# Mart402 - Terms of Service & Privacy (v1.4.1, 2026-09-27)

## 1. Service
Mart402 provides machine-callable APIs (extraction, document parsing, verification,
structured extraction, Japan statutory calculations, image analysis). Payment is per call via x402 (USDC on Base). Quotes are free.
Since 2026-09-27 (v1.4) a payment is verified before the call runs and settled
on-chain only after the call succeeds: a call that does not succeed is never
charged (its response says billed_usd 0). If settlement fails after a successful
call (for example the balance moved), the result is held for one hour and the
payment terms are re-issued; repeating the same request with a fresh payment
delivers the held result without re-execution. A wallet whose payments repeatedly
verify but do not settle, or whose LLM-backed calls repeatedly fail on unusable
input, is refused for 24 hours (payer_temporarily_blocked). Prices shown in the
402 challenge are final for that call; cached results are discounted before
payment, never after.
Receipts: every paid response carries an X-Receipt-Id; GET /v1/receipts/{id}
re-fetches the stored result free for 30 days from settlement (payments are never
lost: a lost connection after settlement is recovered by re-fetching or by resending
the same payment header, which is idempotent). Since 2026-09-27 every 402 carries a
signed offer and every settled response a signed receipt (x402 offer-receipt
extension, key at /.well-known/did.json); the offer's validUntil (5 minutes) is a
declaration of how long the quoted terms are held, not a payment deadline - a valid
payment presented later is still accepted if the price is unchanged.

## 2. Caller responsibilities
- You are responsible for the rights to any URL or document you submit, and for the
  lawful use of results in your jurisdiction.
- robots.txt of target sites is respected by default. Requests refused by policy
  (robots, blocklist) are not charged.
- Prohibited: unlawful content, harassment, rights-infringing bulk copying,
  attempts to re-identify individuals, and any use violating applicable law.

## 3. Data handling (Privacy)
- We log call metadata (endpoint, timestamps, payment status, user-agent, target URL
  or document hash) for billing and quality.
- Extraction results are cached up to the TTL you request (default 24h). Uploaded /
  fetched documents for parsing are retained up to 1 hour (quote validity), then deleted.
- Sub-processors: Hetzner (hosting; primary in the US (Ashburn), warm standby in
  Germany - receipts are replicated between the two), Cloudflare (network proxy and
  TLS termination for all public hostnames), Coinbase Developer Platform (payment
  settlement), Mistral AI (LLM/vision processing for structured/jp-company, OCR,
  hard-document parsing and the vision products; paid tier - inputs are not used for
  model training by the provider), Google Cloud Vision (second OCR engine for
  ocr/parse consensus, and face-region detection for portrait-analysis/style products;
  no identification). Sub-processor list updated in v1.3 (2026-09-26): Google Cloud
  Vision and Cloudflare were in use but not listed before.
- Image inputs (ocr, meal-vision, portrait-analysis) are processed in memory and are
  not stored after the response is produced; only the response JSON is retained for
  receipt re-fetch. portrait-analysis reports tone attributes only: it performs no
  face recognition or identification and creates no biometric template. Submit only
  photos you have the right to use; for photos of a person, their consent is your
  responsibility (v1.2, 2026-08-09).
- style-report returns a PDF whose draping boards embed the submitted photo; that
  PDF is part of the response retained for receipt re-fetch (v1.3).
- Japan statutory inputs (salary, payslip, crypto transactions and similar) are
  computed in memory and not written to our call logs; as with every product, the
  response JSON (which echoes the inputs it used) is retained for receipt re-fetch.
  Results are calculations under the cited rules_version, not tax, legal or
  investment advice (v1.3).
- We do not sell or share customer inputs. Aggregate, non-identifying statistics may
  be published (e.g., service quality metrics).

## 4. No warranty / liability
Results are provided as-is. Liability is capped at the amount paid for the call in
question. On-chain payments are final; verified service errors are remedied by
re-execution or credit at our discretion.

## 5. Contact & changes
Operator contact: at.tanaka77@gmail.com. Removal requests from content owners are
honored via blocklist. Terms may change; the version at /terms applies at call time.

## 6. Data sources & attribution
Company-registry grounding uses the GLEIF LEI API and the Japan National Tax
Agency (NTA) corporate number system Web-API. Content derived from NTA data is
not certified or guaranteed by the NTA. Required Japanese attribution (NTA
Web-API terms of use):
このサービスは、国税庁法人番号システムWeb-API機能を利用して取得した情報をもとに作成しているが、サービスの内容は国税庁によって保証されたものではない。