What a character sale actually costs — and which parts a registry can remove
Ask a creator why they have never sold a character they no longer use, and the answer is rarely “nobody wanted it”. It is that the sale was not worth the trouble. The trouble has a name in economics — transaction costs — and it comes in three buckets: finding a counterparty, agreeing terms, and making sure the deal is actually performed. A registry cannot change what a character is worth. It can change how much of that value is eaten on the way to a deal.
The lemon problem comes first
A buyer looking at a character has no way to tell a five-year-old design with a documented history from something generated last night. Facing that, the rational move is to assume the worse case and discount accordingly. Sellers with genuinely good assets find the offers insulting and walk away, which leaves a market of exactly the assets the buyers feared. This is Akerlof’s used-car argument, and it describes the current market for AI-adjacent characters almost too well.
The escape is not more marketing. It is cheap, verifiable disclosure: something a good seller can produce and a bad one cannot. A registry entry that has existed since a given date, carries every intermediate version, and can be checked against a public chain without trusting the platform is exactly that kind of signal.
Where the money actually goes
Splitting a hypothetical sale into the classic buckets makes it clear which costs are addressable by design and which are not.
| Cost | What it looks like in practice | What the registry does |
|---|---|---|
| Search | There is no place where characters with clean provenance are listed, so buyers ask around and sellers wait to be discovered. | A market where every listing carries the same comparable record. Partially solved — a marketplace is only as good as its supply. |
| Verification | Who made this? Since when? Has the specification been rewritten? Was it sold to someone else last month? Each answer is a private message and a leap of faith. | Registered date, full version history, works bound to the character, and a public check that reads the chain directly. This is the bucket the design attacks hardest. |
| Definition of the object | “The character” means one thing to the seller and another to the buyer; the argument surfaces after payment, about a file nobody listed. | A specification with a fingerprint, plus a file manifest. The contract points at a number instead of describing an object in prose. |
| Bargaining | Either an unpriced lawyer or a copy-pasted template neither side really reads. | Two parameterised templates, optional clauses as switches, and a redline that shows every departure from standard. Both sides sign the same fingerprint. |
| Payment | Whoever moves first carries the risk. | Nothing yet. The platform does not hold funds, and escrow is the largest remaining cost on this list. |
| Delivery | A zip file, a promise that no copy was kept, and no way to check either. | The pack is moved rather than copied, the seller’s access ends, and the file list is compared against the snapshot taken at signature. |
| Enforcement | If it goes wrong, both sides start by arguing about what was agreed and what was delivered. | An acceptance window with a machine check, a dispute that freezes the deal, a reversal that is possible and recorded, and a deal record either side can hand to a lawyer. |
Why verification is the bucket worth attacking
Because it is the one that scales badly. Bargaining costs fall once, when a good template exists. Delivery costs fall once, when someone writes the procedure. Verification costs recur on every deal, for every buyer, and they are paid in the currency buyers hate most: uncertainty about whether they are about to be the person who finds out the character was sold twice.
Taiwan makes the point cleanly. Copyright registration was abolished in 1998, so there is no official record of who holds what — nothing to search before buying, nothing to update after. The United States keeps a register, but it registers works and takes months, and a character is not a work. Both systems leave the same hole in the same place, and it is a hole a private registry can fill without pretending to be a public one.
The part that is legally interesting
US courts protect a character in itself only where it is distinctive enough; one prong of the test asks whether the character is “sufficiently delineated to be recognizable as the same character whenever it appears”. Read that with a creator’s eyes and it is not a legal abstraction at all — it is the daily work of keeping one character consistent across platforms, which is what the evidence checks on this registry measure in the first place.
The AI angle points the same way. Where purely machine-generated material is unprotected, what carries a claim is the human contribution: the decisions, the versions, the alternatives that were rejected. A provenance log kept from the first day costs almost nothing and is the only form that evidence can take, because it cannot be reconstructed afterwards.
Not legal advice, and deliberately general: the doctrines differ by country and the facts differ by character.
What stays expensive
- Simultaneous exchange. Without escrow, someone still goes first. Reporting payment and confirming receipt narrow the window and create a record, but they do not remove the risk.
- Authorship. A registry records what a seller declares. It cannot tell you they drew it, and any registry claiming otherwise is selling comfort.
- Cross-border enforcement. Choosing a governing law is cheap; suing across a border over a four-figure sale is not. Most of the protection has to come from the procedure, not the forum clause.
- Cold start. Search costs fall only when there is something to search. That is a supply problem, and no amount of design solves it.
The honest summary
A registry does not make a character more valuable, and it does not turn evidence into title. What it does is remove the reasons a sale falls apart before it starts: the buyer who cannot check, the seller who cannot prove, the contract nobody wants to pay a lawyer to write, and the delivery neither side can confirm. Those are the four costs standing between a character sitting in a folder and a character that changes hands.
賣一隻角色實際要花多少，登錄簿又能省掉哪些
問一位創作者，為什麼從來沒賣過那隻早就不用的角色，答案很少是「沒人要」，而是這筆交易不值得那些麻煩。那些麻煩在經濟學裡有名字，叫交易成本，分成三桶：找到對方、談好條件、確保交易真的被履行。登錄簿改變不了一隻角色值多少，能改變的是那個價值在走到成交之前，被吃掉多少。
先講檸檬問題
買家看著一隻角色，分不出一個有五年紀錄的設計，和一個昨晚剛生成的東西。遇到這種情況，理性的做法是假設最壞的狀況、照著殺價。真的握有好資產的賣家覺得出價是侮辱，轉頭就走，留下來的市場正好只剩買家擔心的那種資產。這是 Akerlof 的中古車論證，拿來描述現在的 AI 角色市場，準確得有點過頭。
出路不是更多行銷，而是便宜、可驗證的揭露：好賣家拿得出來、壞賣家拿不出來的東西。一筆從某一天起就存在、帶著每一個中間版本、而且不必信任平台就能對照公開鏈核對的登錄，正是這種訊號。
錢實際花在哪
把一筆假想的交易拆進經典的幾桶，就看得出哪些成本可以靠設計處理，哪些不行。
| 成本 | 實務上長什麼樣 | 登錄簿做了什麼 |
|---|---|---|
| 搜尋 | 沒有一個地方列出來歷乾淨的角色，所以買家到處問，賣家等著被發現。 | 一個每筆刊登都帶著同樣可比較紀錄的市集。只解決一部分，市集好不好，取決於供給。 |
| 查驗 | 誰做的？從哪天開始？規格有沒有被改寫過？上個月是不是已經賣給別人了？每個答案都是一則私訊，加上一次信任的跳躍。 | 登記日、完整版本歷史、綁在角色上的作品，以及一個直接讀鏈的公開查驗。這是設計上攻得最用力的一桶。 |
| 標的定義 | 「這隻角色」對賣家是一回事、對買家是另一回事；爭執在付款之後才浮現，為的是一個沒人列出來的檔案。 | 一份帶指紋的規格，加一份檔案清單。契約指向一個編號，不用散文描述一個物件。 |
| 議價 | 不是一位沒報價的律師，就是一份雙方都沒真正讀過的複製貼上範本。 | 兩份參數化範本、可開關的選用條款，以及一份標出每處偏離標準的紅線。雙方對同一枚指紋簽名。 |
| 付款 | 先動的那一方承擔風險。 | 目前沒有。平台不代管資金，託管是這張清單上剩下最大的成本。 |
| 交付 | 一個 zip，加一句「沒留副本」的承諾，兩者都無從查證。 | 角色包是搬走不是複製，賣家的存取結束，檔案清單與簽署時的快照逐一比對。 |
| 執行 | 出事時，雙方先從「當初約定了什麼、交付了什麼」吵起。 | 一段附機器比對的驗收期、一個能凍結交易的爭議程序、一個可行且有紀錄的復原，以及一份任一方都能交給律師的交易紀錄。 |
為什麼查驗最值得攻
因為它是最不會隨規模攤平的一桶。議價成本只降一次，在好範本出現的時候；交付成本也只降一次，在有人把程序寫出來的時候。查驗成本卻在每一筆交易、對每一位買家都重複發生，而且付的是買家最討厭的貨幣：不確定自己會不會就是那個發現角色被賣了兩次的人。
台灣把這一點講得很直接。著作權登記制度 1998 年廢止之後，就沒有誰持有什麼的官方紀錄：買之前查不到，買之後也更新不了。美國有登記簿，但登記的是作品、要花幾個月，而角色不是作品。兩套制度在同一個位置留下同一個洞，那是一個私人登錄簿可以填、又不必假裝自己是公家登記的洞。
法律上有趣的部分
美國法院只在角色本身夠鮮明時才保護角色本身；判準之一問的是這個角色有沒有「被描繪到不管出現在哪裡，都認得出是同一個」。用創作者的眼睛讀這句話，它根本不是法律抽象概念，它就是讓一隻角色跨平台保持一致的日常工作，也正是這個登錄簿的佐證檢查在量的東西。
AI 這一面指向同一個方向。純機器生成的素材不受保護時，撐得起主張的是人的貢獻：那些決定、那些版本、那些被否決的替代方案。從第一天就留的來歷紀錄幾乎不花成本，而且是這種證據唯一可能的形態，因為它事後補不回來。
不是法律意見，而且刻意講得籠統：法理各國不同，事實也因角色而異。
還是很貴的部分
- 同時履行。沒有託管，總得有人先動。回報付款和確認收款縮小了窗口、留下了紀錄，但沒有消除風險。
- 作者身分。登錄簿記錄的是賣家聲明的內容。它沒辦法告訴你那是他畫的，任何宣稱做得到的登錄簿，賣的是安慰。
- 跨境執行。選一個準據法很便宜；為一筆四位數的買賣跨國提告不便宜。多數保護得來自程序，不是管轄條款。
- 冷啟動。搜尋成本只有在有東西可搜時才會降。那是供給問題，再多設計都解決不了。
誠實的總結
登錄簿不會讓一隻角色更值錢，也不會把證據變成權利。它做的是拿掉一筆交易在開始之前就散掉的理由：無法查證的買家、無法舉證的賣家、沒人想付錢請律師寫的契約、雙方都無法確認的交付。這四個成本，正是橫在「躺在資料夾裡的角色」和「換了手的角色」之間的東西。