carsensor.net Used-Car Resale Value Scout

Searches carsensor.net (Recruit) used-car listings by freeword and returns each listing with full spec/price/dealer data, plus a bargain-rate score computed against carsensor's own official market-price matrix for that exact maker/model/year. Pay-per-event pricing.

Features

• Search carsensor.net used-car listings by freeword
• Get price, year, mileage, and repair history per listing
• Score listings against carsensor's own market-price data
• Find undervalued used cars with a bargain-rate score
• Get dealer contact info and vehicle spec details

⚠️ Disclaimer (please read)

• This Actor automatically retrieves information that is publicly available on carsensor.net. You are responsible for how you retrieve and use this data.
• The site owner's Terms of Use, robots.txt, or access blocking could change without notice, which could stop this Actor from working or change its output unexpectedly.
• valueScore is a heuristic estimate, not a guarantee. It compares a listing's price to a count-weighted average derived from carsensor's own aggregate market-price matrix for that maker/model/year (grade-level, not exact-trim-level — see "Valuation logic" below). It does not account for individual vehicle condition, equipment differences, or actual negotiated sale prices. Always verify the actual vehicle before making a purchase or resale decision.

Facts confirmed during pre-build research (as of August 10, 2026)

• Terms of Use: carsensor.net's general Terms of Use (「カーセンサー.net」ご利用規約) do not contain a site-wide "no commercial use" clause for the website itself — the one commercial-use restriction in the ToS (Article 6) is explicitly scoped to the smartphone app product only. However, a separate "リクルートWEBサービス利用規約" governs Recruit's official carsensor.net Web API (opt-in, requires account/API-key registration) and explicitly prohibits monetizing data obtained through that official API. This Actor does not use that official API — it scrapes the public HTML site — so that restriction does not apply to it, but it's worth knowing it exists if you ever consider the official API route instead.
• robots.txt: User-agent: *, no Crawl-delay. Disallows are almost entirely user-action/internal endpoints (/usedcar/search.php, mylist, favorite_*, inquiry*, etc.) — not the listing/detail pages this Actor uses. This Actor deliberately avoids /usedcar/search.php and instead reaches listings via /usedcar/freeword/ search, which is not disallowed.
• CDN / bot protection: Plain curl requests (including a bare curl/8.0 User-Agent, no browser emulation) to the top page and to /usedcar/search.php returned normal HTTP 200 with Server: Apache and no Cloudflare-style markers (no cf-ray, no JS challenge). Only verified with single, low-frequency requests — sustained high-volume behavior has not been tested.

Privacy / PII design

The dealer contact info shown on a listing page (shop name, address, phone number) is carsensor's own labeled business-level public information — the phone number is a 0078--prefixed call-tracking number tied to the listing, not a dealer employee's personal line. This is analogous to how our PR TIMES Actor treats a company's own labeled "Company Overview" block, and is not treated as personal contact information needing redaction.

That said, two fields are free text written by the dealer and could, in principle, contain personal contact info (e.g. "ご質問は担当◯◯まで" written directly into a listing's remarks) — and a small sole-proprietor shop's registered name can occasionally be an individual's name rather than a company name. This Actor applies a safety net (src/privacyFilter.js) to exactly those fields right before output:

• title (listing title)
• description (the dealer's free-text remark field)
• seller.name (dealer/shop name — fallback only, in case of a sole-proprietor shop)

Each is scanned for email-address / phone-number-like regex patterns and person/contact labels (e.g. "担当:"); any match nulls out that field, and the detection is recorded in the output's personalInfoRedactions field. seller.phone / seller.address / seller.url are intentionally excluded from this safety net, since they are carsensor's own labeled business contact info for the dealer, not an individual's direct line.

No listing with an actual personal-info match was found during pre-build spot checks (see the sample output below — personalInfoRedactions: []), but this is a safety net for cases the spot check didn't happen to cover, not a guarantee that the underlying pages never contain such text.

What does this Actor do?

• Searches carsensor.net's freeword search (/usedcar/freeword/{keyword}/) — the same as typing a term into the site's search box — and fetches each candidate listing's detail page.
• For each listing, extracts:
  • Price (車両本体価格 / 支払総額), correctly handling the "応談" (price on request) sentinel — when a listing has no fixed price, price.isNegotiable is true and both price fields are null instead of the raw placeholder value (999,999,999) carsensor uses internally.
  • Year (with Japanese era), mileage, repair/accident history (修復歴), and a spec block (body type, transmission, displacement, color, doors, seating, drivetrain, warranty, inspection status, one-owner, non-smoking).
  • Maker/model (with carsensor's internal codes, from the page's breadcrumb structured data) and dealer contact info.
• Looks up carsensor's own 市場価格(相場)マトリクス for that maker+model (its /usedcar/souba/ page) and computes a valueScore: how this listing's price compares to the market-price baseline for its specific model year.
• Turning on onlyWithValueScore limits output and billing to listings where a score was actually computed.

Who is this for?

Rather than a raw scrape of listing fields, this Actor's differentiation is turning carsensor's own publicly-available market-price matrix into an immediately usable bargain-rate number per listing — structured output an AI agent or a human resale-flipping decision-maker can act on directly, without having to separately open and cross-reference the 相場 page themselves (same stack as our BOOTH, SUUMO, and PR TIMES Actors: JavaScript ESM, Apify SDK, Pay-per-event pricing).

Two things worth being upfront about, decided during pre-build research (see project notes):

• This is not pitched on "exclusive data." carsensor has no per-vehicle third-party inspection/ appraisal data (no AIS/JAAA-style report), and both its 修復歴 (repair history) field and its price×year/mileage market matrix have close equivalents on goo-net and other major Japanese used-car sites. The value here is turning data that already exists into a ready-to-use score, not surfacing data nobody else has.
• No cross-checking against goo-net or other sources in this version. Combining carsensor with a goo-net-style source (for inventory overlap / price cross-validation) was considered as a differentiator but intentionally left out of v0.1 to keep scope manageable — see "Possible future extensions" below.

Input

Output

The following is actual output from a real run of this Actor against the live site (August 12, 2026, freeword "プリウス") — regenerated after the market-matrix parsing fix described under "Pricing" below, so these numbers reflect the corrected valuation logic.

{
  "sourceUrl": "https://www.carsensor.net/usedcar/detail/AU7231736290/index.html",
  "bkkn": "AU7231736290",
  "title": "トヨタ プリウスα G 5ニンノリ　バックカメラ ドラレコ前後 クルコン",
  "maker": { "code": "TO", "name": "トヨタ" },
  "model": { "code": "220", "name": "プリウスα" },
  "price": { "basePriceYen": 1100000, "totalPriceYen": 1223000, "isNegotiable": false },
  "year": 2016,
  "yearEra": "H28",
  "mileageManKm": 5.4,
  "repairHistory": "なし",
  "spec": {
    "bodyType": "ミニバン",
    "transmission": "その他AT",
    "displacement": "1800cc",
    "color": "パールマイカ",
    "doorCount": "5",
    "seatingCapacity": "5名",
    "driveType": "2WD",
    "warranty": "付き / 1年 / 距離無制限",
    "inspectionStatus": "車検整備付",
    "legalMaintenance": "付き",
    "oneOwner": "－",
    "nonSmokingCar": "－"
  },
  "region": "大阪府/大阪北部",
  "seller": {
    "name": "ネッツトヨタニューリー北大阪（株） 千里店",
    "url": "https://www.carsensor.net/shop/osaka/100644015/",
    "address": "大阪府吹田市山田西１－３６－１０",
    "phone": "0078-6002-061253",
    "businessHours": "10:00～18:00"
  },
  "description": "当店は駐車場完備！お気軽にマイカーでお越しくださいませ。ご購入後の走行距離は無制限。全ての中古車に1年間の無償保証がついています。",
  "image": "https://ccsrpcma.carsensor.net/CSphoto/bkkn/750/408/UT0053750408/UT0053750408_002L.JPG",
  "marketBaseline": {
    "soubaUrl": "https://www.carsensor.net/usedcar/souba/TO_S220/",
    "sampleSize": 78,
    "estimatedMarketPriceYen": 1384615,
    "method": "souba-price-x-year-matrix-weighted-average"
  },
  "valueScore": { "bargainRatePercent": 20.6, "verdict": "割安", "confidence": "normal" },
  "disclaimer": "carsensor.net上の公開情報を基に自動生成した参考データです。...",
  "scrapedAt": "2026-08-12T00:00:00.000Z",
  "personalInfoRedactions": []
}

An example of the negotiable-price case (verified against a real "応談" listing, AU5867522762): price comes back as

{ "basePriceYen": null, "totalPriceYen": null, "isNegotiable": true }

Valuation logic

1. Maker/model resolution (src/carsensorClient.js): a detail page's breadcrumb structured data (BreadcrumbList JSON-LD) contains an entry for the model-level page in the form /usedcar/b{makerCode}/s{modelCode}/index.html. This is parsed to get maker.code/model.code.
2. Market-price matrix lookup: carsensor's own 相場(souba) page for that maker+model is at /usedcar/souba/{makerCode}_S{modelCode}/ (verified directly, e.g. bLE/s012 → souba/LE_S012/). That page embeds three cross-tab matrices (price×year, price×mileage, year×mileage) as links whose query string carries the exact bucket boundaries (PMIN/PMAX/YMIN/YMAX) and count. This Actor uses the price×year matrix (identified via the fed=...pricemodelyear... marker in each link's query string, which is more robust than relying on DOM tab position).
3. Baseline estimate (src/valuation.js): for the listing's exact model year, all matching price buckets are pulled from the matrix and combined into a count-weighted average price — this is marketBaseline.estimatedMarketPriceYen, with sampleSize reported alongside it as a transparency/ confidence signal. Open-ended top/bottom buckets (e.g. "415万円以上") don't have a real upper/lower bound, so their midpoint is estimated with a fixed +15%/-15% heuristic adjustment off the known boundary — this is a rough approximation, not a statistically derived figure.
4. Bargain rate: (estimatedMarketPriceYen - basePriceYen) / estimatedMarketPriceYen * 100. Verdict is 割安 (undervalued) at +15% or more, 割高 (overvalued) at -15% or less, 相場並み in between. Fixed thresholds, not tuned against any ground truth.
5. Confidence: low when the matched year's sampleSize is under 5 listings (a thin market for that year makes the weighted average less trustworthy).

Known limitation: the market matrix is at the grade level (e.g. all Lexus CT trims combined), not matched to the listing's specific trim/equipment — so valueScore should be read as "vs. this model's market overall for that year," not "vs. this exact trim." Mileage is not currently factored into the baseline even though carsensor's souba page also exposes a price×mileage matrix — a future version could combine both matrices for a tighter estimate (see below).

Pricing

A scored listing costs $0.08 in total; an unscored listing (negotiable price, or no matching market data) costs $0.015.

How the price was set (measured on Apify, August 10, 2026)

This Actor was deployed (apify push) and run three times with production-like input (freeword search, 25-30 items each), and Apify's measured real infrastructure cost (usageTotalUsd, covering Compute Units, data transfer, and storage writes) was checked.

About 92-93% of the cost is ACTOR_COMPUTE_UNITS; the rest is negligible dataset-write/data-transfer cost. All three runs happened to compute a value score for every listing (both freewords are common models with abundant market data), so we don't have a real-world run isolating listing-extracted-only cost — but architecturally, value-score-computed's only marginal infra cost is the souba/market-price fetch, and that's cached per maker+model for the whole run (typically 2-3 distinct models per 25-30 listings), so its true marginal cost per scored item is a small fraction of the ~$0.00012/item blended figure above, not something separately measurable at this scale.

Note on this measurement: the first run surfaced a real parsing bug in the market-matrix logic (carsensor's souba page links include row/column "total" cells alongside real per-cell data, which were initially miscounted and produced a handful of implausible scores). This has been fixed and reverified — sample sizes now exactly match carsensor's own displayed totals as a sanity check. The pricing below is based on the two post-fix runs (2 and 3); run 1 is included in the cost table only for its infra-cost/ timing data, not its (superseded) computed values.

Following the same approach as our SUUMO Actor (which priced $0.08 against a measured infra cost of about $0.0004/item, roughly a 200x margin) and our PR TIMES Actor ($0.015 base / +$0.045 signal bonus):

• listing-extracted is set to $0.015 — the same tier as PR TIMES's release-classified, since both represent a comparably rich plain structured record (full spec/price/dealer data here vs. official classification + company profile there). Against our measured ~$0.00012/item, that's roughly a 125x margin, well above the 100x floor.
• value-score-computed is set to $0.065 (total $0.08 for a scored listing) — anchored to the same total figure as our SUUMO Actor's per-record price, on the same reasoning: this is the "one investment/resale decision" component, priced on the value of the decision it enables rather than its (near-zero incremental) infra cost. A used-car flip's typical margin is smaller than a typical real-estate deal, so pricing at parity with (not above) the SUUMO figure felt like the right calibration rather than pricing higher just because we could. Against measured cost, the full $0.08 scored-listing price is roughly a 667x margin — comfortably clearing the 100x floor by a wide margin, which is expected since this price is value-anchored, not cost-anchored.

Status: Pay-per-event pricing is enabled in Apify Console and verified live — a real run showed pricingInfo.pricingModel: "PAY_PER_EVENT", chargedEventCounts matching the run exactly (no missed or double charges), and no automatic/implicit charge events beyond the three declared above.

Load management

There's a default 1500ms delay between every HTTP request this Actor makes (search result pages, listing detail pages, and souba/market-price pages all share one pacing clock via carsensorClient.js's politeGetHtml). Configurable via requestDelayMs, but it cannot be set below 1000ms for safety. Souba/market-price pages are fetched at most once per distinct maker+model per run (the soubaCache Map in src/main.js) — a freeword search tends to return many listings of the same model, so this meaningfully cuts down on repeat requests.

FAQ

What if carsensor changes its page layout? The selectors and URL patterns in src/carsensorClient.js were verified against real carsensor.net response HTML (freeword search pages, listing detail pages, souba/market-price pages) as of August 10, 2026. carsensor's markup may change over time; if this stops working, run npm run test-connectivity to inspect the actual response and adjust accordingly.

Why does a search only return a few pages of results? Freeword search covers only maxSearchPages pages (default 5) of results per run — there's no "full sync" mode in this version.

Why is valueScore sometimes missing? It can't be computed for listings priced "応談" (negotiable — no fixed price to compare), for maker/models whose breadcrumb data doesn't resolve to a b{code}/s{code} pair, or where the souba page has no matching data for that model year.

Does the bargain score account for mileage or exact trim/grade? Not currently. The market matrix is grade-level, not trim-level, and mileage isn't factored into the baseline yet (see "Valuation logic" above).

Possible future extensions (not implemented in v0.1)

• Cross-checking against a goo-net-style source: combining carsensor's listings with another aggregator's for inventory overlap detection and independent price cross-validation was considered as a differentiator during pre-build research, but intentionally deferred to keep this version's scope manageable.
• Mileage-adjusted baseline: combining the price×year and price×mileage (and year×mileage) matrices for a tighter market-price estimate than the price×year-only baseline used today.
• Region-adjusted baseline: carsensor's souba page also supports a regional filter; this Actor doesn't currently use it, so the baseline is national, not local to the listing's prefecture.