This site contains affiliate links. We may earn a commission at no extra cost to you.

Bad Ends and Happy Ends: Judging How a Doujinshi Finishes Before You Buy

You have found a work you like the look of. The art suits you, the price is fine, the premise is the one you were after. The question that stops your hand is not whether you can read the Japanese. It is whether the last five pages are going to leave you feeling the way you wanted to feel.

Two works can share a premise, a tag row, a page count and a price, and finish in opposite directions. One restores what was lost, the other does not, and a reader who wanted the first will not be consoled by the artwork of the second.

The storefront will not answer this for you, and that is structural rather than an oversight. This guide covers why the ending is missing from the product page, the tags that genuinely carry it, the Japanese words in titles and blurbs that give it away, what page count and format make possible, what the preview images can and cannot show, and one limit of our own readability figures that matters specifically to anyone buying a work for its ending.

Why the product page does not tell you how it ends

Every other decision you make about a work is supported by a field. Price, release date, file size, format, circle — all recorded, all structured, all reliable. The outcome is not among them, for four reasons that stack.

The description is advertising, written by the person selling. A circle writes a blurb to make you want the work. Blurbs therefore describe the setup and stop: who these people are, what has begun to go wrong, what the situation is on page ten. The last act is exactly the part a seller has no reason to disclose. Read the blurb for premise and register; do not read it for outcome, because outcome is not what it was written to carry.

Revealing the ending would damage the product. This is the same convention that governs every storefront selling narrative anywhere. The information is deliberately withheld, and no amount of careful reading will extract something the writer chose not to write.

There is no field for it. Storefront metadata is built for things that are true on upload and never change. “How it ends” is a judgement about content, not a property of the file, and nothing in the structured data was designed to hold it.

Machine translation cannot recover it. This is the trap for overseas readers specifically. Run the description through a translator and you get an accurate English rendering of a blurb that never mentioned the ending. The output reads like complete information because it is fluent. Fluency is not coverage — translation cannot return what was never in the source. If you have been treating a translated blurb as the whole story, that is the assumption to drop first.

What is left is inference, and inference here works better than you would expect, because these storefronts carry one piece of vocabulary built precisely for this problem.

The tags that actually carry the ending

Tags are the most reliable part of a product page, because the circle picks them from a fixed list rather than writing them as prose. Most of them describe content. A small number describe outcome and register, and those are the ones to read first.

逆転無し is the single most useful label in the genre, and the English rendering hides it. The storefront’s own English catalogue returns it as No Reverse, which reads like a note about plot twists and is in fact a commitment about the last act: the situation does not get undone. It tells you the ending without spoiling a single event, which is unusually well-designed tagging. Netorare, netori and BSS covers where it sits among the other story tags.

鬱, returned in English as Depression, is a warning rather than a flavour note. It is worth separating from the tag above. No Reverse is about events; Depression is about how the work feels while those events happen. A work can carry either without the other, and a work carrying both is telling you plainly what it is.

Two entries from our own catalogue show the difference in practice. RJ01681936 carries a reverse-NTR tag alongside No Reverse: the roles are inverted and the situation stays inverted, which is a specific shape you can commit to before buying. RJ01663763 carries Depression next to its relationship tag, over 63 pages — nothing in the tag row promises a downbeat ending, but the register is declared, and no amount of attractive artwork changes a declared register. (What the RJ number is explains the code itself.)

Now the rule that matters more than any individual tag:

Presence is informative. Absence is not. A work tagged No Reverse is making a commitment. A work without that tag is making no commitment at all — it may reverse, it may not, or the circle may simply not have thought the label worth adding. Tags are entered once at upload and rarely revised, and circles tag what feels salient to them rather than filling every applicable box. Reading a missing tag as a promise is the most common way this method fails.

The practical consequence is asymmetric and useful: you can rule works out reliably, and you cannot rule them in. If a downbeat finish is what you are after, the tags will get you there. If you need the opposite, the tags only narrow the field, and the rest of this article is how you close the gap. DLsite tags in English covers the wider vocabulary these sit inside.

Words in the title and blurb that give it away

Circles do sometimes state the ending — usually because they know their readers care and want the right ones to buy. When they do, it is a short set of terms that recurs. These appear in titles, in the description body, and occasionally in a bracketed note at the very end of a blurb.

How to search for these without reading Japanese, in about thirty seconds:

1. Copy a string from the table above — you are copying characters, not reading them
2. Open the product page in your browser and press the find-in-page shortcut (Ctrl+F, or Cmd+F on a Mac)
3. Paste and look at the hit count. A hit tells you the word is on the page and roughly where it sits
4. Repeat for 前編, 完結 and 逆転 before anything else. Those three decide the most

前編 in a title means the ending is not in this file. Buying it alone gets you a setup and a stop. The other half usually exists and is listed by the same circle, so the fix is to search the circle’s page for 後編 before you buy — but the marker is two characters at the end of a title, and a translated title will often not preserve it.

メリーバッドエンド has no clean English equivalent, and flattening it costs money. It describes a finish that is genuinely happy from inside one character’s position and clearly not from outside it. The decision rule is simple: if you are buying because you want a specific feeling at the end, treat this term as a mismatch with both a clean resolution and a clean collapse, and machine translation will not warn you, because it usually renders the phrase as one or the other.

Paste single terms into a translator, not paragraphs. Translation fails on these pages because of genre convention, not vocabulary. One term in isolation usually comes back right; a paragraph of marketing copy turns the convention-laden parts to mush. The Japanese words you will meet on a product page covers the rest of the vocabulary that turns up in the same places.

What length, format and series membership make possible

None of these tell you the ending. All of them tell you what endings are structurally possible, which narrows things more than it sounds.

A reversal needs pages. Undoing a situation requires three movements: the situation established, the loss landing, and a response playing out. Three movements do not fit into twenty pages alongside everything else the work is doing. Short works can still resolve — but if you are buying a 20-page work hoping for a full reversal arc, the arithmetic is against you, and the tags will not warn you about arithmetic. Judging real length is its own skill, because the number on the listing is sometimes the trial version’s page count rather than the work’s: how many pages a doujinshi is covers that trap.

CG collections often have no ending to have. A set of illustrations may carry a sequence and an implied arc without ever committing to a resolution, because resolution is a property of narrative and the format is not always narrative. If the ending matters to you, check the format before the tags. CG collections and manga doujinshi sets out how the two differ in practice.

A series entry does not have to finish anything. The ending you are asking about may sit two releases away, or may not exist yet. 完結 in a listing is the circle saying the arc closes here. Its absence, again, promises nothing. Doujinshi series and reading order covers how to find where a series starts when the numbering is not in the title.

Anthologies and compilations have several endings. A bundle carries the union of everything inside it, so a tag row on a compilation describes the collection rather than any one story in it. Reading outcome tags off a bundle produces confident nonsense — a No Reverse tag there may belong to one story out of eight.

What the preview images can and cannot show

The samples are the best evidence on the page, and they are systematically unable to answer this particular question.

Sample pages come from the front of the work. They are chosen to sell it, which means they are chosen from the setup. You will almost never be shown the ending, and when a late page does appear it is there as a hook rather than as disclosure.

What the previews genuinely do show:

• The opening register. Whether the work is playing this comically, coldly or heavily. This is evidence about the work’s default setting, not a guarantee about the last act
• How much of the story is being carried by dialogue. A work that talks constantly is a work whose ending you will need to read rather than look at
• Where the free sample was cut. Circles often cut at a moment of change, so the position of the cut tells you roughly how much of the work is setup — a sample that ends before anything has happened suggests a long runway and a compressed finish

What they cannot show:

• The last act. By design
• Whether the register holds. The samples come from the one part of the work that has no reason to change, so they are silent on exactly the stretch where a change would matter

Preview pages before you buy covers how to read the samples for everything else they do carry.

Where our readability figure stops helping

This is the limitation to know about if the ending is the reason you are buying, and it follows from how the measurement works rather than from anything about the works themselves.

Every readability figure on this site is measured by OCR from the free sample images — four pages per work — and what comes back is the amount of text, not its meaning: how much of the page the text covers, roughly how many characters it holds, and how confident the OCR was. The reason line published with each work states what that came to, in characters per page. RJ01681936 above measures around 80 characters per page and is on the heavy side; RJ01663763 measures around 17 and is light.

The confidence number is the interesting one, because the OCR failing is itself information. Typeset text inside speech bubbles comes back with high confidence. Handwritten monologue and drawn-in text come back with low confidence, or are not picked up as text at all. That is the difference between our two heavier bands: plenty of text that OCR reads cleanly means a translation app will cope, and plenty of text that OCR struggles with means it will not.

Two facts about that method collide on this question. The samples come from the front of the work, and the front is where the artwork does the establishing. Monologue, epilogue and the occasional page of the circle’s own text after the story stops all arrive at the back. So for a work bought specifically for its ending, the published figure is a floor rather than an average, and it is a floor on the stretch that decides everything.

To be exact about the limit: we are not claiming to have measured anyone’s last five pages, because the samples do not contain them. We are saying the measurement covers the front, the text at the back is a different kind of text, and you should plan for that gap rather than assume it away.

Three things follow, and all three are actionable.

1. If the ending is what you are buying for, buy from the lower text bands. Not because those works are better, but because the part you care about is the part most likely to defeat you elsewhere
2. Spend your translation effort at the back. If you hand-translate anything — and you may not need to translate anything else — translate the last five pages. That is where the answer is, and it is a small enough job to actually finish
3. Expect your phone’s camera translation to work worse at the end than it did at the start, because handwritten monologue is precisely the text that returns low confidence

Works already carrying an English version remove the problem entirely, ending included: works with an English version. Works measured as carrying almost no text remove it a different way — if there is no monologue, there is nothing to miss — though you then have to read the outcome from the artwork alone: minimal text. The OCR OK band is where a translation app is enough for the front of the work, and the text-heavy band is where a good ending is most likely to be lost on you. The 174 works here sort into English, minimal text, OCR OK and text-heavy — 14, 66, 50 and 44 works, with the counts moving as works are added.

The check to run before you buy

In order, and none of it needs Japanese:

1. Read the story tags first. Look specifically for No Reverse and Depression. Presence commits the circle to something; absence commits them to nothing
2. Find-in-page for 前編 and 完結. Two searches, five seconds, and they catch the most expensive mistake — buying half a story
3. Check the format. A CG collection may have no ending to have, whatever the premise suggested
4. Check the page count against what you want. Twenty pages rarely has room for a full reversal arc
5. Read the samples for register and for where they cut, and accept that they will not show you the finish
6. Check the text band last, and if you are buying for the ending, treat the figure as a floor and budget translation effort for the final pages

The honest summary is that none of this gives you certainty, and anyone offering certainty about the ending of an untranslated work they are also selling you should be read carefully. What the method gives you is the ability to rule out the wrong purchase — to see, before you spend, that a work has promised not to resolve, or that the file in front of you is only the first half, or that the outcome you care about is written in the densest Japanese in the book.

That is usually enough. A work that was clearly labelled as something you did not want is a different problem from a bad work, and it is the one this vocabulary solves.

The Japanese terms above are the ones circles and storefronts actually use; English tag labels are the storefront’s own renderings, and their exact wording varies between listings and over time — 逆転無し is returned as No Reverse in the English catalogue and appears elsewhere as No Reversal. No storefront publishes a field for how a work ends, so everything here is inference from labels, length, format and samples rather than a guarantee. Readability figures are measured from sample pages, which come from the front of a work, and counts are read live from the catalogue and change as works are added.