Tackleberry Japan Used Tackle Market — Used vs New Population Split

How much Japanese fishing tackle is actually out there second-hand, and at what price? One call per keyword answers it with population counts reported by the shop itself — not an estimate from the 60 listings a page happens to render.

This Actor reads ＠ベリーネット (b-net.tackleberry.co.jp), the online arm of タックルベリー — Japan's 200-plus-store used fishing-tackle chain, ~122,000 items in stock, no login — and returns one compact market record per keyword:

• bySaleType — the exact number of listings in each of the shop's three mutually exclusive stock classes. Measured for シマノ on 2026-08-02: 34,397 listings = 33,891 used + 454 new + 52 outlet. Those three add up to the unfiltered total to the item, and the Actor asserts it on every run (saleTypeSplitConsistent)
• priceJpy per class — for used gear the quartiles are read out of the population (¥2,311 / ¥9,240 / ¥16,748 / ¥30,204 / ¥277,200), not averaged off page one
• usedPriceCdf — a population price histogram built from the shop's own price_max counts: 3,047 used listings under ¥5,000, 9,024 under ¥10,000, 19,170 under ¥20,000, 30,951 under ¥50,000
• usedVsNew.medianRatio — the used median against the new median, published only when both are population figures
• sampleStats.conditionRankMix — the A / AB / B+ / B / B- / C grade spread of the sampled used gear
• Optionally every sampled listing (title, price, list price, discount, stock class, condition rank, URL)

The differentiator is the stock split with real denominators. Anyone can list fishing tackle; nobody on the Store can tell a JDM reseller that a brand has 33,891 used listings against 454 new ones, or that a quarter of the used stock sits under ¥10,000 — the numbers you need before deciding whether a model is worth sourcing for eBay or Mercari.

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

Note on the domain. www.tackleberry.co.jp is the chain's store locator and carries no products. The shop is b-net.tackleberry.co.jp, which is what this Actor reads.

Pricing — from $0.02 per keyword, no subscription

Event	Price	When
Keyword market summary (stats-computed)	$0.02	Per keyword analyzed
Individual listing (listing-returned)	$0.002	Only if you enable Include individual listings

A default run (1 keyword, summary only) costs $0.02. A keyword that returns nothing is never charged.

Input

Field	Example	Notes
keywords	["シマノ"]	Brands or model numbers. Japanese works best — this is a Japanese shop. Each costs $0.02
saleTypes	[]	Which stock classes to count. One request each. Empty = all three
sortBy	"price_asc"	price_asc / price_desc / newest / rank. Keep price_asc for population quantiles
priceLadder	[]	price_max rungs for the used CDF. One request each, max 8. Empty = 5000/10000/20000/50000
maxItemsPerKeyword	120	Listings kept for the sampled blocks. Does not affect the counts or the quantiles
includeIndividualItems	false	Also emit each sampled listing (+$0.002 each)
convertToUsd	true	Adds USD stats at the current exchange rate

The one thing to understand: every number says where it came from

This Actor never lets a population figure and a page sample blur together. Each block carries its own basis:

basis	Meaning	Where it is used
population	The shop's own hit count for a filter	saleTypeCountsBasis — the used / new / outlet counts
population_quantiles	The actual listed price at rank ceil(N·q), read by jumping to the page that holds it	Used prices on a price_asc run
population_median	Min and median jumped out of the population; the IQR deliberately not computed	Classes over one page that do not justify four requests (the new class)
population_bands	Interpolated inside band counts that are themselves exact	usedPriceCdf, and the used-price fallback
exact	The whole class fit on the pages fetched, so the statistic is the population	The 52 outlet listings above
sample	Describes only the cards that were parsed	sampleStats (rank mix, discounts, new-arrival share)

Why sort-and-jump rather than the count ladder — both were verified

Both population techniques work on this shop, which is unusual, so the choice is deliberate:

• The price-ascending sort (orderby=5) is monotonic across pages. Verified live on 2026-08-02 for the 33,891 used シマノ listings: pages 142 / 283 / 424 returned ¥9,240 / ¥16,748 / ¥30,204, and price_max=2000 returned 0 listings, confirming that page 1's head (¥2,311) really is the population minimum.
• The price_max ladder returns exact counts, which makes a population histogram.

The rank-jump is used for the headline quartiles because it returns prices that actually exist in the inventory — an item you can go and look at — whereas the ladder interpolates inside a band. The gap is real and measurable: on the same population the ladder puts the median at ¥17,808 against the rank-jump's measured ¥16,748, i.e. 6% high, because it has to assume prices are spread evenly inside the ¥10k–20k band and they are not.

The ladder is not wasted, though. It is published in its own right as usedPriceCdf (the shape of the distribution is a different product from its quartiles), and it is the fallback: if the monotonicity guard ever trips — a page jump the shop clamps, a probe whose values are not non-decreasing — the used prices fall back to population_bands, and only if that fails too, to sample. The claim always shrinks to what was actually measured.

Output example (type: "market_summary", real measured output)

{
  "type": "market_summary",
  "keyword": "シマノ",
  "sortUsed": "price_asc",
  "sortCode": "5",
  "totalFound": 34397,
  "sampledListings": 120,
  "saleTypeCountsBasis": "population",
  "bySaleType": [
    { "code": "1", "key": "used",   "expectedLabel": "#中古品",      "labelSeen": "#中古品",      "labelMatchesCode": true,
      "count": 33891, "share": 0.9853, "sampledListings": 60,
      "priceJpy": { "min": 2311, "p25": 9240, "median": 16748, "p75": 30204, "max": 277200, "count": 33891 },
      "priceJpyBasis": "population_quantiles" },
    { "code": "2", "key": "new",    "expectedLabel": "#新品",        "labelSeen": "#新品",        "labelMatchesCode": true,
      "count": 454, "share": 0.0132, "sampledListings": 60,
      "priceJpy": { "min": 484, "p25": null, "median": 4488, "p75": null, "max": null, "count": 454 },
      "priceJpyBasis": "population_median" },
    { "code": "3", "key": "outlet", "expectedLabel": "#アウトレット", "labelSeen": "#アウトレット", "labelMatchesCode": true,
      "count": 52, "share": 0.0015, "sampledListings": 52,
      "priceJpy": { "min": 330, "p25": 880, "median": 1021, "p75": 1741, "max": 51590, "count": 52 },
      "priceJpyBasis": "exact" }
  ],
  "saleTypeSplitSum": 34397,
  "saleTypeSplitComplete": true,
  "saleTypeSplitConsistent": true,
  "usedPriceCdf": {
    "basis": "population_bands",
    "saleType": "used",
    "rungs": [
      { "upperBound": 5000,  "count": 3047,  "cumulativeShare": 0.0899 },
      { "upperBound": 10000, "count": 9024,  "cumulativeShare": 0.2663 },
      { "upperBound": 20000, "count": 19170, "cumulativeShare": 0.5656 },
      { "upperBound": 50000, "count": 30951, "cumulativeShare": 0.9133 }
    ],
    "bands": [
      { "from": 0,     "to": 5000,  "count": 3047 },
      { "from": 5000,  "to": 10000, "count": 5977 },
      { "from": 10000, "to": 20000, "count": 10146 },
      { "from": 20000, "to": 50000, "count": 11781 },
      { "from": 50000, "to": null,  "count": 2940 }
    ]
  },
  "usedVsNew": {
    "medianRatio": 3.7317,
    "basis": "population_quantiles/population_median",
    "note": "... catalogue-mix ratio, not a residual-value ratio ..."
  },
  "sampleStats": {
    "basis": "sample",
    "sampledListings": 120,
    "usedSampled": 120,
    "conditionRankMix": {
      "sampleSize": 120,
      "ranks": [
        { "code": "a",  "label": "A",  "count": 4,  "share": 0.0333 },
        { "code": "ab", "label": "AB", "count": 7,  "share": 0.0583 },
        { "code": "bp", "label": "B+", "count": 25, "share": 0.2083 },
        { "code": "b",  "label": "B",  "count": 34, "share": 0.2833 },
        { "code": "bm", "label": "B-", "count": 45, "share": 0.3750 },
        { "code": "c",  "label": "C",  "count": 5,  "share": 0.0417 }
      ]
    },
    "discountVsRegularPct": { "count": 0, "median": null },
    "newArrivalShare": 0.0333
  },
  "requestsMade": 13,
  "checkedAt": "2026-08-01T15:56:34.141774+00:00",
  "sourceUrl": "https://b-net.tackleberry.co.jp/products/list?name=%E3%82%B7%E3%83%9E%E3%83%8E&disp_number=60&orderby=5",
  "usedPriceUsd": { "min": 14.58, "p25": 58.29, "median": 105.65, "p75": 190.53, "max": 1748.58 },
  "exchangeRateJpyUsd": 0.006308
}

What this Actor does not do

• No sold prices. These are current asking prices on a retailer's shelf. There is no transaction history on this shop and none is invented
• No PII. The shop is a company selling its own stock — there are no sellers, no accounts, no contact details. Nothing of that kind is collected or emitted
• No maker[] filter. It returns 1.4 MB in ~11 seconds; name= covers the same ground five times faster, so the Actor never sends it
• No listing dumps by default. The product is the statistic; individual listings are opt-in and separately priced
• No stored dataset. Every run fetches live; nothing is retained between runs
• No browser, no proxy. Plain HTTP, 256 MB

Notes on the data

• usedVsNew.medianRatio is a catalogue-mix ratio on a broad keyword. The measurement above (3.73) does not mean used tackle costs 3.7× new tackle — it means that this shop's 454 new items are mostly accessories while its 33,891 used items include rods and reels. Pin the keyword to a single model and the same number becomes a genuine residual-value ratio. The note field in every record says so
• Stock moves by the minute. The unfiltered count read 34,401 / 34,399 / 34,397 in the space of a day, which is why every record carries checkedAt and why saleTypeSplitConsistent tolerates nothing but reports the gap rather than hiding it
• labelMatchesCode is a tripwire, not decoration. sale_type[]=1 means "used" today; if the shop ever re-numbers its own filter, every statistic here would silently invert. The Actor therefore carries the shop's own #中古品 / #新品 / #アウトレット card tag next to the code it asked for and flags any disagreement
• The condition grade lives in an icon filename (com-rank-bm-square.svg = B-). There is no rank filter, so the grade mix can only ever be a sample — labelled as such — even though everything else here is a population figure
• Used gear carries no list price, so discountVsRegularPct is usually empty on a used-heavy keyword. Discounts are an outlet/new phenomenon on this shop, and reporting count: 0 is the honest version of that
• sortBy is always sent explicitly. The shop's own default order is 新着順 (newest), which would skew every price statistic; the sort actually used is echoed back as sortUsed
• Run it without a proxy. Measured 2026-08-02 with the internal dc-ip-probe (runs X1kho2Wzj37b74t0w, qBpVfzQB4pwpjjrM5, 96imjNnaeywlMsdSd): plain datacenter IPs get 200 on every combination of filter, sort and deep pagination. No anti-bot vendor is in front of it
• The shop is slow — 1.5–4 s per request typically, 4–12 s from a datacenter IP, with occasional 20 s+ outliers. A default run is 13 requests and took 68 s end to end when this README was measured. A soft 85-second budget protects the 120-second target: if it is reached, the price ladder is cut from the tail first, the affected statistic drops to a weaker (still labelled) basis, and the record carries truncatedForTimeLimit: true. Run fewer keywords per call rather than raising the budget