Chung仲

Grading doesn't get enough credit as an invention. Before PSA, every card was a private argument about condition. After it, two strangers on opposite sides of the world could agree on what they were pricing without ever seeing the card. Sealed slab, cert number, public pop report. This market built itself a unit of account decades before anyone thought to put one on a chain.
It just never got around to building a settlement layer.
That gap usually gets told as a friction story. Shipping, insurance, customs, the two weeks between deciding to sell and actually having the money. All true, all solvable, and all the boring half.
Here's the part people skip. The grade is a claim about condition. It says nothing about supply, and nothing at all about whether a single human being wants the card. Pop reports are public and they only move one direction. Every submission window quietly fattens the denominator that every comp gets marked against.
For vintage Pokémon that denominator barely budges. The raw pool is finite and thoroughly picked over, so a PSA 10 population behaves like a stock. For modern print runs it's a flow, and a quick one. That, not age or hype, is why Pokémon at thirty and One Piece at four don't trade like the same asset.
But supply is the smaller half of what a card is worth.
The other half has no ticker. Why a Poncho-Wearing Pikachu still holds a bid nine years on while numerically scarcer promos from the same era just sit there. Why a Dodgers crossover Luffy, a card with nothing to do with how anyone actually plays or collects One Piece, became something people chase. Grade is verifiable, population is public, and neither one sees this coming. Taste does: the collective, slow-forming, weirdly stubborn judgment about which cards deserve to be the ones that matter.
That isn't a nice sentiment bolted onto the analysis. It's the analysis itself, because taste decides whether a card has a second buyer at all. A card nobody wants is illiquid at any population. A card everyone wants is liquid at any population. Liquidity is downstream of taste, every time. (@Plus_Ultra_715 says this a lot and he's not wrong!)
Which is why the scale matters. Plenty of fair market values here rest on a handful of sales a year, and marks that thin hold up beautifully until two holders exit in the same quarter. And that has never happened at any meaningful size.
Prices get scored continuously. Liquidity gets scored once, on the way out. Record sales, population milestones, all the first kind. The second is untested, because this market has only ever known money coming in.
Taste isn't an asset class. It's what decides whether one exists at all. Price is a demand problem: visible, tradeable, mean-reverting. Liquidity is a taste problem in structural clothing, and taste answers slowly. You can't iterate your way to it, and no amount of infrastructure will manufacture it for you.
oh yeah i collected three Riolus with consecutive PSA cert numbers on @renaissxyz how's that for taste lol

La Chine a la première interface cerveau-ordinateur implantable (bci) approuvée, les délais d’essai les plus rapides du secteur et davantage d’argent investi que partout ailleurs en dehors des États-Unis. Plus de 5 Md RMB au premier semestre 2026, plus que 2024 et 2025 réunis.
La plupart des gens y voient une histoire réglementaire. Un examen éthique plus léger, des hôpitaux qui disent oui, un État qui a décidé de passer à l’action. C’est une partie du tableau. Mais cela occulte la moitié la plus intéressante.
Regardez ce qui a réellement été construit.
La plupart des programmes chinois qui sont parvenus jusqu’aux patients se situent hors du cortex. Des électrodes au-dessus de la dure-mère ou reposant dessus, sans être enfoncées dans le tissu cérébral. Le dispositif qui a remporté la toute première approbation au monde ne touche jamais le cerveau : il fonctionne grâce à une alimentation sans fil à travers le crâne, plutôt que sur une pile. Il n’y a donc rien à réinsérer ni à remplacer.

Note de recherche Chung n°8 - l’avantage en robotique
Le fossé en robotique n’est pas le modèle. C’est la collecte des données.
Tout le monde surveille la course au matériel et la course aux benchmarks. Mais ce n’est peut-être pas là que se trouve la véritable histoire.
Réfléchissez à la façon dont GPT a été entraîné. Le texte existait déjà. 20 ans de l’internet, des trillions de tokens, là, avant même que quiconque se mette en tête de l’entraîner.
Les robots n’ont pas un tel avantage. Vous ne pouvez pas récupérer « comment remplir un verre d’eau sans le renverser » depuis une page web. Chaque mouvement qu’un robot apprend doit être produit dans le monde physique, une répétition à la fois.