CHANGELOG.md

Changelog

All notable changes to AI Price Radar are documented in this file.

[3.7.51] - 2026-09-07

Added

• Added POST /api/v1/admin/source-candidates/cleanup endpoint to purge invalid candidates (no_match, validation_failed, disabled, rejected) while strictly protecting promoted candidates and source intakes.
• Added "清理无效候选" (Batch Cleanup Invalid Candidates) button with confirmation prompt and execution feedback in the admin Source Discovery panel.

[3.7.50] - 2026-09-07

Added

• Added Business Cooperation & Advertising page (/advertise) and official contact email (info@ai.pricememo.cn).
• Added backend SystemSetting key-value model and admin API endpoints (GET /api/v1/admin/settings, PATCH /api/v1/admin/settings).
• Added admin panel toggle switch for advertise_enabled to dynamically control public visibility of commercial cooperation links and gate /advertise.

[3.7.49] - 2026-09-05

Fixed

• Kept Dujiao-Next discovery disabled in complete remote refreshes unless explicitly enabled.

[3.7.48] - 2026-09-05

Fixed

• Made the production Importer image use the configured Python package mirror so dependency installation is reproducible during deployment.

[3.7.47] - 2026-09-05

Fixed

• Preserve failed crawler discovery batches for a retry instead of silently dropping candidates.

[3.7.46] - 2026-09-05

Fixed

• Hardened admin and worker authentication, email delivery, source-intake state transitions, and notification retries.
• Prevented hidden, stale, unpublished, disabled-platform, non-finite-price, and cross-product data from leaking into public catalog and history responses.
• Tightened classifier, discovery, JSON-LD/XML parsing, crawler budgets, and pipeline input validation without re-enabling disabled Dujiao-Next publication.
• Fixed Web API error handling, watchlist bounds, public-source URL checks, mobile navigation, and visible focus states.
• Guarded PostgreSQL restore database names against identifier truncation.

[3.7.41] - 2026-09-05

Fixed

• Accurate Product Classification for Free Accounts and Tools:
  • Cleanly stripped exclusion phrases (除Codex, 不可codex, 除Plus, etc.) in brand and tier detection, ensuring titles like (可网页反代，除Codex) are not misidentified as Codex/Plus.
  • Expanded CHATGPT_FREE_MARKERS to recognize g free, g-free, gfree, codex free, outlook/icloud/gmail free, free账密, free成品, free底号, 可升级plus, 开plus专用, 好底号, 未绑卡, and correctly route them to chatgpt-account instead of chatgpt-plus.
  • Prevented session relay indicators (反代专用) from overriding free account markers.
  • Added payment link tools and verification CDK markers (提炼, 代提链, 直卡支付链接, 支付链接, 卡头开通plus必备, 提炼cdk, 实卡号码, 无限接马) to classify into chatgpt-access-service.
  • Enforced a price safeguard for chatgpt-plus (< 8.00 CNY): demotes free helper accounts to chatgpt-account, helper tools/links to chatgpt-access-service, and flags abnormal low prices.
  • Synchronized classifier rules between apps/api/app/services/classifier.py and pipeline/common.py.

[3.7.40] - 2026-09-05

Added

• Admin Offer Sorting Aligned with Frontend:
  • GET /api/v1/admin/offers defaults to sort="frontend", matching the public site ranking (in_stock desc, cny_first, price asc, observed_at desc).
  • Added sort selector in the admin search & filter bar with options for frontend default, latest updated (updated_desc), price low-to-high (price_asc), and price high-to-low (price_desc).
• Admin Real-time Brand and Product Offer Counts:
  • GET /api/v1/admin/stats now calculates and returns product_counts (by product slug) and brand_counts (by brand platform) across all active, public offers.
  • Added real-time count badges to each Brand Chip and Product Chip in the admin navigation rails.
  • Returns X-Total-Count header in GET /api/v1/admin/offers to report exact matching offer count.
  • Displayed 共 {offerTotal} 条报价（已载入前 {offers.length} 条） summary header and added bottom pagination ("加载更多报价") so large product sets can be fully explored.
• Catalog Alignment:
  • Added chatgpt-pro (ChatGPT Pro) to PRODUCT_TABS.OpenAI in apps/web/lib/catalog.ts.

Fixed

• Prevent Scroll-to-Top on Admin Approval and Status Actions:
  • Implemented optimistic in-place state updates for offer actions (patchOffer, reclassifySingleOffer) to prevent DOM node unmounting and focus loss.
  • Added preserveScroll helper to lock and restore window scroll coordinates (window.scrollY) across all administrative updates (offers, shop intakes, user reports).
  • Explicitly marked all action buttons with type="button" and blurred active elements prior to asynchronous state updates.

[3.7.39] - 2026-09-05

Added

• Admin Category Hierarchy Aligned with Frontend:
  • Replaced the admin offers panel's flat product slug list with a brand and product rail matching the public catalog (apps/web/lib/catalog.ts).
  • Added dedicated navigation views for 🚫 受限/已隐藏 and ❓ 未分类商品 with real-time offer count badges.
  • Added brand parameter filtering in GET /api/v1/admin/offers and enriched offer responses with brand and product_name.
  • Added grouped <optgroup> selection by brand hierarchy for offer reclassification.
  • Clearly highlighted restriction reasons for restricted/hidden offers with one-click restore (恢复公开) and hide (隐藏/限制) actions.

Changed

• Paused Dujiao-Next Candidate Discovery & Web Submission:
  • Paused automatic Dujiao-Next discovery in remote refresh scripts (ENABLE_DUJIAO_DISCOVERY=false) and removed dujiao-next search queries from GitHub discovery.
  • Disabled the Dujiao-Next option on the public shop submission form (/shops/submit) with (暂停收录) badge.
  • Filtered out dujiao_next from the frontend catalog source platform filters.
  • Documented discovery pause and frontend disabled status in docs/CONNECTORS.md.

[3.7.38] - 2026-09-04

Changed

• Include Shop Address and Name in Intake Notification Emails:
  • Added merchant's original shop URL (店铺地址：{source_url}) and shop name (店铺名称：{shop_name}) to all applicant notification emails:
    • Auto-approval notifications (shop_request.approved)
    • Manual approval notifications (shop_request.approved)
    • Shop onboarding & publish notifications (shop_intake.onboarded), clearly distinguishing the merchant's 店铺地址 and 本站收录页面
    • Rejection notifications (shop_request.rejected)
    • Scan completion with no products (shop_intake.no_products) and validation failures (shop_intake.validation_failed)
  • Ensures merchants can immediately identify which shop request has been processed and follow the links directly.

[3.7.37] - 2026-09-04

Fixed

• Phone Verification Services Included in Benchmark Comparable Pricing:
  • Added verification_service (手机接码, 实卡接码, 短信验证) to COMPARABLE_DELIVERY_TYPES (is_comparable=True).
  • Fixes issue where the dedicated ChatGPT 手机接码 (chatgpt-access-service) product page displayed 0 offers and "暂无有货价" by default because offerQuery filters for comparable=true.

[3.7.36] - 2026-09-04

Changed

• Reverse-Proxy Tokens Included in Benchmark Pricing:
  • Included session_token delivery type (只能反代, 仅反代, 无账号密码, 发CDK不可网页) into COMPARABLE_DELIVERY_TYPES.
  • Reverse-proxy token accounts now participate in comparable pricing calculations (is_comparable=True) as a recognized account delivery mode.

[3.7.35] - 2026-09-04

Fixed

• Plus Account & SMS Service Precision Classification:
  • Eliminated false routing into chatgpt-access-service caused by substring matching of 接马 inside account status markers (未接马, 需自行接马, 自行接马, 免接马).
  • Removed legacy GENERIC_EMAIL_MARKERS (Gmail, iCloud) fallback from chatgpt-access-service, ensuring accounts with iCloud/Gmail emails (e.g. 韩国-PLUS-icloud邮箱, GP Plus gmail越南渠道) correctly classify as chatgpt-plus.
  • Added support for team / 周限额 in implicit brand detection so 长效周限额team routes accurately to chatgpt-k12.
  • Enhanced delivery type detection: 未接码/未接马 recognized as semi_finished_account (半成品/首登号), icloud and 保首登 recognized as finished_account (成品号).

[3.7.34] - 2026-09-04

Changed

• Codex Classification Merged into ChatGPT Plus & Respective Tiers:
  • Eliminated the top-level brand == "codex" prefix hijacking that forced all Codex-tagged items into codex-access.
  • Reclassified Codex accounts into their true underlying tiers: chatgpt-plus (for Plus/Sub2API/RT/CPA), chatgpt-go (for Codex Go), chatgpt-k12 (for Team), and chatgpt-account (for Free).
  • Preserved Codex, Sub2API, 带RT as scenario tags.
• Dedicated SMS Verification ("手机接码") Category:
  • Differentiated account attribute markers (已接码, 已接马, 已绑手机) from independent verification services (代接码, 手机接码, 实卡接码, 接码卡密).
  • Allowed SMS verification services through classifier and renamed public frontend tab 辅助服务 to 手机接码.
  • Removed Codex standalone tab from OpenAI navigation header.

[3.7.33] - 2026-09-04

Security

• CRLF & Email Header Injection Protection:
  • Sanitized subject, recipient, and dedupe_key across the outbox notification pipeline (apps/api/app/services/source_intake.py and apps/api/app/services/outbox.py).
  • Added strict email format validation in ShopRequestCreate forbidding CRLF, control characters, commas, and quotes.
  • Added strict sanitization to shop_name and note stripping all control characters and angle brackets to prevent header and template injection.
• SSRF Hardening:
  • Strengthened normalize_public_https_url in apps/api/app/services/source_platform.py to forbid internal hostnames, loopback/private/link-local/multicast IPs, and internal TLD suffixes (.local, .internal, .lan, .home, .corp, .intranet, .priv, .arpa).
  • Verified parameterized SQL queries across all repositories.

[3.7.32] - 2026-09-04

Added

• Automatic Shop Intake Approval: Enabled automated shop intake approval via SHOP_INTAKE_AUTO_APPROVE (True in production). When a store request completes automated security and platform detection (ldxp, dujiao_next, woocommerce, 16688, merchant_json, schema_org), it is automatically approved into the worker validation/publishing queue without requiring manual admin intervention.
• Admin Auto-Approval Email Notifications: Automatically dispatches a notification email (shop_request.auto_approved.admin) to configured administrator emails (SHOP_INTAKE_ADMIN_EMAILS) whenever a shop request is automatically approved, providing full store details, detected platform, and direct links to the admin console.

Changed

• Applicant Notification: Added explicit notifications to applicants confirming automatic approval upon successful probe detection.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.32 after CI passes.
• Rebuild and restart api.
• Set SHOP_INTAKE_AUTO_APPROVE=true in .env on production.

[3.7.31] - 2026-09-04

Added

• Admin Shop Intake Platform Modification & Manual Approval: Added platform selection dropdown to the admin console shop intake view (POST /api/v1/admin/source-intakes/{id}/platform), allowing administrators to switch an intake's platform type directly (e.g. from other to ldxp, dujiao_next, 16688, woocommerce, merchant_json, schema_org).
• On-Demand Platform Re-Detection: Added POST /api/v1/admin/source-intakes/{id}/redetect to re-trigger automatic platform detection for an intake using the latest detection rules.
• Intelligent Platform Auto-Upgrade on Approval: Enhanced POST /api/v1/admin/source-intakes/{id}/approve to automatically recognize supported platform URLs (such as wzyp.cn -> ldxp) and approve them directly without throwing 409 errors.

Fixed

• Source Detector LDXP Domain Coverage: Updated detector/probe.py to include wzyp.cn and www.wzyp.cn in LDXP_HOSTS, preventing new store applications on wzyp.cn (like shop #37) from being misclassified as 其他独立站 (other).

Safety and data scope

• No database schema migrations required.
• Existing and future wzyp.cn shop intakes can be approved and validated seamlessly.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.31 after CI passes.
• Rebuild api, web, and source-detector.

[3.7.30] - 2026-09-03

Fixed

• Detail-Driven Classification & Brand Detection: Enabled brand and tier recognition fallback to product detail page descriptions (raw_products.raw_json->>'description') when storefront titles are cryptic or non-standard.
• Universal Non-Product Exclusion: Systematically rejected pure tutorials (教程, 保姆教程, 图文教程, 反代教程), test items (测试商品, 不要拍, 不可拍), SMS verification ad services (接码渠道), virtual cards (虚拟卡, 0刀卡), and referral boost links from being classified into subscription products.
• Reverse Proxy / Sub2API Token Isolation: Offers offering only Sub2API/RT JSON tokens without login credentials (只能反代, 无账号密码) are now correctly classified as codex-access (session_token) rather than chatgpt-plus.
• ChatGPT Pro Integrity: Filtered out team bug sub-accounts (Team bug 子号) and API quota credits (20X 额度｜50美金) from chatgpt-pro and chatgpt-pro-20x.
• Multi-User Carpool Isolation: Expanded SHARED_POOL_MARKERS to catch 拼车, 共享账号, 多人共享, and 车位, ensuring carpool offers are tagged shared_pool and is_comparable = false, preventing them from distorting individual account comparison prices.
• Category Group False Positives: Preserved genuine subscriptions in merchant storefront categories ending in 分组 (e.g. Grok分组).

Safety and data scope

• Synchronized apps/api/app/services/classifier.py and pipeline/common.py.
• No database schema migrations required.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.30 after CI passes.
• Rebuild api, web, and importer.

[3.7.29] - 2026-09-03

Added

• Added admin console tabs for reviewing restricted (status=restricted) and unclassified (status=unclassified) offers, with live count badges on the stats bar.
• Added admin search and target product filter to the offer management table.
• Added support for manual reclassification (including unclassifying back to None) and single-offer auto-reclassification via POST /api/v1/admin/offers/{id}/reclassify.
• Added visual display of merchant original category and restriction / hidden reasons in the admin offer view.

Fixed

• Strengthened classifier and pipeline normalization to reject API relay groups (e.g. plus分组), relay model channels (e.g. (cx,5,4)), and non-20 dollar credit quotas from being falsely classified as chatgpt-plus, chatgpt-pro, or codex-access.

Safety and data scope

• No schema migrations required.
• Existing restricted offers can now be inspected, audited, and reclassified directly from /admin.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.29 after CI passes.
• Rebuild api, web, and importer.

[3.7.28] - 2026-09-03

Added

• Added support for wzyp.cn and www.wzyp.cn storefront hostnames in the ldxp platform detector and crawler normalization.
• Added https://wzyp.cn/shop/KFLA to public source discovery seeds and crawler candidate database.
• Added wzyp.cn guidance to source intake copy in the web application.

Safety and data scope

• Preserves token.casefold() as source_key and token as shop_token under the ldxp platform namespace.
• No database migration is required.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.28 after CI passes.
• Rebuild api, source-detector, web, importer, and crawler.

[3.7.24] - 2026-08-28

Fixed

• Preserved the flat token-list contract of GET /api/v1/shops and added a paginated shop-card endpoint for directory pages.
• Excluded hidden products and shops without current public offers from public metadata, source pages, and sitemap entries; added pagination for shop directories.
• Unified API and pipeline 16688 classification with source-category context, rejected non-product aliases, and retained valid API-credit classification.
• Rotated 16688 discovery categories within the global page budget so one category cannot starve the others.

Safety and data scope

• The 16688 default approval behavior is unchanged: newly discovered offers still follow the existing approval policy.
• No database migration is required.

Deployment

• Follow docs/QUICK_DEPLOY.md and deploy only tag v3.7.24 after CI passes.
• Rebuild api, source-detector, web, importer, and crawler.
• Run one complete multi-source refresh because the release changes crawler and pipeline behavior.

[3.7.21] - 2026-08-27

Fixed

• Enforced attempt matching before accepting an idempotent LDXP onboarding response, so a stale onboarding report cannot be mistaken for a retry of the current attempt.

[3.7.20] - 2026-08-27

Fixed

• Prevented completed LDXP intake attempts from being re-reported on later inventory scans, eliminating false production 409 errors while retaining the metadata required for publication onboarding.
• Made same-attempt scan-result retries idempotent after an intake reaches a closed state; a newer attempt remains rejected as stale.
• Restored LDXP application onboarding after a successful atomic multi-source snapshot, so validated applications with public offers are marked as published.

[3.7.19] - 2026-08-27

Fixed

• 16688 discovery now uses the platform's public AI source marketplace to resolve public goods to canonical /shop/{shop_no} URLs before the existing detector and review flow.
• Unified source discovery now runs before legacy Dujiao revalidation, and the 16688 and Common Crawl adapters run before the high-volume Bing adapter so Bing cannot consume their discovery opportunity.

[3.7.17] - 2026-08-27

Fixed

• Updated Common Crawl discovery regression coverage for the new platform-reserved budget semantics.

[3.7.16] - 2026-08-27

Fixed

• Scheduled unified source discovery now obtains its worker key inside the Compose crawler container instead of incorrectly requiring the systemd host service to load the production .env.
• Common Crawl discovery reserves candidate capacity for 16688, so high-volume LDXP URLs cannot consume the complete run budget before 16688 shop paths are queried.

[3.7.15] - 2026-08-27

Fixed

• The isolated source detector now falls back to another already-validated public DNS address when an initial socket connection fails, allowing 16688 sources to work on IPv4-only egress networks that resolve IPv6 first.
• A successful fallback address is pinned for the rest of the detection run without re-resolving DNS or weakening the existing public-address and TLS hostname checks.

[3.7.14] - 2026-08-27

Added

• Added public 16688 shop intake, source detection, approval, atomic publication, and the 16688 connector for public shop and goods APIs.
• Normalized 16688 aliases such as /shop/HARVEY to the canonical shop number and scoped shop tokens and product keys by platform so same-named shops do not collide.
• Extended automatic discovery through Bing and Common Crawl for 16688 shop URLs, while keeping discovery auto-approval disabled by default.

Changed

• Added the v11 database constraint migration for 16688 source intakes and discovery candidates.

[3.7.13] - 2026-08-22

Changed

• Refreshed the full Web experience with the Signal Ledger visual system, clearer hierarchy, more consistent typography, responsive layouts, and unified interaction states.
• Reworked public-facing copy to describe observed prices and data freshness more precisely across product, guide, watchlist, submission, policy, and administration surfaces.
• Connected catalog search terms to the public catalog API and preserved available offers when product metadata is incomplete.

[3.7.12] - 2026-08-22

Fixed

• Each shop scan now runs inside a supervised browser Worker with a hard wall-clock deadline. A wedged Playwright page or renderer is terminated with its Chromium process group, recorded as a transient failure, and scanning continues with a fresh browser.

[3.7.11] - 2026-08-20

Fixed

• The crawler now stores only current matches and run summaries. The unused per-scan product_snapshots history no longer grows the operational SQLite database and delays each publication copy.
• Ten-minute inventory refreshes now update only LDXP data while carrying other published sources forward, so a transient external-source timeout cannot block inventory publication.

[3.7.10] - 2026-08-20

Fixed

• Compact publication uses a deferred SQLite transaction so the production SQLite runtime does not request a write lock on the attached read-only crawler database.

[3.7.9] - 2026-08-20

Fixed

• The compact crawler publication helper now runs on the production host's Python 3.6 runtime.

[3.7.8] - 2026-08-20

Fixed

• Scheduled publication now copies only the three current crawler tables used by the publisher instead of validating, copying, and revalidating the full multi-gigabyte history database on every refresh.

[3.7.7] - 2026-08-20

Fixed

• Persistent Chromium teardown now stops the Playwright connection directly instead of waiting indefinitely for BrowserContext.close() after all shops were scanned.

[3.7.6] - 2026-08-20

Fixed

• Browser-replayed shop API requests now abort at the configured crawler timeout instead of allowing one unresponsive source to block every later scan and publication.
• Browser refreshes remove stale Chromium singleton symlinks left by a terminated crawler, and inventory refreshes receive a one-hour service budget so a completed scan can finish atomic publication.
• Release metadata is aligned on 3.7.6 across the API, Web package, lockfile, and repository version marker.

[3.7.4] - 2026-08-08

Fixed

• Restored text and icon color utilities on form controls so dark action buttons remain readable.

[3.7.3] - 2026-08-08

Changed

• Redesigned the Web pages around a unified product, catalog, offer, guide, and source-review UI system.
• Reworked public copy to distinguish current observations from live data, separate empty and unavailable states, and remove internal workflow wording from user-facing surfaces.
• Clarified watchlist/Atom subscription behavior, correction privacy, offer grouping, information coverage, and source intake outcomes.
• Restored the public author-support entry and kept the existing community prompt wording unchanged.

Fixed

• Restored support QR configuration defaults for local and production Web builds.
• Updated admin actions and source-discovery labels so buttons describe the action instead of repeating the current state.

[3.7.2] - 2026-08-06

Fixed

• LDXP sources in blocked or challenge_required no longer remain permanently excluded after a transient source-level challenge.
• Blocked sources receive bounded retry times and can be forced immediately with --retry-blocked; consecutive source-level failures still stop the scan through the existing circuit breaker.
• Regression coverage now verifies that an all-blocked batch is recorded as failed rather than successful.

Changed

• Crawler self-tests and pytest coverage now run in CI.

[3.7.0] - unreleased

Added

• Unified source discovery engine: seed/Bing/GitHub/Common Crawl adapters submit normalized candidates to a PostgreSQL candidate pool (source_discovery_runs, source_candidates, v10 migration).
• Source Detector qualification of discovered candidates with bounded public samples and AI product classification, plus strict auto-approval for Dujiao-Next and WooCommerce and manual review for Schema.org and Merchant JSON.
• Internal candidate claim/lease/result APIs, idempotent promotion into source_intakes, and admin discovery funnel/controls.
• Production Dujiao discovery now runs GitHub sources with optional token, full AI keywords, and env-driven budgets.

[3.6.0] - 2026-08-03

Added

• WooCommerce Store API connector with exact minor-unit pricing, complete pagination validation, and safe variation fallback.
• Schema.org sitemap and product-page JSON-LD connector with bounded discovery and same-origin HTTPS validation.
• Dujiao-Next qualified candidates are auto-approved, and GitHub public repository homepages are a new passive discovery source.
• Source intake, detector, pipeline publication, Web labels, and a v9 database migration for the new source platforms.

Changed

• Detector platform probing order is now Dujiao-Next, WooCommerce, Merchant JSON, then Schema.org.
• Directly submitted sitemap and product-page URLs are preserved exactly through detection and publication.
• WooCommerce products that are not purchasable never count as in-stock or enter lowest-price comparisons.

[3.5.0] - 2026-08-03

[3.5.0] - 2026-08-03

Added

• Dujiao-Next connector support with brand-aware shop metadata, paginated products, variants, currency preservation, and reviewed-source publication.
• Public-fingerprint discovery with bounded candidate quotas, stale revalidation, isolated platform detection, and administrator-controlled intake routing.
• Merchant JSON intake publication and persistent multi-source refresh across LDXP, Dujiao-Next, and approved merchant feeds.

Changed

• Complete catalog publication is atomic across all sources and now runs in the dedicated Importer image.
• Intake publication distinguishes raw records, classified offers, and fresh public offers; only a truly visible offer marks a source as published.
• Published sources remain in later complete refreshes, while disabled or review-required sources leave the next snapshot.
• Product brand and source platform are exposed separately across the API and Web application.

Security

• Detector, Pipeline connectors, and Dujiao discovery share a bounded HTTPS client that pins validated public IPs while preserving TLS SNI and certificate verification.
• Public intake submissions no longer fetch user-controlled URLs inside the API process; the Detector has no database credentials or default-network access.
• Merchant feed shop identity is derived from the canonical feed URL, and public shop/product links reject credentials, fragments, control characters, and non-HTTPS schemes.
• Detector egress is designed for a production firewall policy that permits only public TCP/443 destinations.

[3.4.0] - 2026-08-02

Added

• Optional GitHub Star and author-support entry points in the public footer.
• Low-frequency, session-aware community prompts that stay disabled on administrator and shop-submission routes.
• Accessible WeChat Pay and Alipay support dialog with keyboard dismissal and mobile layouts.

Changed

• Production Web containers can mount support QR images from data/support as read-only runtime assets.
• Production preflight validates both public HTTPS QR URLs whenever author support is enabled.

Security

• Payment QR images and production support configuration remain outside the public Git repository.
• The support dialog does not display or configure a payee name and never records payment information.

[3.3.1] - 2026-08-02

Added

• Administrator intake emails now include a direct link to the matching review item in the admin panel.
• Final onboarding emails now include the published public shop page.

Changed

• Admin intake links still require the administrator key, then scroll to and highlight the referenced request after authentication.

Security

• Administrator links contain only the intake identifier and never include the administrator key.

[3.3.0] - 2026-08-02

Added

• Durable shop-intake records with explicit review, validation, onboarding, rejection, and retry states.
• Admin controls for approving, rejecting, retrying, and inspecting source-intake notification delivery.
• Applicant and administrator email notifications through Resend, with SMTP fallback and a transactional outbox worker.
• LDXP intake bridges for crawler and pipeline jobs, protected by a dedicated worker credential and leased claims.
• Idempotent migrate_shop_intake_v6.py migration for historical shop requests and notification outbox storage.

Changed

• Shop submissions now require a valid contact email and return a stable request identifier for duplicate requests.
• Production preflight now requires administrator recipients, a separate intake-worker key, and a complete Resend or SMTP configuration.
• Production Compose and deployment guidance now include the notification worker and the v6 intake migration.

Security

• Intake-worker access is isolated from the administrator API key.
• Source validation failures are sanitized before storage or email delivery.
• Resend credentials remain environment-only and are never written to application logs.

[3.2.1] - 2026-07-30

Changed

• Rewrote homepage, catalog, product-detail, About, methodology, shop, watchlist, and footer copy in user-facing language.
• Replaced internal pricing and crawler terminology with clearer descriptions such as recent in-stock low, common price, quote coverage, and source update status.
• Simplified grouped-offer labels, anomaly warnings, source links, update timestamps, and product FAQs without changing pricing or ranking behavior.

[3.2.0] - 2026-07-29

Added

• Official price references, verification dates, product data-quality scores, source scan-health facts, and daily aggregated price/stock trends.
• Browser-local watchlists and privacy-preserving Atom price/restock subscriptions.
• Public methodology, privacy, terms, security, developer, and correction-log pages.
• Public correction summaries with optional merchant responses while keeping reporter contacts private.
• Generic connector protocol, merchant HTTPS JSON Feed importer, submission flow, fixtures, and tests.
• Full delivery, period, warranty, fulfillment, freshness, stock, and price-range filters across catalog and product pages.

Changed

• Default catalog ranking now prioritizes data quality and freshness before price.
• Homepage and directory counters use one published-snapshot scope and show trusted/comparable context.
• Price history presentation uses daily aggregates instead of connecting unrelated raw observations.
• Source-health labels explicitly describe crawler availability rather than merchant reputation.

Security

• Merchant Feed submissions require public HTTPS URLs and reject localhost, internal hostnames, and private/reserved IP literals.
• Public correction endpoints exclude raw report messages and contact information.

[3.1.0] - 2026-07-29

Added

• Trusted-price scoring derived from comparable inventory and delivery-type medians.
• trusted_offer_count and median_price in public product responses.
• Per-offer is_trusted_price indicator while retaining anomaly warnings.
• API pricing unit tests and GitHub Actions checks for API and Web builds.
• Version, security, contribution, and release documentation.

Changed

• Product cards and product details now use the trusted lowest price as the primary price.
• Extremely low or strongly off-median offers remain visible as source evidence but no longer lead the primary ranking.
• Related/all-in-stock lowest price remains available through related_lowest_price for backward compatibility.
• Homepage copy now distinguishes trustworthy rankings from raw low prices.
• Product structured data reports trusted offers rather than every comparable offer.

Fixed

• Prevented ¥0.01 promotion, balance, trial, or restricted offers from becoming the headline price.
• Prevented anomalous representatives from being selected ahead of trustworthy offers inside grouped results.

[3.0.0]

• Initial public architecture with FastAPI, Next.js, snapshots, classification, grouping, and price history.