HobbyLink Japan (HLJ) Gunpla & Figures — Price + Availability Stats in One Call

What does this kit go for in Japan, and can you actually get it? One call per keyword answers both — because for hobby goods the price alone is useless. A ¥4,500 Master Grade that is discontinued and a ¥4,500 Master Grade that is in stock now are not the same product to anyone who resells.

This Actor reads HobbyLink Japan (HLJ — the Shizuoka hobby exporter that has shipped Gunpla, plastic model kits, figures, paints and tools worldwide since 1988) and returns one compact market record per keyword:

• priceJpy — min / Q1 / median / Q3 / max of the sampled listings, in yen, plus the same in USD at the current rate
• stockBreakdown — counts by HLJ's own state code: instock / preorder / futurerelease / backorder / orderstop / discontinued. This is the field that decides whether a keyword is a sourcing opportunity or a dead catalogue entry
• inStockRatioSampled and orderableRatioSampled — two different numbers on purpose, see the note below
• onSaleRatioSampled and saleRatePct — how much of the keyword is discounted, and by how much when it is
• releaseYearTop — the release-date spread, so you can see whether a keyword is current stock, a 2026 preorder wave or a back-catalogue tail
• Optionally every sampled item (title, price, list price, savings, discount %, stock state, release date, URL)

No login, no API key, no browser, nothing stored between runs.

Pricing — from $0.02 per keyword, no subscription

Event	Price	When
Keyword market summary	$0.02	Per keyword analyzed
Individual item	$0.002	Only if you enable Include individual listings

A default run (1 keyword, summary only) costs $0.02. A keyword that returns zero listings is never charged.

Input

Field	Example	Notes
keywords	["RG Nu Gundam"]	HLJ's catalogue is indexed in English — product and line names work best. Each keyword costs $0.02
pagesPerKeyword	2	1–8. One page = 24 items = 1 page request + 1 batch price request. Default 2 (48 items)
includeIndividualItems	false	Enable to also get each sampled item as its own record (+$0.002 each)
convertToUsd	true	Adds USD stats at the current exchange rate. A failed lookup never fails the run

What basis is each number on?

Every record carries priceJpyBasis, so you never have to guess:

priceJpyBasis	What it means	When you get it
exact	The pages read covered the whole result set — these are true statistics over every matching listing	Narrow keywords (hits ≤ pagesPerKeyword × 24)
sample	Ordinary sample statistics over the listings actually read, count = how many of them had a price	Everything else

There is deliberately no third option. HLJ's search returns a relevance order and offers no price sort, so there is no rank whose value is a population quantile. Sibling Actors of ours (bookoff-market-checker, digimart-instrument-market-checker) do read population quartiles — they can, because those platforms expose a monotonic price-ascending sort. HLJ does not, and this Actor will not invent a population claim it cannot support. What you get is an honest sample of the first N relevance-ranked hits, clearly labelled, with totalListingsFound next to it so you can see the sampling fraction yourself.

Everything else in the record (stockBreakdown, the ratios, saleRatePct, releaseYearTop) is likewise sample-based over sampledListings.

inStockRatioSampled vs orderableRatioSampled — read this one

HLJ's price feed carries an is_in_stock flag that is true for backordered items too — it means "you may place the order", not "it is on the shelf". Taken at face value it reports 100% availability on a keyword that is 38% backorder. So this Actor splits it:

• inStockRatioSampled — share of sampled items whose HLJ state code is literally instock. This is the one you want.
• orderableRatioSampled — share for which HLJ will accept an order at all (in-stock plus backorder/preorder). Reported separately, under a name that says what it is.

Output example (type: "market_summary")

{
  "type": "market_summary",
  "keyword": "RG Nu Gundam",
  "totalListingsFound": 273,
  "sampledListings": 24,
  "priceJpy": { "min": 500, "q1": 500, "median": 500, "q3": 4125, "max": 60000, "count": 24 },
  "priceJpyBasis": "sample",
  "pricedListings": 24,
  "stockBreakdown": { "instock": 15, "backorder": 9 },
  "inStockRatioSampled": 0.625,
  "orderableRatioSampled": 1.0,
  "onSaleRatioSampled": 0.0,
  "saleRatePct": null,
  "releaseYearTop": [[2025, 12], [2021, 7], [2026, 4], [2023, 1]],
  "checkedAt": "2026-08-01T05:33:51.318685+00:00",
  "sourceUrl": "https://www.hlj.com/search/?Word=RG+Nu+Gundam&page=1",
  "priceUsd": { "min": 3.15, "q1": 3.15, "median": 3.15, "q3": 26.02, "max": 378.48 },
  "exchangeRateJpyUsd": 0.006308
}

That ¥500 median is real and it is instructive: a RG Nu Gundam search matches the ¥500 water-slide decal sheets and detail parts for the kit as well as the ¥4,500 kit itself, and relevance order puts them near the top. Search a full product name (RG 1/144 Nu Gundam) when you want the kit's own price, and read q3/max alongside the median on any accessory-heavy keyword. The Actor reports what the search returned rather than quietly filtering to make the number look nicer.

What this Actor does not do

• No population quantiles. HLJ exposes no price sort, so no rank can be claimed as a percentile of the whole result set — see the basis table
• No sold prices. HLJ is a shop: these are asking prices, the yen price a buyer would pay today, not completed-sale comps
• No item dumps by default. The product is the statistic; individual items are opt-in and separately priced
• No login-only data. Everything comes from public search pages and HLJ's own public price endpoint
• No stored dataset. Every run fetches live; nothing is retained between runs
• No customer or seller PII. HLJ is a single retailer — there are no third-party sellers to profile

Notes on the data

• Prices are not in the search HTML. HLJ renders the result cards server-side but leaves the price span empty and fills it in with JavaScript. This Actor therefore pairs the card list with HLJ's own batch price endpoint (/search/livePrice/?item_codes=…, up to 40 codes per call) instead of running a browser — which is why it fits in 256 MB and finishes in seconds. A generic HTML scraper pointed at these pages returns titles with blank prices
• Some items genuinely have no price (unannounced or delisted products). They are still counted in sampledListings and in stockBreakdown; pricedListings tells you how many made it into the price statistics. They are never counted as ¥0
• priceJpy is the sell price — what a buyer pays today. When an item is discounted, the pre-discount price is available per item as listPriceJpy, with savingsJpy and saleRatePct
• saleRatePct is the median over discounted items only. Including the large majority of the catalogue that is not on sale would drag it to 0 and hide the only number a bargain hunter wants. null means nothing in the sample was on sale
• releaseYear is parsed from HLJ's release_date (e.g. "July 2026" → 2026). Future years are normal and meaningful here: preorders for kits not yet released are a large part of this catalogue
• Stock state codes are HLJ's own (instock, preorder, futurerelease, backorder, orderstop, discontinued, …). Codes we do not recognise are passed through in stockBreakdown rather than dropped, so a new HLJ state never silently disappears from the split. Items HLJ's price feed did not answer for appear as unpriced_or_unknown
• Requests are throttled to ≥1.2 s and the run carries a wall-clock budget; if it is reached, later keywords are sampled over fewer pages and their records carry truncatedForTimeLimit: true