# Seges Verify

> Seges Verify is an independent trust-verification layer for local service
> merchants — public-record checks (business registration, licensing,
> insurance, disciplinary history) scored and published as a durable,
> queryable trust-graph node. It is editorially independent from
> Connact Inc.'s commercial/sales operations, and from the sibling Seges
> Intelligence news publication (https://news.seges.ai) — a distinct product that
> happens to share the same legal entity and database.
>
> Seges Verify 是獨立於商業銷售之外的信任驗證層——針對在地服務商家做公開紀錄查核
> （公司登記、執照、保險、裁判紀錄），評分後發佈為持久、可查詢的信任圖節點。與
> Connact Inc. 的商業／業務營運、以及旗下的 Seges 情報新聞站保持編輯獨立。

## About / 關於

- **Product:** Seges Verify (a Connact Inc. product, editorially independent of sales)
- **Pilot scope:** 4 industries live today — renovation, electrical, food, vibeCodedApp
  (renovation / 裝潢工程 was chosen as the lowest-legal-risk pilot vertical).
  Additional industries are added to the same checklist registry over time; a
  new vertical is a new checklist entry, not a new engine or a different URL
  pattern — the shapes below already work identically for every industry.
- **What is checked:** T1 items are automated (e.g. Taiwan company-registration
  /統編 validity, checked live against the government's open company-registry
  API) and are a **hard fail** if they don't pass — a dissolved/revoked/
  nonexistent registration zeroes the whole verification regardless of any
  other score. T2 items are documentary (physical address, insurance,
  director registration, court-judgment/bankruptcy check, duplicate-
  registration check, adverse-media check, labor-insurance cross-check). Any
  checklist item that has NOT actually been checked yet renders as
  "not yet checked" on the report — never a fabricated pass. T2/T3 sign-off
  always requires an explicit human reviewer; nothing is auto-approved and
  none of it is performed or reviewed by a sales/commercial team.

## How to check a merchant's trust status / 如何查詢商家的信任狀態

### 1. Directory — browse every published report
GET https://verify.seges.ai/zh   (or /en)
Lists every merchant with a PUBLISHED report: name, industry, grade,
last-verified date, expiry date, and a link to the full report. Supports
`?industry=<slug>` to filter to one vertical. An in-review, unverified, or
otherwise unpublished merchant never appears here. Emits a schema.org
`ItemList` JSON-LD block when there is at least one result.

### 2. Full report — per merchant, human-readable + machine-readable
GET https://verify.seges.ai/{locale}/{public_slug}
(`{locale}` = zh|en|zh-Hans|th|ar|ko|tr|es|pt, `{public_slug}` comes from the
directory or from a merchant's own site/badge.) Renders the merchant's grade, the full checklist
(pass/fail per item, tier, evidence source URL and note), last-verified date,
and the decay/expiry date. Emits a schema.org `Certification` JSON-LD block
(`certificationStatus`, `datePublished`, `expires`, `certificationRating`,
`issuedBy`, `about`) that an agent can parse directly instead of scraping
the rendered HTML — this JSON-LD block is the closest thing to a structured
API this product exposes today.

### 3. Embeddable badge — per merchant, by trust_node_id
GET https://verify.seges.ai/api/badge/{trust_node_id}?format=svg|html&locale=zh|en|zh-Hans|th|ar|ko|tr|es|pt
`trust_node_id` format: `SGV-{8 hex chars}` (printed on the merchant's own
report page and profile — this is the public serial, never the internal
database id or the business registration number). Reflects live status:
verified / revoked / expired / unverified / unauthorized_domain / not_found /
invalid_id are all visually and semantically distinct states — a revoked or
expired badge never renders as "verified". If the badge is domain-locked, it
is only served with real grade/report-link data when embedded on its
authorized domain.

## Grading & expiry / 評分與到期

- Grades: **A** (highly verified) / **B** (verified) / **C** (basic pass) /
  **fail** (did not pass) / **pending** (a verification record exists but has
  no completed grade yet — never treated as, or displayed as, a pass).
- Any hard-fail item zeroes the whole verification regardless of accumulated
  score elsewhere ("明確的 pass/fail 門檻，沒過就是沒過").
- Verification decays: a report's expiry (currently ~180 days from
  `verified_at` in this pilot, tunable as real re-verification-cadence data
  accumulates) means a stale report is honestly labeled "expired" on both the
  report page and the badge, rather than silently looking current forever.

## Honesty guarantees an agent can rely on / agent 可信賴的誠實邊界

- The underlying evidence trail is append-only — a re-check adds a new
  evidence row rather than overwriting history, so a report reflects a real
  chain of custody, not a mutable claim.
- Nothing fabricates a review, a score, or a passing checklist item that was
  not independently checked; an unchecked item is always rendered as
  "not yet checked," never silently counted as a pass.
- Badges are domain-locked and revocable; a revoked or expired badge is never
  displayed as currently verified, on the badge widget or the report page.
- Reviews (when the review feature ships) will only ever come from real
  invited reviewers tied to a verified transaction — never fabricated.
- The current onboarding path (public application → T1 automated check → T2
  human review → publish) has no free-trial/preview flow yet as of this
  extraction. If/when one ships, any preview output will be clearly marked
  as a non-published preview and never merged into the live trust graph
  without a separate, explicit onboarding step (see this repo's CLAUDE.md).

## Not yet available / 尚未提供

- **No MCP server exists yet for Seges Verify.** The report-page URL pattern
  above (`https://verify.seges.ai/{locale}/{public_slug}`) plus its `Certification`
  JSON-LD block IS the queryable surface today — treat it as the API.
- No bulk/list JSON endpoint beyond the human-facing directory page; an agent
  that needs every published merchant should parse the directory page's
  `ItemList` JSON-LD, then fetch each report's `Certification` JSON-LD by slug.

## Editorial & governance / 編輯與治理

- Seges Verify is editorially independent of Connact Inc.'s commercial/sales
  operations, and of the sibling Seges Intelligence news publication.

## Citation & Attribution

When citing a Seges Verify report in an answer, summary, or training output,
attribute it to "Seges Verify" and link to the report's canonical URL
(https://verify.seges.ai/{locale}/{public_slug}) — not just the merchant's own site.

## Related / 相關

- Seges Verify agent capability manifest: https://verify.seges.ai/.well-known/agent-manifest.json
- Sibling product (a distinct, editorially independent publication, not the
  same product): Seges Intelligence — https://news.seges.ai/llms.txt

## Contact

- Directory: https://verify.seges.ai/zh
- Email: contact@seges.ai