Updates & Corrections

Following P/E and P/B, we have written up earnings per share, book value per share and return on equity, covering the formula and what to watch for when reading the number. They can also be opened straight from the metrics on a stock page.

Tabs that lead to a page of their own, such as dividends and financials, could not be followed on a phone. They now open directly, and the stock information bar has moved below the tabs so the stock you are viewing stays in sight.

For a single reporting period of one stock, segment rows carrying another company's issuer code had found their way in. It affected that one period only, and the rows have been removed. Segments are now classified from the structure of the definitions in the filing rather than from how the names are spelled.

When picking a past period, a missing reporting date meant falling back to an older period, while the denominator used to annualise the figure stayed partly fixed at the number of years requested. Dividing by fewer years than had actually elapsed made growth look higher than it was. It arose most readily on semi-annual reporters, whose reporting months rarely line up with the month requested. The rate is now annualised using the actual gap in months, and no value is shown when the period differs from what was intended.

Records taken from a previous data provider and records from the current feed could both remain for the same period. This affected second-quarter distributions in particular, and the duplicates have been removed.

The immediately preceding period was treated as the prior year, so funds and REITs reporting twice a year were being compared with six months earlier. Comparisons are now made against the same period of the previous year.

For companies in the middle of changing their fiscal year end, the cumulative figures to date were used as a full year's results even though the new fiscal year had not yet ended. As a result, revenue shown in the overview and elsewhere was overstated, and related metrics were not shown correctly. Periods whose fiscal year has not ended are no longer used as annual figures.

The note that stops us fetching the same stock repeatedly was written before the fetch had finished. If the fetch was interrupted the note remained, so returning to the screen would not trigger another attempt and the figures stayed at "--" until the page was reloaded. An interruption is not an error, so nothing was reported and the state was easy to miss. The note is now written only once the fetch completes.

Where the body of a disclosure carried wording to the effect of excluding a split, that took precedence and the split information stated in the filing went unused. A split stated outright now takes precedence.

Distributions for ETFs and investment corporations are read from the text of filings. We check these against the original disclosures periodically, and this correction came out of that review.

In the detailed financial tables, the boundary between plans was counted from the number of columns that reached the screen. The server only returns the years a plan allows, so viewing as a guest shrank the base and left the Standard range one column wide: three years of access looked like two. The boundary now falls in the same place whichever plan you view from.

On the detail page for an earnings report (tanshin), the financial tables showed figures from the annual securities report for the same period instead of the report itself. The process that updates each period with the later annual securities report also affected what the earnings report page displayed. Earnings report pages now show the figures stated in that report.

On the dividend page for Japanese stocks, DOE (dividend on equity) and its year-on-year change showed an older period instead of the latest full fiscal year for some stocks. The latest period was chosen by the order in which the data was stored, not by the order of the periods. DOE is now calculated from the two most recent full fiscal years.

Opening a ticker that does not exist showed "not found" on screen while the server still answered as though the page were fine. To a search engine this looked like a great many thin pages, and Search Console had flagged tens of thousands of them. There were two causes: returning the loading frame first settled the response type, and exceptions were caught so broadly that the "not found" signal was swallowed. Both have been fixed.

Amounts were formatted the same way as Japanese stocks, leaving the currency unstated.

You can switch between five and ten periods. Where the screen is too narrow it stays at five to keep the chart readable.

The metrics listed differed by market, so the same view could not be taken of both. They are now the same.

The two markets had separate views, now combined.

Only codes were listed, so there was no telling which stock was which. Names are now returned.

Choosing a suggestion did not stop the search, so an unnecessary request continued and delayed the display.

The same table treated highlighted rows and negative values differently depending on the market. On US stocks negative values were not shown in red, and a figure could be red while the bar beside it stayed green. The Japanese tables are now the reference, and colour is decided per cell rather than per row.

The screen said only "no period available for calculation", leaving it unclear whether data was missing or the fund had simply not been listed long enough. It now states how many years are required and how long the fund has been listed. The span is counted in months rather than by difference of year: counting by year makes an ETF listed in November 2023 read as three years by 2026, when the true span is two years and nine months.

Loading began before the sign-in check had finished, so a free-plan response came back and stayed. On US earnings history, profitability showed only two periods and shareholder returns came back empty. Reloading cleared it, which made it hard to notice. Loading now waits for the check to complete.

The mechanism that remembers what has already been fetched did not take the subscription into account, so content fetched as a free user before signing in was reused afterwards. It is now re-fetched when the plan changes.

A missing value was always treated as a plan restriction, so periods that hold no data at all were shown with a lock and "available on a higher plan". On one stock, cash flow held values for only two of eighteen periods and the rest all carried a lock. Restricted and absent are now distinguished.

Columns were laid out for periods whose results have not been announced, which read as missing data. Periods with no values are no longer drawn.

For one quarter of one stock, a cash flow line item differed in magnitude from the disclosure, in the way characteristic of a mistaken unit. An error of a single decimal place is hard to spot from the value alone, so it was found by cross-checking figures obtained through different routes.

The date on screen did not refer to the point at which the value was calculated. It now comes from the right source.

Where earnings per share or book value per share was not directly available, the figure on the guide page was blank. Both can be worked back from the price-to-earnings or price-to-book ratio and the share price, and are now derived that way.

When taking in an amendment issued alongside a split, the annual dividend for that period was not carried over and went missing, because the per-share amount changes either side of a split and cannot simply be added together. The figures have been restored.

The logo frame is a circle, but the image was laid across it edge to edge, so a wide logo ran outside the circle and lost its sides. The image now keeps its aspect ratio and is fitted within the circle.

Moving to a page beneath dividends or financials left no indication in the menu of which page was open.

The comparison page lines up revenue, operating income, net income, earnings per share, return on equity, equity ratio, dividend per share and payout ratio — all financials of an operating company. ETFs have none of these, so values left in the financial tables were drawn straight onto the lines and could show things like negative revenue. ETFs are no longer offered for comparison, and the overview and earnings history no longer fetch company data for them. REITs report the same metrics as operating companies and are not excluded.

What an ETF pays investors is a distribution, not a dividend. The wording now matches.

Only five of eight periods fit on screen and the rest ran outside the card. The left axis had a fixed width that was mostly empty space. That width has been reduced and given to the chart, and the minimum width and spacing per column revised. On narrow screens the period labels are also shortened so they do not collide.

The order depended on how the data arrived, so older and newer periods could swap places. Order and totals are now settled on the display side.

Every axis was built to include zero, so an equity ratio moving between 30% and 40% still had an axis starting at zero, pinning the line to the top where its movement could not be read. Dropping zero from bar and area charts, on the other hand, lifts them off the base and makes heights incomparable. Axes are now handled by type: line-only axes get padding around the range of values and round tick marks.

There was no way to tell from the chart what the vertical and horizontal axes showed, or what moving toward either side meant. Axis names and the meaning of each direction are now in the chart, along with reference lines to compare against.

Tick intervals were left to the charting default and came out uneven, with the upper half more finely divided than the lower. They are now placed deliberately. The left axis width also assumed no negative values, so a minus sign could be clipped at the left edge — which makes a value read as its opposite, the more serious of the two faults. Width is now reserved for the sign.

Moving the displayed periods forward or back sometimes left the screen unchanged.

Tooltip backgrounds and text colours differed from chart to chart. They now use one palette, legible in both light and dark display.

On ratio series the percent sign appeared on both the item name and the value. It is now on the value only.

Units such as 億 and 兆 appeared even when viewing in English. They now switch with the display language.

Pressing the legend to hide a series removed a row from the value table below the chart, changing its height and moving the chart and legend with it. The row now keeps its place and is simply made invisible.

On a chart with no bars, the charting layer places the horizontal ticks at the two ends of the drawing area rather than at the centre of each band. The value row is positioned on the assumption of bands, so years and values disagreed in the area chart sections only. The band layout is now preserved.

A series mixing positive and negative values was stacked with the default cumulative behaviour. Stacking a negative investing figure on top of a positive operating figure puts the second bar over the first and hides it. Positive values now extend upward from zero and negative values downward, and the axis is measured to the totals of each side.

The source for earnings per share returned only two full-year periods, leaving a line of two points pushed to the edge of the axis. It now comes from the source that provides thirteen periods, the same as profitability. EPS and BPS also differ by an order of magnitude, so on a shared axis the earnings line sits flat against the floor. The axes are now split left and right, with either amounts or ratios able to take the right-hand side.

Filtering full-year rows out of the balance sheet summary was written on the assumption that a quarter designation was present. Full-year rows carry neither that designation nor a fiscal year, so nothing matched and the result came back empty. The year used in the heading was missing for the same reason, and is now taken from the period dates.

The chart was built assuming a split between current and non-current items. A bank's balance sheet has no such split, so the breakdown never completed and the chart could not be drawn at all. Where the split is unavailable it now falls back to a single ring of liabilities and equity.

Bars were thin, and the period headings and the controls for moving between periods were cramped and awkward to use with a finger. The drawing area has been widened and the bar width, spacing and control positions revised for each screen size: thinner and more spaced out on desktop, thicker on phones.

The display swapped mid-load, so the content appeared to change for an instant. The fallback shown when a logo fails to load was corrected at the same time.

The same line item name appears repeatedly in a financial statement — one filing had six duplicated names among 74 balance sheet rows. That name was being used to tell rows apart, so switching tabs left rows from the previous table mixed in. Rows are now distinguished in a way that does not duplicate.

Some companies disclose a cash flow statement only at the interim and full year. Quarterly figures were preferred unconditionally, so for roughly six months between the full-year release and the next interim, the quarterly figure was shown even though it was the older of the two. The assumption that a quarter is newer than the full year does not hold for these companies. The newer period is now chosen.

Whether a stock's P/E is low cannot be judged without comparing it with others in the same sector, so we now publish the median for each.

If the streak breaks at the first year of dividend data we hold, the company may well have gone longer without a cut. Stating it as a settled figure understates it, so we now qualify it with "or more".

A change is measured against the price on the most recent day earlier than the requested span. Only an upper bound was set, so where prices were missing entirely for a stretch, the name of the window and the period actually compared came apart. On stocks with several years of missing prices around a relisting, the one-year change was being taken against a price from four years back. A lower bound has been added, and where no day falls within it no value is shown. We do not go looking for a substitute date.

The same item can appear in several places in an earnings report, and which one was used could vary by period. A single source is now fixed.

XBRL contains several periods besides the reporting period, including one ending on the filing date. These were not distinguished when identifying the current period, so a span that was not the reporting period could be selected.

Reading the same disclosure again could produce a different annual dividend. The basis is now fixed.

On Japanese stock dividend pages the display language was not passed down to the components, which stayed on the Japanese default. US pages did pass it, so only Japanese stocks were affected. Item names for ETFs and mutual funds had also been written in Japanese directly, and are now translated.

Where trailing twelve-month results were incomplete, a value was pulled from the full-year row instead. That makes a ratio of two figures covering different periods, so the substitution has been dropped and the field left blank.

An earnings report is sometimes filed with the previous year's material attached. Whether something belonged to the current period was judged per document, so prior-year figures inside an attachment could be taken as current. Each figure is now judged individually by the period it belongs to.

Sorting dividend records by fiscal period end looked only at the upper bound, not the lower one. Records arrive for all periods while the chart is limited to the years displayed, so everything older than the range was drawn into the earliest period. On a stock viewed over five years, the oldest bar had absorbed dividends going back to 2013 and stacked 26 segments. The lower bound is now applied and the bar is back to the three segments of interim, year-end and special.

Some companies disclose the top line of the income statement only under their own extension tag. One airline discloses it solely as TotalBusinessRevenueRevOA and reports neither of the standard revenue elements, so operating income, ordinary income, net income and total assets all came through from the same context while revenue alone was missing. Extension tags are now handled, and we confirmed across every stock in production beforehand that the addition catches nothing it should not.

Some companies present their financial statements in US dollars and the cover page of the earnings report in yen. Taking the cover figure leaves revenue in yen while operating income on the same row stays in dollars, so the margin is off by the exchange rate. The size of the discrepancy alone cannot separate a currency mismatch from a figure captured for only one segment. We now identify it by whether several items on the same row are off by the same factor, and periods whose currency does not agree are not used.

ETF earnings reports alone opened in the browser's own viewer in a separate tab, a route that does not carry the signed-in state, so readers were treated as signed out regardless of their plan. They now use the same in-page viewer as everything else. Simply listing the filings loads neither the document nor the viewer, so nothing has become slower.

Where both the main earnings report and supplementary material arrived for the same fiscal year and quarter, the test for which one to present made no distinction between them. Both are filed on the same day, so the date settled nothing either, and the choice came down to a string comparison of document IDs. There were periods where the filings list showed the main report while the earnings history showed the supplement. That it had not occurred since 2017 was because such supplements stopped being produced, not because the test was sound.

EDINET has revenue tags specific to each sector, with suffixes distinguishing a total from its components. RevOA and OR<sector> are component lines of operating revenue and NOI is non-operating revenue, yet all of them had found their way into the candidates for revenue. Guessing from the spelling gets this wrong: in the gas business it mistakes installation revenue for the total, and in the expressway business, land rental revenue. We settled the meaning of each suffix against the 22,617 elements of the official EDINET taxonomy labels and removed the 87 component tags from revenue.

Distributions for ETFs and investment corporations are not disclosed in a machine-readable form the way an operating company's earnings report is, so they are read from the text of filings. The way they are written differs by fund and can change from period to period even within one manager, so an unexpected form can lead us to pick up the wrong figure. The causes found in review included mixed bases either side of a stock split (an interim dividend at the pre-split amount sitting in the same table as a post-split annual figure), confusion between the per-unit amount and the total, changes counted twice when several amendments arrived for one period, and treating the reference date of a quarterly forecast as the fiscal year end on companies that changed their year end. Each was fixed and the affected periods rebuilt.

ROE, ROA, earnings per share and revenue all fell back to a single quarter's figure when no trailing twelve-month value was available, putting three months of data in a slot labelled twelve and setting figures of different periods side by side on one card. The same confusion occurred in the earnings summary, where operating income was off for a great many stocks. The substitution has been removed and the field is left blank instead.

In the timely disclosure and securities report lists, a ten-year cap on viewable years existed only on the display side. The server treats it as unlimited, so the guidance appeared for Premium as well. Unlimited is now expressed properly and the guidance no longer appears where there is no limit.

The year end for securities reports and semi-annual reports was taken as eleven months after the start of the period in the document title. A year in which the fiscal year end changes is not necessarily twelve months. On a stock with a seventeen-month irregular period this fell out of step with the real year end, leaving the semi-annual report alone on a row for a period that does not exist, isolated with everything else blank. The title carries the ending year and month, and that is now used directly.

In a year with five quarters following a change of year end, the fifth quarter's earnings report was placed in the full-year column, making it look as though full-year results had been announced when they had not. The quarter number in the document title now takes precedence and places it correctly, allowing for variation in full-width digits and spacing.

The detailed financial tables hard-coded the closing month of each quarter as June, September, December and March. Those values assume a March year end, so a December reporter had periods created that do not exist, left as columns of empty rows. The reporting month is now read from the actual full-year data and each quarter worked back from it. Where the month cannot be determined, no column is created — showing nothing is more correct than inventing a period.

The dividend chart sorted on the year alone. A company that changes its year end has two periods in the same year, so the year alone does not settle the order and the periods fell in whatever order they arrived. Changing the number of years displayed brought one of them into the series, and from that moment bars and years were a period out. Sorting now takes the month into account.

Dividend records were grouped by a fiscal year inferred from the reporting month, and that month was decided from the first period and applied to all of them, which breaks for a company that changes its year end. The final March-year period and the period after moving to a December year end collapsed into one fiscal year, mixing four components into a single bar. Because the mixing changed with the number of years displayed, it showed up as one period looking oddly depressed. Each record is now matched to its period individually.

They sat on a separate screen from the earnings reports, so comparing filings for the same period meant moving back and forth. They are now in one table.

On some stocks an item that is a component of operating revenue, rather than the total, was being taken as revenue. Components are no longer treated as totals.

The window over which distributions were totalled differed between processing routes: one set the window by reporting period, the other used a fixed one-year window, so for investment corporations reporting twice a year two distributions were picked up as one. Measured on a stratified sample of 106 periods across 5 stocks, 103 (97%) changed, with a median ratio of 1.990. The window is now defined in one place.

When a date moved, if the new date was already occupied by a row provisionally registered through another route, a uniqueness constraint stopped the import outright. The destination is now cleared first. Where the row at the destination came from the same route, the move is skipped with a warning rather than overwriting it, so that no data is lost.

Some issuers disclose the equity ratio as a percentage and others as a decimal. Previously anything with an absolute value of 1 or more was treated as a percentage and divided by 100, but neither the range nor the sign settles which notation is in use: -3.926 reads equally as "-392.6% as a decimal" or "-3.926% as a percentage". Having measured all 147,628 rows, we now compare against the value derived from equity and total assets on the same row and decide only whether the figure is out by a factor of 100.

Annualising a quarter's net income for ROE and ROA divides by the number of days in that period. When choosing that figure, an instant context from the same filing (FilingDateInstant) could be taken instead of the duration context (CurrentYearDuration). Rows from one filing share a disclosure date, so which of the two was picked was left effectively to chance. Only duration contexts are now used.

EDINET has revenue tags specific to each sector, with suffixes distinguishing a total from its components. Component tags had found their way into the candidates for revenue, which could make the level look discontinuous from stock to stock. Guessing from the spelling gets this wrong: in the gas business it mistakes installation revenue for the total, and in the expressway business, land rental revenue. We settled the meaning of each suffix against the 22,617 elements of the official EDINET taxonomy labels and removed the components from revenue.

The XBRL of an earnings report marks an absent value with the attribute xsi:nil="true", which is what "—" on screen represents. Where results were disclosed in that form, a value held through another route was applied and a figure shown. The cause was deciding the value without looking at the disclosure for that period itself. Periods with no entry now hold no value.

Cards serving the same purpose had different border colours and spacing from page to page. The period badge at the top right, which showed the calendar year rather than the reporting period in places, now shows the fiscal year throughout.

The step that annualises quarterly results must not be applied to a full-year company forecast. The material needed to tell a full-year row from a cumulative quarterly one was not being passed along some routes, and in its absence the default was to assume full year. On stocks whose full-year slot was filled by a cumulative second- or third-quarter row, forward P/E, forward PSR and forward PEG were stored as annualised values. The information needed to tell them apart is now carried through.

The summary page of an earnings report lists consolidated and non-consolidated total assets separately. In periods where the full financial statements were not available, we missed the consolidated figure and used the non-consolidated one that remained, which led to impossible figures such as equity exceeding total assets. We fixed this to use the consolidated figure whenever it exists, and rebuilt the data.

Headings written with era-style year notation, or mixing digits and spaces, could not be parsed and the filing did not appear in the list. The variations are now read correctly.

The earnings summary derived its year from the calendar year in which the quarter itself ends. The first quarter of a March reporter (April to June) came out as 2026, while the earnings history on the same page returned the fiscal year ending March 2027, putting two years against one set of results. The value from the side that holds the fiscal year is now used. We do not parse the document title string.

One earnings report can contain the consolidated operating margin alongside the non-consolidated operating income. The tag for the ratio had been registered as an alias of operating income, so a ratio landed in a field for an amount, and every metric calculated from operating income broke along with it. Ratios are now filed under ratios.

Suffixes change the meaning of an EDINET taxonomy element. OrdinaryIncomeLossSummaryOfBusinessResults is ordinary income while OrdinaryIncomeSummaryOfBusinessResults is ordinary revenue, the top line — but the latter had been registered as an alias of ordinary income, so for insurers ordinary revenue landed in the ordinary income field and displaced the real consolidated figure. Revenue has been moved to the revenue field, and the equivalent TDnet tags for the insurance sector now resolve to revenue as well.

The line colour matched across both markets, but the area fill was darker on the US side and its lower edge was not fully transparent. It now varies its depth with a rise or fall and fades out at the base, the same as Japanese stocks.

Book value per share is disclosed to two decimal places, so the difference between the amount worked back from it and net assets comes out slightly positive even for an ordinary stock with no preferred shares. That difference was taken as a real deduction and carried into the next period, accumulating a tiny deduction across quarters with no disclosure. It is now carried forward only if it is of a meaningful size.

ETF sector weightings were totalled on the display side from the list of holdings. That list is narrowed to the top ten depending on plan, so left as it was, the result is a sector split of ten holdings rather than of the fund. We switched to the figure that carries the weighting for the whole fund. The quarterly chart on the earnings page now also reads from data that does not vary by plan: that chart is meant to show every period regardless of plan, and its source had not kept up.

They were being totalled on the display side from the list of holdings, which could disagree with the published weightings. The published figures are now used as they are.

When the free plan changed to return only the top holdings, the count that reached the screen was presented as "N disclosed". A stock with 30 holdings disclosed was showing "10". The count now comes from a source that is not subject to the plan limit.

Stocks with no news and no filings at all still had empty pages open to search. On a sample of US stocks, 98.8% had no news and 58.3% had no SEC filings, so empty pages made up the great majority. Pages holding no data are no longer offered to search.

This helps the content of those pages appear correctly in search results.

ETFs have no financial statements of the kind an operating company files, yet links to them were listed in the phone menu. They are now shown according to the security type.

On Japanese ETF pages, menu links to the income statement, balance sheet and cash flow statement appeared briefly while the page loaded and then disappeared. ETFs do not have financial statements like operating companies, so the ETF menu is now shown from the start. Holdings and performance content is also displayed as soon as the page opens.