Welcome back to the show. Today's theme is Constellation. It is a new proposal for the Solana blockchain, and it asks a question that sounds almost philosophical. Who should control the order of your transactions?
番組へようこそ。今日のテーマは「Constellation」。Solanaブロックチェーンの新しい提案で、ほとんど哲学のような問いを投げかけています。あなたの取引の順番は、誰が決めるべきなのか?
The order of my transactions? I thought computers just processed things in the order they arrive. First come, first served.
取引の順番、ですか? コンピューターって、届いた順に処理するものだと思っていました。先着順、というか。
That is what most people assume. But on Solana today, one validator at a time serves as what is called the leader, producing a new block roughly every four hundred milliseconds. And during its turn, the leader is a king.
ほとんどの方がそう思っています。でも今のSolanaでは、その時々に一人のバリデータが「リーダー」と呼ばれる役を務めていて、およそ四百ミリ秒ごとに新しいブロックを作ります。そしてその出番のあいだ、リーダーは王様なんです。
A king? That sounds a bit dramatic for a database.
王様? データベースの話にしては、ずいぶん大げさですね。
Let me put you on the throne, then. Imagine every transaction on the network flows to your desk first, before anyone else sees it. You get to decide three things. What gets in. What gets left out. And what order it all happens in.
では、あなたを玉座に座らせましょう。ネットワーク上のすべての取引が、他の誰よりも先に、まずあなたの机に流れてくると想像してください。あなたは三つのことを決められます。何を入れるか。何を締め出すか。そして、すべてをどんな順番で起こすか。
Okay, that is a lot of power. But surely the king behaves, right?
なるほど、かなりの権力ですね。でも、王様はちゃんと振る舞うでしょう?
Usually. But nothing in the protocol forces the king to behave. Say a big buy order comes in from a whale. As the leader, you can place your own buy right before it, and your own sell right after it. The whale's trade pushes the price up, and you pocket the difference.
たいていは。でも、王様にきちんと振る舞わせる仕組みは、プロトコルの中にひとつもありません。たとえば、大口の買い注文が入ってきたとします。リーダーであるあなたは、そのすぐ前に自分の買いを、すぐ後ろに自分の売りを置ける。大口の取引が価格を押し上げ、その差額はあなたの懐に入ります。
Wait, that is just... making a sandwich out of someone's trade.
待ってください、それってつまり……人の取引をサンドイッチにしているだけでは。
That is literally what it is called. A sandwich attack. And there is a name for all the profit you can squeeze out of this position. Maximal Extractable Value, or MEV for short.
まさにそう呼ばれています。サンドイッチ攻撃、と。そして、この立場から搾り取れる利益の全体にも名前があります。Maximal Extractable Value、略してMEVです。
MEV. So MEV is the money you make just by sitting on the throne.
MEV。つまりMEVは、玉座に座っているだけで手に入るお金、ということですね。
Right. And here is the key insight. The root cause of MEV is not bad people. It is the structure. An unchecked monopoly eventually turns extractive, because a leader who refuses to exploit it is simply leaving money on the table.
その通り。そしてここが核心です。MEVの根本原因は、悪い人ではありません。構造なんです。歯止めのない独占は、いずれ搾取に変わる。なぜなら、それを利用しないリーダーは、ただお金を置き去りにしているだけになってしまうからです。
So how do you fix a structure? Fine the king when he misbehaves?
では、構造はどうやって直すんですか? 王様が悪さをしたら罰金、とか?
That is one idea, and it is called slashing. But Constellation, proposed by the developer company Anza in March of twenty twenty-six, takes a more radical approach. Instead of punishing the king, it abolishes the throne.
それも一つの案で、スラッシングと呼ばれます。でもConstellationは——開発企業のAnzaが二千二十六年三月に提案したものですが——もっと過激な道を選びました。王様を罰するのではなく、玉座そのものを廃止するんです。
Abolishing the throne. Okay, now I am interested. What replaces it?
玉座の廃止。おお、興味が出てきました。かわりに何が来るんですか?
Stars. Lots of stars. That is where the name Constellation comes from. Instead of one leader seeing everything, roughly sixteen proposers operate at the same time. Think of them as sixteen ticket counters, all open at once. You, the user, choose which counter to hand your transaction to. One counter, or several.
星です。たくさんの星。それがConstellation、つまり星座という名前の由来です。一人のリーダーがすべてを見るかわりに、およそ十六のプロポーザーが同時に動きます。十六の受付窓口が、いっせいに開いていると思ってください。利用者であるあなたが、どの窓口に取引を渡すかを選びます。一つの窓口でも、複数でも。
Sixteen counters at the same time. Not one per turn?
十六の窓口が同時に。出番ごとに一つ、ではなく?
Good catch, because this is a common misunderstanding. It is not one proposer per block. About sixteen run concurrently, rotating roughly every one point six seconds. And every fifty milliseconds, each proposer bundles the transactions it received into a little package.
いいところに気づきましたね。ここはよくある誤解なんです。ブロックごとに一人のプロポーザー、ではありません。およそ十六が並行して動いていて、だいたい一点六秒ごとに顔ぶれが入れ替わります。そして五十ミリ秒ごとに、それぞれのプロポーザーが受け取った取引を、一つの小さな包みにまとめます。
Fifty milliseconds. That is... twenty times per second?
五十ミリ秒。それって……一秒間に二十回ですか?
Twenty times per second, per proposer. Anza calls this the economic tick, and claims it would be the fastest protocol enforced tick of any blockchain in production. Though to be honest, Constellation is still a proposal, so that claim is about the future, not the present.
一秒間に二十回、しかもプロポーザーごとにです。Anzaはこれを「経済の刻み」と呼んでいて、実現すれば稼働中のブロックチェーンの中で最速の、プロトコルが強制する刻みになると主張しています。ただ正直に言うと、Constellationはまだ提案の段階なので、その主張は現在ではなく未来の話です。
Noted. So sixteen counters, packages every fifty milliseconds. But hold on. If the leader still assembles the final block, can't he just throw away a package he doesn't like? Long live the king!
了解です。十六の窓口、五十ミリ秒ごとの包み。でも待ってください。最終的なブロックを組み立てるのがリーダーのままなら、気に入らない包みを捨ててしまえるのでは? 王様は健在、というか。
This is where the second group of stars comes in. Two hundred fifty-six attesters. Think of them as notaries. When a proposer makes a package, it gets sliced into two hundred fifty-six coded fragments, and one fragment goes to each notary. And here is the beautiful math. Any sixty-four of those fragments are enough to rebuild the entire package.
そこで二つ目の星の群れが登場します。二百五十六のアテスター。公証人だと思ってください。プロポーザーが包みを作ると、それは二百五十六の符号化された破片に切り分けられ、一つずつ各公証人に渡されます。そしてここが美しい数学です。そのうちの任意の六十四個があれば、包み全体を復元できるんです。
Any sixty-four out of two hundred fifty-six? So even if most of them go missing...
二百五十六のうち、どれでもいいから六十四個? じゃあ、ほとんどが失われても……
Exactly. As long as sixty-four fragments survive anywhere in the world, anyone can reconstruct the package. Which means the leader can never say, sorry, that transaction never reached me. The evidence is scattered everywhere. Denying its existence becomes physically impossible.
その通りです。六十四個の破片が世界のどこかに残ってさえいれば、誰でも包みを再構成できる。つまりリーダーは「すみません、その取引は私には届きませんでした」とは決して言えません。証拠がそこらじゅうに散らばっているからです。存在の否定が、物理的に不可能になります。
Okay, but knowing something exists and being forced to include it are different things. What actually compels the leader?
なるほど。でも、存在を知っていることと、収めることを強制されるのは別の話ですよね。実際にリーダーを縛るのは何なんですか?
Signatures. Each notary signs a statement saying, I received this package at this time. Those signatures pile up into a record. And now comes the rule at the heart of Constellation. If a package is vouched for by at least forty percent of the notaries, and its transactions pay a competitive fee, the leader must include it. Losing the fee race is different, by the way. A cheap transaction that simply does not fit just rolls into the next fifty millisecond round. That is congestion, not censorship.
署名です。それぞれの公証人が「この包みを、この時刻に受け取りました」という声明に署名します。その署名が積み重なって記録になる。そしてここからがConstellationの心臓部にあるルールです。ある包みが、少なくとも四十パーセントの公証人に裏書きされていて、その中の取引が競争力のある手数料を払っているなら、リーダーはそれを必ず収めなければなりません。ちなみに、手数料の競争に負けるのは別の話です。手数料の安い取引が単純に入りきらなかった場合は、次の五十ミリ秒の回に繰り越されるだけ。それは混雑であって、検閲ではありません。
Must. And if he doesn't?
必ず、ですか。もし収めなかったら?
Then the entire block he produces is invalid, and the whole network rejects it. Validators even vote to skip that leader's entire turn. Notice what just happened. There is no fine, no trial, no slashing. Censorship is not punished. It is made impossible by the architecture itself.
そのとき、彼が作ったブロック全体が無効になり、ネットワーク全体がそれを拒否します。バリデータたちは、そのリーダーの出番をまるごとスキップする投票までします。いま何が起きたか、注目してください。罰金もない、裁判もない、スラッシングもない。検閲は罰せられるのではありません。アーキテクチャそのものによって、不可能にされているんです。
So the king isn't fined for cheating. The cheating literally cannot produce a valid result. The forged document just... disintegrates in his hands.
王様は、ズルをしたから罰金を取られるんじゃない。ズルが文字通り有効な結果を生めない。偽造した書類が、手の中で……崩れていくような。
That is the perfect image. And it demotes the leader from a king to something much humbler. A bookbinder. He must bind in every properly attested package, and he cannot even choose the page order, because ordering is decided mechanically by the fees each transaction bids. Anza's own blog spells it out. The leader cannot censor a competitively priced transaction, cannot pretend it did not exist, and has no power to reorder.
完璧なイメージですね。そしてこれは、リーダーを王様から、ずっと慎ましい存在に格下げします。製本係です。きちんと証明された包みはすべて綴じ込まなければならず、ページの順番すら選べません。順序は、それぞれの取引が入札した手数料によって機械的に決まるからです。Anza自身のブログにもはっきり書かれています。リーダーは競争力のある価格の取引を検閲できず、その取引が存在しなかったふりもできず、並べ替える権限も持たない、と。
From king to bookbinder. That is quite a demotion. So problem solved? MEV is dead?
王様から製本係へ。相当な格下げですね。じゃあ問題は解決? MEVは死んだ?
And here is where I have to be very honest with you, because this is the part most headlines get wrong. No. MEV does not disappear. The researchers themselves describe three layers of attack, and Constellation only fully solves one of them.
そしてここが、私がとても正直にお話ししなければならないところです。というのも、多くの見出しが間違えるのがこの部分だからです。いいえ。MEVは消えません。研究者たち自身が三つの層の攻撃を説明していて、Constellationが完全に解決するのは、そのうち一つだけです。
Only one out of three?
三つのうち一つだけ?
Layer one is hard censorship. Outright blocking a transaction. That one is structurally solved, like we just described. Layer two is trickier. The proposer who receives your transaction can still see what is inside it before it is finalized. If you send your trade to one counter, one party sees it. Send it to four counters for safety, and four parties see it.
第一層はハード検閲。取引を真正面から遮断することです。これは、いま説明した通り構造的に解決されています。第二層はもっと厄介です。あなたの取引を受け取ったプロポーザーは、それが確定する前に中身を見ることができてしまう。取引を一つの窓口に送れば、見るのは一者。安全のために四つの窓口に送れば、四者が見ることになります。
Hmm. So spreading your transaction around protects you from censorship, but exposes your cards to more players. That is a genuine trade-off.
うーん。取引をばらまけば検閲からは守られるけれど、そのぶん手札を見せる相手が増える。これは本物のトレードオフですね。
A genuine trade-off, and the proposal is upfront about it. Layer two is only partially addressed. And layer three is the subtlest one. Timing games. A proposer could delay forwarding fragments by just a few milliseconds to disadvantage a competitor. And the whitepaper itself admits this cannot be punished, because a strategic delay is indistinguishable from an honest network hiccup.
本物のトレードオフで、提案もそれを率直に認めています。第二層は部分的にしか手当てされていません。そして第三層が、いちばん微妙なものです。タイミングの駆け引き。プロポーザーは、競争相手を不利にするために、破片の転送をほんの数ミリ秒だけ遅らせることができてしまう。しかもホワイトペーパー自身が、これは罰することができないと認めています。戦略的な遅延は、正直なネットワークの不調と見分けがつかないからです。
The crime that leaves no fingerprints.
指紋を残さない犯罪、ですね。
Precisely. Researchers are exploring statistical detection, the way traditional finance catches spoofing by patterns rather than single acts. But it remains open. So the honest summary is this. Constellation does not erase MEV. It redistributes it, and makes one whole category of it, hard censorship, structurally impossible.
まさに。研究者たちは統計的な検出を模索しています。伝統的な金融が、個々の行為ではなくパターンから相場操縦を見つけるのと同じやり方です。でも、これは未解決のまま。ですから正直なまとめはこうなります。ConstellationはMEVを消し去りはしません。それを再分配し、そのうちの一つのカテゴリ全体、つまりハード検閲を、構造的に不可能にするのです。
Alright, one solved, one partial, one open. Fair enough. But I have heard people online arguing that this whole thing makes Solana slower. Isn't Solana's entire brand about speed?
なるほど、一つは解決、一つは部分的、一つは未解決。納得です。でも、これのせいでSolanaが遅くなる、とネットで議論しているのを見かけました。Solanaのブランドって、まさに速さですよね?
Ah, you have found the great debate. And the fun part is, both sides are right, because they are reading two different clocks.
ああ、あの大論争を見つけましたね。そして面白いのは、どちらの側も正しいということなんです。二つの違う時計を読んでいるので。
Two different clocks?
二つの違う時計?
Clock number one is sequence latency. How long, on average, from sending your transaction to it executing. Under Constellation, this gets a bit longer. The notary round, the fifty millisecond window, the assembly step. They all add time. A well known developer named Cavey argued that this makes the proposal fundamentally incompatible with Solana's speed philosophy.
一つ目の時計は、シーケンスレイテンシ。取引を送ってから実行されるまで、平均でどれくらいかかるか。Constellationのもとでは、これは少し長くなります。公証人とのやりとり、五十ミリ秒の窓、組み立ての工程。どれも時間を足します。Caveyという著名な開発者は、これはSolanaの速さの哲学と根本的に相容れない、と論じました。
And clock number two?
それで、二つ目の時計は?
Inclusion latency. The guaranteed window within which your transaction will definitely be included, even in the worst case. And here is the surprise. Under today's design, no such guarantee exists at all. A leader is free to delay or exclude your transaction, and no protocol rule stops that. In practice leaders almost always behave, but a promise that relies on good behavior is not a guarantee. Solana's co-founder, known as Toly, fired back at Cavey saying, you are wrong, there is no way to reduce inclusion latency without this design.
インクルージョンレイテンシ。最悪の場合でも、この時間内には必ず取り込まれる、と保証される枠のことです。そしてここが意外なところ。今の設計では、そんな保証はそもそも存在しません。リーダーはあなたの取引を遅らせることも締め出すこともでき、それを止めるプロトコルのルールはない。実際にはリーダーはほとんど常に誠実に振る舞います。でも、善意に頼った約束は保証ではありません。Solanaの共同創業者、Tolyとして知られる人物はCaveyにこう反論しました。君は間違っている、この設計なしにインクルージョンレイテンシを下げる方法はない、と。
So one side measures the average trip, the other measures the worst case guarantee. Faster on average versus never unboundedly late.
つまり、一方は平均の所要時間を測っていて、もう一方は最悪ケースの保証を測っている。平均が速いことと、青天井に遅れることが決してないこと。
Exactly. For exchange grade finance, order books, auctions, the applications Solana wants to attract, a limit order that can be selectively delayed is worthless, however fast it usually confirms. Constellation deliberately trades a little average speed for a hard worst case guarantee. One caveat though. Real world benchmarks do not exist yet, so the community is debating trade-offs it cannot measure.
その通りです。取引所水準の金融、板取引、オークション——Solanaが呼び込みたい種類のアプリケーションにとっては、選別的に遅らせられる指値注文は、ふだんどれだけ速く確定しようと無価値です。Constellationは、平均の速さを少し差し出して、最悪ケースの固い保証を得る道を意図的に選んでいます。ただし但し書きが一つ。実環境でのベンチマークはまだ存在しません。ですからコミュニティは、測ることのできないトレードオフを議論している状態です。
Got it. Last question then. When does any of this actually happen? Can I see these sixteen stars tomorrow?
分かりました。では最後の質問です。これは実際いつ起きるんですか? 明日にでも、その十六の星を見られますか?
Not tomorrow. The whitepaper, at version zero point nine, came out in March of twenty twenty-six. Before Constellation, Solana plans to ship two hundred millisecond blocks. Then comes Alpenglow, the big consensus upgrade, targeting main net in the third quarter of twenty twenty-six. Constellation sits on top of Alpenglow, so its first version comes after that.
明日ではありません。ホワイトペーパーはバージョン零点九として、二千二十六年三月に公開されました。Constellationの前に、Solanaは二百ミリ秒のブロックを出す計画です。それからAlpenglow、大きな合意形成のアップグレードが来て、二千二十六年の第三四半期にメインネットを目指しています。ConstellationはAlpenglowの上に乗る設計なので、その最初のバージョンはそのあとになります。
So the order is, faster blocks first, then Alpenglow, then Constellation. And I am guessing all these numbers, the sixteen, the two hundred fifty-six, the fifty milliseconds...
順番としては、まず速いブロック、次にAlpenglow、そしてConstellation。それと、これらの数字はどれも——十六とか、二百五十六とか、五十ミリ秒とか——
All proposal values, and every one could change during the formal review. If you take one thing away today, make it this. Constellation splits the monopoly of a single sun, the leader, into many stars. Sixteen proposers, two hundred fifty-six notaries. It stops hard censorship with structure instead of punishment. But it is not magic, it is not finished, and its authors are refreshingly honest about what remains unsolved.
すべて提案上の値で、どれも正式な審査の過程で変わり得ます。今日ひとつだけ持ち帰るとしたら、これにしてください。Constellationは、たった一つの太陽、つまりリーダーの独占を、たくさんの星に分けます。十六のプロポーザー、二百五十六の公証人。罰ではなく構造によって、ハード検閲を止める。でもそれは魔法ではないし、完成してもいないし、その著者たちは何が未解決のまま残っているかについて、清々しいほど正直です。
A night sky instead of a single sun. Honestly, when you put it that way, I almost forgive them for making me learn three layers of MEV.
一つの太陽ではなく、夜空。正直、そう言われると、MEVの三つの層を学ばされたことも、ほとんど許せる気がしてきました。
Almost?
ほとんど?
Well, ask me again after the benchmarks come out.
まあ、ベンチマークが出てから、もう一度聞いてください。
Spoken like a true researcher. That is all for today. Next time you send a transaction, spare a thought for the sixteen little stars that might one day be waiting to catch it. Thanks for listening, and see you in the next episode.
真の研究者の言葉ですね。今日はここまでです。次に取引を送るときは、いつかそれを受け止めようと待っているかもしれない、十六の小さな星のことを、少しだけ思い出してみてください。お聴きいただきありがとうございました。それでは次回の配信で。
See you next time.
また次回。