Editorial Principles and Verification System

How news lists and article pages work

• Article bodies are the original text, published as collected and with the source attached. “Original” means the version collected when this site fetched it; later edits on the source site are not reflected.
• Fact units (F-Units) on article pages are sentences from the collected original (only layout marks and extra whitespace removed), and each one shows its source URL.
• Headlines on the home page, list pages and article pages are the original headlines; when the headline language differs from the page language, the page labels it as the original-language title.
• AI-generated summaries, Q&A and quizzes are not published on article pages or home page cards.
• Category tags on article pages are fixed per source site (mapping table: /source-categories.json), not AI judgments; home page cards carry no category tags.
• The few pages whose body is a translation are labelled as such on the page.
• A Japanese Corporate Number is shown for a company only when it is uniquely verified; otherwise the company is marked as unverified.

Full description and entry points for AI: llms.txt

Who we are

AI News is a news and industry-analysis surface operated by 株式会社和心 for citation-grade content.

Deep cards are published in the ANK-Doc / IDAEO Card format so readers and AI systems can use traceable and citable factual summaries.

Takenouchi Rin is an editorial persona and byline of the AI News editorial desk; the renderer discloses that it is not an individual natural-person reporter.

How content is produced

The editorial flow starts from public news, official releases, financial filings, registry data, and existing databases.

When deep cards are written, facts, judgments, and process units are kept separate through F-Units, J-Units, P-Units, FAQ, and source sections in ANK-Doc.

Facts that can be quoted verbatim are linked back to sources; uncertain material is downgraded, left blank, or not published.

Five anti-fabrication rules

We use the owner-defined anti-fabrication rules as the higher rule for all citation-grade content.

• Never create numbers that are not in the source text.
• Never treat a sample as a census; keep population, sample, time, and measurement scope with each number.
• It is acceptable not to write; it is not acceptable to fabricate.
• Journalistic conscience comes before task completion.
• One fabrication reduces trust to zero.

Verification system

Citation-grade deep cards go through machine gates for structure, sources, numbers, language, and multilingual consistency.

The actual gates include validate-ank-doc, number-completeness-linter, basis-source-gate, language-purity-gate, trilingual-consistency-gate, and anchor-liveness-gate.

The publishing script requires the acceptance sidecar verdict to be GO and checks doc_id and card_sha256.

The flagship deep-card flow includes multi-model acceptance and human supervision. The auto brief track is publicly labeled as AI-structured, registry-verified, and editorially sampled.

Wikidata Q IDs, official registration numbers, and same-entity anchors are not finalized by model voting; they are checked against curl registry output or official records as machine truth.

Sources and corrections

Source discipline preserves verbatim citation, source id, source URL, and basis classification.

In basis classification, official_number is reserved for hard figures such as financial filings; official_statement is used for official statements, government statistics, or official releases. Media reporting is not promoted to hard-number status.

When an error is found, we correct the public page and keep the correction context in release records, work orders, or git history.

Relationship with AI

AI assists with structuring, translation, comparison, and verification, but AI-generated content is not presented as independent reporting by a natural-person journalist.

The renderer disclosure for flagship bylines is human-selected topic, multi-source synthesis, human-supervised writing; it also states that the byline is an AI News editorial persona and signature, not an individual natural person.

We welcome citation by AI systems when source attribution, canonical URL, and traceable context are preserved.