What the evidence supports, and what it does not
A reproducible record of what a non-Japanese reader can actually see in a Japanese company’s disclosure — and whether its capital decisions are explained well enough to judge. Every claim below is stated at the strength the evidence currently supports, and no further.
Every score traces to a document you can open
128 companies scored on all 34 questions, each answer checked against a named document on a stated date. An answer we looked for and did not find is recorded with a note of where we looked, and counts as zero.
The database refuses to store a score without a citation, and refuses to store a “not found” without a note describing where we looked. A company’s figure is published as a floor and a ceiling, and the distance between them is exactly how much remains unverified — so a thin score announces itself instead of hiding inside an average.
Every score we have changed is published, including the one we lowered against a company and had to restore when the evidence did not hold. Open any question and read the scores →
Every capital decision checked against what it leaves unexplained
Each monitored corporate action is tested against the questions a reader would need answered to judge it. Where the company’s own disclosure stops, the open question is recorded — labelled as analysis and never attributed to the company.
What is the post-exit ROIC and ROE bridge, including remaining
restructuring costs and the timing of benefits?
IR clarification sought: state the closing status and, where
disclosure obligations permit, give a comparable before/after segment
and capital-employed baseline.
The discipline behind it is what makes it citable. Each of the five answers per action is labelled sourced, partial or analyst_question, and any factual answer must link a primary source — issuer filing, TDnet or EDINET first, never a press aggregator.
A missing deal case is only called a disclosure gap when a source explicitly confirms the non-disclosure. Otherwise it is recorded as not-yet-verified. That distinction is what stops a gap being claimed where the company did disclose — on page 14.
It already surfaces patterns rather than one-offs. Two unrelated companies, two different deal types, the same refusal: THK withholds its transfer price citing non-disclosure obligations; Daifuku withholds the EISENMANN acquisition price citing confidentiality, offering only “less than 5% of consolidated net assets.” That is a sector question, not a company question.
Reading quality does not proxy for decision quality
Scored both ways, the two companies invert. The one that reconciles its medium-term plan best explains its deals worst, and the reverse.
| Company | Plan reconciliation | Deal explanation |
|---|---|---|
| Daifuku 6383 | 4 | 2 |
| THK 6481 | 0 | 3 |
Two companies is an anecdote, and we label it as one. But it is a testable anecdote with a clear prediction: if disclosure quality genuinely proxied for decision quality, these lines would run parallel. They cross. At n = 20 the same result would mean a disclosure score cannot stand in for capital-allocation quality — and that the two need measuring separately, which is the thesis this work is built on.
What we do not claim
Stated plainly, because you would find each of these in an afternoon and we would rather you heard it from us.
- No predictive power. Not yet.
- The hypothesis is that English disclosure quality precedes foreign ownership change. Testing it needs two scored periods per company; most of the cohort has one. Until then this is a measurement tool, not a signal, and we will not present it as one.
- The capital monitor covers 112 companies.
- 134 selected actions as of 21 October 2026, drawn largely from a semiconductor and advanced-manufacturing universe. Each company's action universe is a selection, not a complete search, so the scores must not enter a cross-company ranking. 10 of them now have both scored actions and a disclosure score. The two-company finding above was drawn from the first two and has not yet been re-run on the others.
- The disclosure index covers 128 companies, and they are a selection.
- They are a research selection, not a sample of the market, and review dates differ by company. Each company page states its date and every correction to its record.
- The v1.1 question set is a reconstruction.
- 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. That is recorded in the database itself, not only in the documentation.