The Cap That Runs Out First

Ricoh (7752) filed its buyback status report for the July reporting month on 7 August 2026. In the line printed 自己株式取得の進捗状況(％) — acquisition progress, in percent — the company put two numbers: 31.23 and 42.47. Same programme, same reporting month, same board resolution of 12 May 2026. One is measured on the share count, the other on the amount paid. They are 11.24 points apart, and the line carries both.

The document is the 自己株券買付状況報告書, the status report on a company's purchase of its own shares. EDINET indexes it as document type 220, and Ricoh's cover cites 金融商品取引法第24条の6第1項. It runs to three numbered sections — 1【取得状況】, then 2【処理状況】 for disposals and 3【保有状況】 for the holding at month end — and this piece stays inside the first. Ricoh's copy runs to three pages, with the acquisition section on the second and the other two on the third. Under (2)【取締役会決議による取得の状況】 the table is headed 区分, 株式数(株), 価額の総額(円), so each of its rows prints twice, once in shares and once in yen: 取締役会(2026年5月12日)での決議状況 (取得期間 2026年5月13日〜2026年11月30日), the terms of the board resolution; 報告月における取得自己株式(取得日), which takes a row per date on which shares were acquired; 計, the month's total; 報告月末現在の累計取得自己株式, the cumulative position at the end of the reporting month; and then the progress line. Above the table, (1)【株主総会決議による取得の状況】 reads 該当事項はありません. How often the report must be filed, and for how long after a resolution, is a question for the rule rather than for the printing, and this piece stays with the printing.

Ricoh's resolution of 12 May 2026 authorised up to 23,000,000 shares and up to ¥25,000,000,000 (US$150 million), over a window running from 13 May to 30 November 2026. In the July reporting month the company acquired 2,766,100 shares for ¥4,161,554,500 (US$25.41 million), on twenty dated rows. At the end of the reporting month the cumulative position stood at 7,183,600 shares and ¥10,617,481,650 (US$64.82 million). The two printed percentages follow from the lines above them: 7,183,600 against a ceiling of 23,000,000 is 31.23 percent, and ¥10,617,481,650 (US$64.82 million) against ¥25,000,000,000 (US$150 million) is 42.47. Each is arithmetic on numbers printed a few lines up. Neither is an estimate.

There are two percentages because the authorisation has two ceilings, and a purchase draws on both at once: each share subtracts one from the count and some yen from the amount. The two columns advance together only when the buying has drawn on the two ceilings in the same proportion as the resolution set them — 23,000,000 shares against ¥25,000,000,000 (US$150 million). Draw harder on the amount and the amount column runs ahead of the count. Draw harder on the count and the count runs ahead. Neither ceiling is the binding one until one of them is reached, and the report carries both figures, so which one is running down faster can be read off the page while the window is still open.

Ricoh's two figures are 42.47 and 31.23, and their quotient is 1.36. That single number is the whole of the separation: the amount has been drawn down 1.36 times as fast as the count. Nothing else is needed to read the pair, and nothing else can be got out of it — the form prints the two percentages and the totals behind them, and the ratio is arithmetic on the two figures printed.

Across the five business days from 3 to 7 August 2026, 318 of these reports were filed with EDINET. A machine pass over the progress line took two percentages out of 287 of them. The pass had been told to accept only figures carrying a decimal point or a percent sign, which is what keeps a table full of dates and note numbers from being read as percentages, and two filings print a pair it could not see on those terms. Tabio (2668) prints 100 beside 85.08; the pass took the second and left the first. Prestige International (4290) prints 0 and 0, and it took neither. Read off the page, those two join the rest: 289 filings print both figures. Of the remaining 29, 27 print a dash in that line rather than a percentage, one prints a single figure, and in one the acquisition section reads 該当事項はありません. All 318 are accounted for.

The line belongs to the resolution, not to the report. A company running two board resolutions at once can print the table twice and the progress line twice, and ten of the 318 do. In eight of the ten the second line prints a pair; in two it prints a dash. The pass took the first pair in each, so the week has 297 pairs on its pages and 289 in every count below, which are one pair per filing. Where a number counts what a machine took rather than what a filing printed, the sentence says so.

Three of the 289 are investment corporations, which print the same line in different words. Over columns headed 投資口数（口） rather than 株式数（株）, the line reads 自己投資口取得の進捗状況（％）, and the resolution row is 役員会 rather than 取締役会: ＫＤＸ不動産投資法人 76.7 and 100.0, 大和証券オフィス投資法人 31.4 and 50.1, GLP投資法人 39.2 and 40.7. Two figures, one field, a different noun.

In 269 of the 289, the two numbers differ by at least 0.005 points, the tolerance used here for calling them the same. That is 93 percent. The direction is lopsided: in 202 the yen figure is ahead, in 67 the share figure is ahead. Twenty are the same. The rounding is left as each company printed it, one decimal in some and two in others, which means agreement is measured on the companies' own terms rather than on a recomputed basis.

Agreement is a narrow condition. It requires the cumulative purchases to have drawn on the share ceiling and the amount ceiling in exactly the proportion the resolution set between them — not approximately, but at whatever precision the filing prints. That happened 20 times out of the 289, about 7 percent. Seven of the 20 print zero on both sides: zero stands in the same relation to either ceiling, so a resolution with nothing yet bought under it agrees with itself. Three of the seven cover a month that closed before the acquisition window opened, among them Prestige International, whose report covers 28 to 31 July while the window opened by its 28 July resolution runs from 3 August. That leaves 13 in which the equal pair sits on purchases made, 13 out of 289.

At the wide end, the yen side runs far ahead. KOKUSAI ELECTRIC (6525) printed 37.73 on the count and 99.99 on the amount, a separation of 62.26 points. Cybozu (4776) printed 41.27 and 99.99, 58.72 points apart. Niterra (5334) — 日本特殊陶業株式会社 on the cover, Niterra Co., Ltd. in its 英訳名 field — printed 53.0 and 100.0, a gap of 47.0. In each, it is the amount-based figure that sits at or near 100 while the count is well short of it. Two of the three windows closed with the reporting month: KOKUSAI ELECTRIC's ran 14 May to 31 July 2026, Cybozu's 15 May to 31 July 2026. Niterra's runs to 31 October 2026.

A yen column at 99.99 or 100.0 while the count reads between roughly 38 and 53 percent is a specific arrangement: as printed, the amount column stands at or within a hundredth of a point of 100 and the count column does not. What the pair alone cannot say is why. A 62.26-point separation is consistent with a resolution whose two ceilings sit far apart in ratio, and equally consistent with shares costing more than that ratio allowed for. Each of the three prints its own two ceilings in the resolution row of the same table, so the split could be made from the document; this piece does not make it, because making it means dividing a yen amount by a share count.

Sixty-seven filings run the other way, just under a quarter of the 269. Aidma Holdings (7373) printed 100.00 and 50.31: the count is at its ceiling, 450,000 shares against an authorisation of 450,000, while the amount column stands at half, and the note under the same table records that acquisition under the 10 July board resolution ended with the purchase of 30 July. IMAGE MAGIC (7793) printed 100.0 and 66.2, on a window that closed with the reporting month. Tabio, of the bare 100, printed 85.08 beside it.

The acquisition table takes a row per date, and the dates are where the reading thins out. In 88 of the 318 the scan of those rows came back with no dated rows at all; in six more it returned no count. Tabio is one of the 88, and its table prints a dash where the dated rows would be, with a progress line reading 100 and 85.08 below it. Where the scan did return rows, in 224 filings, the median is 20. So in a single filing week the acquisition table runs to twenty-odd rows where shares were bought on twenty-odd days and prints a dash where none were bought at all, and in a smaller residue the scan returned no dated rows while still carrying a monthly figure. In some of those the acquisition table's 計 row prints only dashes and the figure the extract holds is the total printed below it in ２【処理状況】; in others the acquisition table prints the month's total itself, on a second resolution's table, or over dates set in a form the scan's date pattern declines. That residue counts the extract, not the page. The week is not a single reporting month either: among the periods printed on these covers are April 2026, December 2025 and July 2025.

Between the extremes the pattern holds without being dramatic. Uyemura (4966) resolved on up to 260,000 shares and ¥4,000,000,000 (US$24 million); in the reported month it took 24,800 shares for ¥566,888,000 (US$3.46 million), and printed 65.31 and 95.80, a gap of 30.49 points. Nippon Television Holdings (9404) resolved on up to 5,200,000 shares and ¥12,000,000,000 (US$73 million); the reported month brought 558,400 shares for ¥1,614,856,737 (US$9.86 million), and the printed pair is 72.62 and 91.21, 18.59 points apart.

Yamazen (8051), with ceilings of 13,000,000 shares and ¥15,000,000,000 (US$92 million), acquired 1,002,000 shares for ¥1,745,805,084 (US$10.66 million) in the reported month and printed 47.1 and 66.9, a separation of 19.8 points: the count near half way, the amount about 1.4 times further along. Osaka Gas (9532), with ceilings of 28,000,000 shares and ¥80,000,000,000 (US$490 million), took 863,400 shares for ¥4,747,910,804 (US$28.99 million) and printed 11.90 and 23.09, 11.19 points apart, the count a ninth of the way through and the amount very nearly twice that.

The near-ties look no different in construction. NH Foods (2282), authorised for 7,000,000 shares and ¥40,000,000,000 (US$240 million), acquired 1,193,600 shares for ¥7,456,356,400 (US$45.52 million) and printed 19.94 and 21.63, 1.69 points apart. Amada (6113), authorised for 25,000,000 shares and ¥50,000,000,000 (US$310 million), took 2,218,800 shares for ¥6,488,281,850 (US$39.61 million) and printed 9.71 and 14.28, a gap of 4.57. Nanyo (7417), on ceilings of 700,000 shares and ¥1,100,000,000 (US$6.7 million), took 71,700 shares for ¥128,659,200 (US$0.79 million) and printed 55.9 and 54.5, the count ahead by 1.4.

The same reading works one report at a time. Uyemura prints 65.31 and 95.80, a ratio of 1.47. Osaka Gas prints 11.90 and 23.09, a ratio of 1.94 — the amount running down at nearly twice the rate of the count. Nanyo prints 55.9 and 54.5, a ratio of 0.97, the only direction in which the count is the one running out first. The three ratios come from six printed percentages and no other input, and they are cumulative: they describe the programme to date, not the month just reported.

Read across the seven, the separation does not follow the progress. The gaps run from 1.4 points to 30.49, and they do not sort with the counts they sit beside: a pair a tenth of the way through its share ceiling shows a wider separation than a pair more than half way through. The two columns have to be read together. One does not fix the other.

Two things set the separation: where the resolution placed its two ceilings relative to one another, and what the shares cost while the buying happened. Both inputs are printed — the ceilings in the resolution row, the shares and amounts in 報告月末現在の累計取得自己株式 — but the pair of percentages by itself does not separate them, and separating them is the division this piece does not do. The point is not the quotient. The point is that in 289 of one week's 318, the same field carried both figures, and on ten of those pages the line was printed twice: eight times with a second pair, twice with a dash.

Ricoh's window runs to 30 November 2026. In the report for July the count reads 31.23 and the amount 42.47, on one line, in that order.

Currency conversion: yen amounts are converted at ¥163.8 = US$1, the rate for 24 July 2026 taken from Frankfurter's default series, which blends the central-bank sources it tracks rather than quoting any one of them alone; the value returned for that date is 163.82. The single fixed rate is applied to every figure and every period, including the figures restated a second time, with no adjustment for rates prevailing on other dates; that is a stated convention, not a claim about the rate on the days shares were bought. Dollar figures carry no more significant figures than the yen figures they translate: a ceiling printed as ¥25,000,000,000 becomes US$150 million, while an amount printed to the yen, ¥1,745,805,084, becomes US$10.66 million. Percentages, point differences, quotients and counts of companies, filings, lines or pairs are stated as printed or as counted, and are not converted.