How a score is made, and what it is not

The work asks three questions about the same reader in sequence: whether they can read the company, whether they can judge its capital decisions, and whether they ultimately invest. These may not move together, and the third is a hypothesis that has not been tested.

• A defined search per question — navigate the English IR site, the last two quarters of results materials, the latest annual and governance reports, the latest AGM notice, two years of material announcements, plus one external search.
• A document not found within that scope is recorded as not-found, with a note saying where we looked — never silently as zero.
• Questions not yet reached score nothing and are excluded from the denominator, so a thin score announces itself instead of hiding.
• Every company score is published as a floor and a ceiling. The gap between them is exactly how much remains unchecked.
• Rubric versions are frozen once a cohort is scored against them; changes fork a new version rather than silently revising history.
• Scores are verified in a rendered browser, not by fetching HTML. A page that looks empty to a script can be complete to a reader, and text present in the source can be hidden from one. Both have produced wrong scores here, and both are recorded in the change log.

34 questions in five assessment areas, each scored 0 to 5. The areas are weighted as below.

Each question is scored 0 to 5. Where a question has its own criteria, those take precedence.

A Japanese–English publication gap scores 5 for the same day, 4 for 1–7 days, 3 for 8–30 days, 2 for 31–90 days and 1 beyond 90 days.

The full rules: what earns each score on all 34 questions, the special cases, the evidence required and every dated change. Scoring rules →

Market cap is used only to place each company in a size band (under ¥100bn, ¥100–500bn, ¥500bn–2tn, over ¥2tn) for the size filter and the comparison with peers. It plays no part in any score, so it is not central to the index. We refresh it once a quarter, at each cohort release, from one public source for every company. A company moves between bands only then, not with each day's share price.

The second programme scores corporate actions rather than documents, so it uses its own four checks. Each is rated 0 to 5 by an analyst against the company’s filings, with a written reason and a dated source:

• Was the promise specific? A dated, measurable commitment scores higher than a stated intention.
• Can the steps be traced? Decision, execution and completion each evidenced in their own disclosure.
• Were the economics explained? Capital employed, expected return, risks and alternatives, in numbers.
• Did it report back against the target? Dated actuals reconciled with the original promise, and material misses explained.

A company’s figure is the mean of its rated checks, normalised to 100. Three rules keep it honest, and they mirror the disclosure index: a check whose outcome is not yet due is excluded rather than counted as a failure; a check awaiting verification produces a range rather than a point estimate; and a question nobody has searched is never entered as a zero.

Every score is provisional, because the search for a company’s material actions is a selection rather than a complete sweep. That is why the monitor publishes no ranking, and why its figures must not be compared across companies. A company is given a figure out of 100 only once three or more of its actions are reviewed; with fewer, each action’s four ratings are shown instead. See the companies →

It scores the explanation, not the decision. A buyback rated 5 is one a reader can follow from promise to outcome. Whether it was a good use of the money is a different question, and not one this index answers.

Intended

Self-diagnosis and budget justification by IR teams; a published benchmark for anyone reading Japanese companies in English; academic or policy research into disclosure practice.

Not investment advice

Nothing here is a recommendation to buy, sell or hold any security. Scores measure whether a decision was explained, never whether it was a good decision.

No league table

Coverage is uneven between companies, and a ranking built on uneven coverage would mislead. Scores are comparable within a rubric version and a cohort run, not across them.

No predictive claim

Whether better English disclosure precedes foreign buying is the open question, not a finding. The test requires two scored periods per company separated by at least six months; today it returns no companies at all, which is the correct answer.

Coverage is incomplete and uneven. 128 companies is a research cohort, not a market. The capital-allocation programme covers 112 companies and 134 corporate actions, and its action universe is a selection rather than a complete search, so its scores must not be used for cross-company comparison.

The rubric is a documented reconstruction. The v1.1 question set was rebuilt from the recorded pillars, weights and amendments after the original was lost. It is consistent with all of them and it is not the original. This is recorded in the database itself, not only here.

Sources are public. Every observation derives from material the companies themselves published — IR websites, exchange filings and statutory reports. No non-public information is used, sought or accepted.

AI assistance is used for document retrieval, comparison and drafting. Every score is reviewed by a person against the primary source before publication, and the review date is recorded with the observation.

Errors are published, not buried. The change log records every score changed, including the ones changed against us.