Inspiration
Before sharing personal information or making a purchase, a shopper may want to understand what a company has disclosed about past incidents. Finding that context can require connecting a news report, a corporate announcement, and a shopping website operating under a different brand or domain.
Morashiter brings those pieces together. Its purpose is to make public incident disclosures easier to find, review, and use at the moment they become relevant.
The project starts with a maintained collection of disclosures from Japanese companies and organizations. The Chrome extension brings that information into the browsing experience. The PayPal demonstration explores a further step: using source-linked information to support an informed purchase decision.
What it does
Morashiter collects news that can lead to incident announcements, retrieves official source material, and supports human review before publication.
Its public catalog presents structured disclosure records, including the organization, incident description, publication date, reported impact, affected information categories, and links to official sources. Records distinguish confirmed disclosures from possible exposure where those distinctions are available. They also include service interruptions described in the underlying announcements.
The Chrome extension connects this catalog to registered shopping domains. On the live Gin no Sara website, the extension displayed a badge showing two matching disclosure records. Opening it showed the associated operator, publication dates, response information, and links to official announcements.
A separate fictional store, Aozora Goods, demonstrates how this information experience can extend to checkout. A shopper reviews fictional notices and an AI purchase memo, then chooses whether to defer the purchase or continue through buyer-approved PayPal Sandbox checkout.
How it is built
The collection and review backend uses PHP/Laravel and PostgreSQL. News discovery and official-source retrieval are separate stages. Administrators review candidate domains before accepting them as official sources, and the interface keeps unresolved searches and retrieval failures visible.
Sakura AI Engine structures official announcements into reviewable drafts. The review screen places the source URL alongside the proposed organization, impact, affected information, publication date, response status, and summary. Human reviewers can inspect and approve or reject those drafts.
Approved incident revisions feed a static public site. Publication is separate from approval, with release history and support for corrections and withdrawals. A scheduled publication process runs at 03:30 JST and records when there are no changes to publish.
Shopping domains have their own mapping to operating companies, supported by evidence URLs. The extension checks exact registered hosts against published data. Its implementation retrieves data using a truncated domain-hash prefix and performs the exact match locally, without sending the full browsing URL.
The checkout demonstration uses Sakura AI Engine with llm-jp-3.1-8x13b-instruct4. The model selects and organizes approved statements from fictional notices; the server validates the statements and source references. PayPal JavaScript SDK v6 and Orders v2 provide buyer approval and server-side Sandbox capture. The AI has no payment authority.
Challenges
A news link may identify an organization without pointing to the relevant announcement. Morashiter therefore needs explicit source verification, unresolved-candidate handling, and a way to add the correct official URL.
Connecting a shopping brand to its operator is another distinct problem. The announcement may live on a corporate website while the shopper visits a separate storefront. Maintaining evidence-backed mappings is essential to making the extension useful.
Incident disclosures also change over time. Possible exposure, confirmed leakage, investigation updates, and service restoration should retain their original context. Publication dates, source links, revisions, and human review help preserve those distinctions.
At checkout, the same discipline applies to payment state: buyer approval, completed capture, and an unknown outcome after an interruption must remain distinguishable.
Accomplishments
As of November 12, 2026, the deployed system contained 56 collected news records, seven Sakura-generated drafts awaiting review, and 30 published disclosure records.
The administration interface showed collection activity, source-verification work, review queues, revision controls, and publication history. This provides an operating foundation for maintaining the catalog.
The Chrome extension also worked on a real shopping website: Gin no Sara displayed two source-linked records associated with its operator. The current merchant registry contains three shopping hosts, establishing a small, concrete starting point for broader coverage.
The separate Aozora Goods implementation adds an AI-assisted review step and PayPal Sandbox checkout, demonstrating a purchase-time application of the underlying information approach.
Engineering lessons
Useful incident information requires more than extracting text. It needs a traceable source, an organizational identity, a date, an interpretation that preserves uncertainty, and a review process.
Collection, approval, publication, and browser matching have different responsibilities. Keeping those stages explicit makes failures and unfinished work visible.
Dates also need precise meanings. A browser lookup timestamp describes when a match was checked; it does not establish when the underlying disclosure was reviewed or whether it remains the latest announcement.
AI is useful within these boundaries: it helps organize source material, while people retain publication responsibility and buyers retain payment authority.
What's next
The next priorities are to expand verified company and shopping-domain coverage, resolve more unmatched official announcements, and improve how related disclosures and subsequent updates are connected.
The interface also needs clearer source-review dates and freshness information, particularly where a record currently directs the reader back to the announcement instead of displaying a specific confirmation date.
For the purchase experience, the next step is to evaluate how reviewed catalog information can support clearer decisions, building on the fictional AI and PayPal demonstration. Broader browser verification and a reproducible reviewer setup will help make that experience easier to assess.
Built With
- docker
- javascript
- laravel
- paypal
- php
- postgresql
- python
- sakura-ai-engine

