THE product endpoint (unified 2026-07-23). One call: current snapshot (price, BSR, rating, buy-box — cache-or-live via mode + max_age_days) plus observed_at, the FULL price-change timeline (price_history, seller-stamped, back to first observation) with a 30-day convenience view, the full in/out-of-stock timeline (stock_history, with stock_state + stock_observed_since), the Amazon-reported bought-past-month badge (bought_past_month int + raw text — measured, not modeled; Amazon marketplaces only), the current all-seller snapshot, and brand stats. country=walmart returns the intelligence blocks (numeric item ID). Partial live scrapes are never billed. Walmart products also carry (2026-09-20, additive) image_url / images[], description, category_path + breadcrumbs[], highlights[] (name/value spec pairs as shown on the PDP), warnings[] (compliance text, e.g. Prop 65), upc, model_number, color, rating / review_count / rating_observed_at (null when the page showed none), was_price / discount_pct, additional_offer_count, per-seller shipping_cost (a cost, 0 = free; never a landed price), and in_stock / as_of beside the unchanged stock_state / observed_at. gtin= resolves Walmart barcodes to item IDs. LIVE Walmart (2026-10-18): country=walmart&mode=live reads the walmart.com product page right now (typically 8-15 s) - price, stock, rating, the featured seller and the sellers the page lists, plus the PDP content above - laid over our stored history for the item (history_available says whether we held any). 2 credits; a partial page (title, no price, in stock), an item walmart.com does not have (404) and a failed or busy fetch (503) are never billed. mode=auto on Walmart works like Amazon: the stored record when observed within max_age_days (0.5), else a live read (2); mode=cache never fetches. A failed live read on an item we hold answers the stored record as source=cache_after_live_fail (0.5) with a note. One product page per call: the cheapest other seller Walmart shows only on a second page is not fetched (additional_offer_count says how many there are). location names the store walmart.com priced the page for (postal_code, city, state, store_id): each read can land on another store, and fresh grocery is priced per store. buybox_price_points_hot (2026-10-14): every all-offers-page observation of the last 30 days {ts, price, seller, seller_id, fulfillment, offer_count}, newest last — sub-daily for hot-lane ASINs (observed every few hours), the walk's and on-demand observations for the rest; null when none. tracking: the hot lane's state for the ASIN {hot, interval_min, rate, source, last_fetched_at, next_due_at}, null when the ASIN is not in the lane. Both on every Amazon answer, included in the record; history_only carries buybox_price_points_hot. The cache snapshot's as_of and the fresh flag behind mode=auto follow the latest observation we hold (not only the nightly summary), so a hot-lane fetch makes the record fresh. buybox_state (2026-10-14, §86): what the lane's newest all-offers read saw — in_stock | no_buybox (offers, but no featured offer) | out_of_stock (no offers) — with buybox_state_observed_at; when that read is newer than the daily history's last point it decides stock_state / in_stock and stock_state_source names it. stock_state, in_stock and the tri-state out_of_stock always agree (2026-10-15). buybox_condition (2026-10-15): new | used | null (no featured offer) with buybox_condition_source (product_page | offers_page) and buybox_condition_detail (the pinned offer's condition text, used_only_buybox, or the used storefront's name — Amazon Resale / Warehouse / Retourenkauf …, which win the buy box like any seller and read as Amazon retail before): live = measured on the fetch; cache = the pinned offer's stored condition at the row's observation, else the storefront name. On LIVE answers variation_status (page_is_sibling | page_is_other | null) + page_asin say when Amazon served another ASIN's page for the one asked (cache answers carry both as null); the stock keys describe the ASIN asked for, the content fields the page served.

PDP content (2026-08-09, Amazon marketplaces): description + description_source (product_description = prose, feature_bullets = "About this item", aplus = brand marketing — bullets read very differently from prose), the full image gallery (images[], ≤20 originals), feature_bullets[], breadcrumbs[] (full category path), the variations matrix (variations[] with child_asin + dimension_values + availability, variation_count, parent_asin — an EMPTY variations is normal, most ASINs have no variants; per-variant PRICE is NOT included, Amazon only sends it on selection), featured reviews (reviews[], ≤50 — the accumulated set Amazon shows a logged-out visitor, NOT the full review history; the count varies per fetch because Amazon does not always render the section; date_raw is verbatim and carries the review country + a localised month name — parse downstream, keep the country), and a tri-state out_of_stock (true = out of stock: Amazon's dedicated out-of-stock buybox on the page, or the newest offers read found no offer - it always agrees with stock_state; false = in stock, null = unknown). bought_past_month null means NO BADGE — Amazon only badges higher-velocity listings — never "we missed it". content_observed_at null with empty arrays means content has not been captured for this ASIN yet. The "Customers say" AI summary/aspect chips are NOT available (lazy-loaded after our snapshot). No extra charge for any content field.

2026-09-27 (additive, Amazon marketplaces, no extra charge): dimensions (item weight + package dimensions as printed plus parsed value/unit and metric weight_g / *_cm), deal_history (list price, savings and coupon/promotion text as segments of consecutive observations, with parsed coupon_amount / coupon_percent and a current view), bought_past_month_history (the badge as segments; null = page observed without a badge), rating_history (one point per observed day: rating_count + rating, with ratings_gained_30d / 90d), and variation_coverage (the variation matrix as last observed with, per child, whether we track it and its latest 30-day price — current state, not history). With offer_history=true every point also carries the offer's shipping (cost), fba and prime for that day, and each series row fba_days / fba_last / shipping_last (null on legacy-era days before 2026-04-17). Explicit null on Walmart. Also since 2026-09-27: sales_estimate is null whenever the bought-past-month badge is present (the badge is the measured figure; the estimate is served only for unbadged listings), and reviews[] lists the marketplace's own reviews (is_local true) before the other-countries set.