Same Smiths, Different Blades

Collectors often describe the designations as a ladder: Jūyō, then Tokubetsu Jūyō, then Jūyō Bunkazai, then Kokuhō, each a step up in quality. That model is convenient for anyone who submits to shinsa, but the designation record does not support it.

This paper compares the five designations using every designated blade in the record: Jūyō Tōken never promoted (10,923), Tokubetsu Jūyō (1,198), Jūyō Bijutsuhin (984), Jūyō Bunkazai not also Kokuhō (765) and Kokuhō (102). Figures 1 and 2 summarise the result.

Tokubetsu Jūyō and Jūyō Bunkazai blades come from smiths of the same standing. The Jūbun blades differ in what has survived of them. They are more often signed, unshortened, in original tachi form and recorded as meibutsu, and this holds when one smith's blades are compared with each other. Jūbi blades come from much the same smiths, with a slightly lower centre at the top quarter of Jūyō. Kokuhō sits above all of them.

The sections below set out the evidence.

Two Systems

The five designations belong to two separate systems, each with a small upper tier, plus a pre-war list with no upper tier.

The NBTHK designates Jūyō Tōken and promotes about one in ten of those blades to Tokubetsu Jūyō. The Agency for Cultural Affairs designates Jūyō Bunkazai and raises about one in ten of those to Kokuhō. The two populations hardly overlap. Only one blade in the record holds both a Jūyō and a Jūbun designation.

The comparable pairs are therefore Jūyō with Jūbun, the entry level of each system, and Tokubetsu Jūyō with Kokuhō, the upper tier of each. Asking whether Jūbun outranks Tokubetsu Jūyō compares the entry level of one system with the upper tier of the other.

Who Made Them

The quality of individual blades cannot be compared directly across the two systems. The NBTHK and the Agency write their descriptions to different conventions, and neither body has examined the other's blades. The smiths who made the blades can be compared.

For every blade we measure its smith's standing with the NBTHK: how often that smith's other Jūyō blades are promoted to Tokubetsu Jūyō, relative to the average rate of 9.8%. A smith at 2× has Jūyō blades promoted twice as often as average, and a smith at ½× half as often. The blade being scored is left out of its own smith's count, and only NBTHK decisions are used, so Jūbun or Kokuhō status has no effect on the score.

The Tokubetsu Jūyō and Jūyō Bunkazai curves are nearly identical. The median standing is 1.58× for Tokubetsu Jūyō blades and 1.53× for Jūbun blades, and the two distributions share 93% of their area. The medians differ by 3%, with a smith-clustered bootstrap 95% interval of −16% to +31%, so the data show no difference between them.

Tokubetsu Jūyō, Jūyō Bunkazai and Jūyō Bijutsuhin draw on largely the same smiths; each pair overlaps by 89 to 94%. Jūbi is slightly lower than the other two. Its median of 1.40× is almost exactly Jūyō's 75th percentile (1.41×), against 1.53× for Jūbun and 1.58× for Tokubetsu Jūyō. Ordinary Jūyō sits well below this group and Kokuhō well above it, at 2.19×.

What Was Kept

Since the smiths are the same, the difference has to be in the blades. The most direct test compares blades by the same smith. 136 smiths have blades in both Tokubetsu Jūyō and Jūyō Bunkazai, and Figure 2 compares the two groups of blades for exactly those smiths.

For these smiths, 83% of the Jūbun blades are signed, against 54% of the Tokubetsu Jūyō blades. The Jūbun blades are also more often in original tachi form (58% against 30%) and more often ubu (55% against 40%), and they are recorded as meibutsu four times as often (7.2% against 1.7%).

The pattern repeats from smith to smith. The gaps are largest for smiths whose signed and unsigned works both survive in number, such as Kaneuji, the Osafune masters, the Rai and Awataguchi smiths, and Norishige. Masamune, Sadamune and Gō Yoshihiro, whose surviving works are almost all mumei, show no gap. That is the expected result: if a smith left no signed blades, there are none for either system to prefer.

What Each Step Selects For

To separate these effects we fitted three logistic models, one for each decision in the two systems, using the same blade attributes in each:

• NBTHK promotion: which Jūyō blades become Tokubetsu Jūyō. This model also controls for the Jūyō session, because blades from older sessions have had more chances to be promoted.
• System: which of a smith's blades are Jūbun rather than Tokubetsu Jūyō. This is a conditional logit, so each blade is compared only with blades by the same smith.
• Agency promotion: which Jūbun blades become Kokuhō.

A signature makes no measurable difference to promotion in either system. The odds ratio is 1.08 for Jūyō to Tokubetsu Jūyō and 0.93 for Jūbun to Kokuhō, and neither is distinguishable from 1. Between the systems the effect is large. Among one smith's blades, a signed blade has about five times the odds of being Jūbun rather than Tokubetsu Jūyō, and the same comparison against that smith's ordinary Jūyō blades gives an odds ratio of 5.5. So the difference arises in which blades enter each system, not in which blades each system later promotes.

Maker standing shows the reverse pattern. It predicts promotion within both systems (the odds rise 2.2 times per doubling of standing for Tokubetsu Jūyō and 1.7 times for Kokuhō) but barely separates Tokubetsu Jūyō from Jūbun (1.17, at the edge of significance). Named swords and meibutsu raise the odds in all three models, most strongly for Kokuhō (about 7 times).

The Other Tail

Figure 7 shows how often each designation's blades have each property.

It is sometimes said that Jūbun reaches lower than Tokubetsu Jūyō at the bottom end. In maker standing it does not. Few Jūbun blades are by low-standing smiths (4.5% of rated blades fall below ½×, against 6.9% for Tokubetsu Jūyō), and the Jūbun spread is slightly narrower.

The lower end of Jūbun is made up of a different kind of object. 12% of Jūbun blades have no smith attribution (Tokubetsu Jūyō: 0.5%), and 8% are archaic or votive forms such as ken, ōdachi, naginata and pre-Heian blades (Tokubetsu Jūyō: 1.2%). Shrines hold 35% of Jūbun blades. Many of these objects were designated for their historical or religious importance rather than as works of a known smith. At the top end, Jūbun includes Kokuhō, since every Kokuhō is legally a Jūbun first.

The Claims Tested

The claims that prompted this paper, checked against the record:

What This Does Not Say

This paper does not grade individual blades. Maker standing describes the smith. Two blades by the same smith, one Jūbun and one Tokubetsu Jūyō, may differ a great deal in quality, and nothing in the public record compares individual blades across the two systems.

It does not show that the Agency looks for signatures. It shows that blades entering the Cultural Properties system are more often signed, unshortened, in original form and recorded as named swords than blades entering shinsa, and that neither body's promotion decisions produce this difference. The reasons are a historical question the numbers cannot answer. They may include which collections were surveyed in the 1930s, what owners chose to submit to the NBTHK, and what a heritage designation is meant to protect.

It does not measure provenance in full. The catalogues record ownership too differently to compare. Meibutsu and named-sword status are recorded consistently, so they are used as a partial measure.

Technical Annex

Corpus. Blades only, from the full designation record as of October 2026. Groups: Jūyō not promoted (10,923), Tokubetsu Jūyō (1,198), Jūyō Bijutsuhin (984), Jūyō Bunkazai not also Kokuhō (765) and Kokuhō (102). A blade with several designations counts in each of its groups.

Maker standing. For a blade by smith s: (TJ_s + a·p) / (Jūyō_s + a) ÷ p, where p = 9.8% (the corpus promotion rate) and the pseudo-count a = 10. The blade itself is removed from its smith's counts. School attributions and smiths with no other Jūyō blade are not scored. Distributions are Gaussian kernel estimates on the log₂ scale (bandwidth 0.18), and overlap is the area two density curves share.

Attributes. Signed: the smith's own mei, including orikaeshi-mei and gaku-mei. Gold or lacquer attributions count as unsigned. Ubu: tang recorded as unshortened, among blades whose tang state is recorded. Named: meibutsu or a recorded gō.

Models. Logistic regressions on signed, ubu, tachi form, named, era (Heian or earlier and Nanbokuchō or later, each against Kamakura) and a "den" attribution flag. The promotion models add log₂ maker standing, and the Jūyō to Tokubetsu Jūyō model adds Jūyō session bands. The Jūbun against Tokubetsu Jūyō model is a conditional logit stratified by smith (1,207 blades from smiths with at least one blade in each designation). Pooled models use standard errors clustered by smith. Blades without a recorded tang state are left out of the models.

Robustness. The order of the designations by median maker standing is the same for every shrinkage prior tested:

For Kamakura blades alone, 78% of Jūbun blades are signed against 50% of Tokubetsu Jūyō blades; 52% against 33% are ubu, and 66% against 41% are tachi.