All notable changes to the Zid Store Scraper are recorded here.
The record shape, the error codes, and the limits quoted in the README are a published contract.
Adding, renaming, or restatusing any of them is a breaking change and must reach the README, the
dataset schema, and this file together.
-
Listing assets. assets/output-preview.png shows real rows from four live Zid stores —
Arabic names, list price beside sale price, live stock, SKUs, option counts, and the one product
page that timed out, shown as the failed row the dataset actually carries rather than edited
out. It is a browser screenshot, because Arabic is cursive and needs real shaping; drawing the
glyphs directly would produce exactly the mojibake this Actor claims to avoid.
scripts/capture_output_rows.py and scripts/render_output_preview.py rebuild it from a live
run, and scripts/publish_readme_assets.py uploads it to a public key-value store — Apify
publishes a README's text and not the files beside it, so a repository path renders broken on
the listing.
-
Store icon, drawn from code by scripts/prepare_brand_assets.py so a redraw is a reviewable
diff. Same geometry as the Salla Actor's mark and a violet price tick against its emerald: one
product with two front doors should read as siblings in a Store grid, and colour is the only
difference that survives 32px.
-
language does something. It was accepted and then ignored entirely — an input that
changed nothing. This Actor reads whatever the storefront serves, so it now does what the Salla
Actor does: reads the languages the shop declares about itself (<html lang> and hreflang
alternates) and puts a LANGUAGE_UNAVAILABLE warning on the store row when the requested one is
not among them. It never translates and never switches the shop. A store that declares no
language is left alone — silence is not a denial, and warning on it would fire everywhere and
mean nothing.
-
maxItems now applies per store, matching the Salla Actor. It was a shared run budget, and
the first platform run showed what that cost: given four stores and maxItems: 4, the first
store took all four and the other three published nothing while reporting failed. Nothing had
failed — the budget was gone. A store truncated by the limit now says so with an
ITEM_LIMIT_REACHED warning, the same code and wording Salla uses.
-
Products already published by an earlier container of a resurrected run count against that
store's limit. Counting only new rows would let a run that died after three products publish a
full allowance again, billing the customer for more than they asked for.
-
First implementation. Reads public Zid storefronts and emits two record types:
product-scraped (one row per product) and store-analyzed (one row per store per run).
-
Custom-domain store detection. About 59% of Saudi Zid stores run on the merchant's own
.com domain, so detection reads page content — several independent platform markers — rather
than the hostname. A page that merely mentions Zid is refused with NOT_A_ZID_STORE.
-
List price and sale price are reported separately. The storefront's embedded product object
carries both; its JSON-LD carries only the effective price, so JSON-LD is the fallback and never
the source of price.
-
Availability is reconciled, never guessed. The storefront's own state and its structured
data are read independently. When they disagree, the row says unknown and carries a warning.
-
Sitemap-first discovery with a category-crawl fallback, chosen on the response body: one
measured store answers sitemap_products.xml with HTTP 400 and an HTML error page.
-
Arabic throughout — product names, categories, and variant options — with sitemap IRIs
percent-encoded before fetching, and a CSV export that opens in Excel with a UTF-8 BOM.
-
scripts/probe_live_store.py, the measuring instrument for the published field-accuracy
scorecard, and scripts/scorecard_stores.txt, the store list it was run against.
-
Published field-accuracy scorecard, measured on 38 live Zid stores and 152 products: name,
price, currency, availability, SKU, images and variants at 100%, categories at 96.7%. 28 of the
38 stores run on custom domains and every one was detected. Numbers and method in the README;
full account in docs/ACTOR_6_MARKET_EVIDENCE.md §8.
