gap, not a source · 2026-09-28
Supports
Nothing. Listed to make the missing evidence explicit.
Cannot support
Whether a bundle rule exists in the cart, how it is scoped per store, and whether it fires correctly are all inside the ordering platform.
The first reading on 2026-09-28 was taken as notes and quotes, not screenshots. So on 2026-09-29 every source page was captured in full, the key pages were saved to the Internet Archive as neutral dated copies, and every confirmed finding was checked again. Deals rotate daily, so counts move. What matters is whether each finding still stands — and one does not.
On 2026-09-29 The index now shows 22 cards (12 on the 28th). Its filters are still Flower, Vapes, Pre-Rolls, Edibles, Concentrates, All Items and Clearance — no location control.
On 2026-09-29 No card on the home page or the specials index carries a start or end date.
On 2026-09-29 Still two different sets. Both deal lists have rotated; one title appears on both surfaces (two did on the 28th).
On 2026-09-29 Same sentence, word for word: "Website cart may not reflect the bundled price but it will be adjusted at time of payment."
On 2026-09-29 Still several wordings on the specials index alone, in mixed case, plus "Exclusions apply. Limited time offer." on the deal page.
On 2026-09-29 Not visible on the 29th. Neither prefix appears on the home page any more, so this one cannot be shown from the new captures.
On 2026-09-29 The home-page paragraph still names Hamtramck; the store locator still does not list it.
On 2026-09-29 The /specials/2-25-disposables/ page still carries a title, a terms sentence and the cart disclaimer, then the store list.
On 2026-09-29 The FAQ still says "up to 15 grams of marijuana may be marijuana concentrate", and "15/$69 Rocket Fuel 1g Live Resin" is still on the specials index.
What the role asks for, what this case study puts in front of you, where it came from, what had to be assumed, what the work cannot prove, the internal source that would close the gap, and the metric that would be used in a real engagement.
Rejected approaches are more informative than chosen ones. Each of these was considered and dropped for a stated reason.
Puff already has ten working effect labels and strain pages. Building a "find something for sleep" tool would have been a second front end over a taxonomy that exists, and the rule that assigns a product to an effect is not observable from outside. Building it would have meant inventing the thing it recommends on.
It would need a crawl of roughly 29,000 store-scoped URLs to say anything defensible, and the interesting failures live in the admin, not on the pages. Kept as a supporting issue rather than the centre.
The site already runs programmatic local pages, category copy and structured data. Recommending "improve SEO" without a specific defect would have been noise, and there is no evidence in this case study that SEO is where the money is lost.
The storefront is a native, store-scoped, price-carrying build. A redesign would have been the easiest thing to draw and the least useful thing to hand a hiring manager, because it demonstrates taste rather than judgement.
Cart adjustment rates, stock levels, search logs and conversion are all knowable internally and not from outside. Filling them in with plausible figures would have made this look more complete and be worth less.
Written as a list rather than a paragraph, because a limitation buried in prose is a limitation that was not really disclosed.
One retrieval, on one day. Both deal surfaces may rotate, and a rate cannot be established from a single observation.
The product sample is modelled. It is not Puff's inventory, and no real SKU, stock level or potency is claimed. Exactly one price in it is real, and it is labelled wherever it appears.
The session stream is modelled and deterministic. It is not Puff traffic and it is not a forecast.
The ordering platform, the cart and its bundle rules were not loaded or inspected. Nothing is claimed about how the cart actually behaves — only about what the company's own page says it may do.
The 14 checks are my design. They are the checks I would argue for in a first week with the promotion table in front of me, and they are the part most likely to change.
Effort estimates on the tickets are indicative. They are not a quotation and were not produced with the agency that would do the work.
The reference project's mobile behaviour could not be verified in the environment used, and is recorded as unresolved rather than observed.
No automated accessibility audit was run against this case study. Keyboard operation, focus visibility, contrast and reduced-motion behaviour were checked by hand; that is not the same as conformance testing.
Nothing here has been validated with Puff Cannabis. If the ordering platform already scopes promotions per store and applies bundle pricing correctly, then the operational half of this case study is unnecessary and what remains is presentational. That is a better outcome than being right, and it is the first question in the 30-day plan.
This is independent work by Mousa Batarseh. It was not commissioned, approved, reviewed or endorsed by Puff Cannabis Company, and it is not a claim that they need it. It does not modify any live system, and no account, form or order was used anywhere in the research.
What I claim: that I can read a business's public surface, find a specific and checkable gap, build something that closes it, show the operational work behind it, define the measurement honestly, and write it up so a reviewer can disagree with me precisely.
What I do not claim: any measured business result, any employment relationship with Puff Cannabis, any endorsement, or any internal knowledge of how their systems behave.
Counted figures in this case study describe its own 28 offers and 63sample records. They are not statements about Puff's trading.
Where to go next
