luxray

I’ve now met with @TuomHolmberg and gone through the errors in my “Collector Crypt Has a Margin Problem” article (teapawt.net/FourPillarsFP/status/2…), as promised. This is the full correction of what was wrong, what the accurate figures are, and what stands. Error 1. The redemption count I claimed 75 user wallets redeemed cards out of ~6,000 June depositors, with the top 4 redeemers accounting for 47.1% of user burns. All of it is retracted, including the “1.22% redemption participation rate” built on it. I counted user-signed Metaplex Core burn instructions on Dune and treated them as physical redemptions. That proxy is structurally blind on CC’s architecture, and I have now verified how, using the recent 2,165 shipment transactions CC provided after our call. ~78% of shipment burns run through CC’s original pNFT collection, which my query never looked at, since it filtered only the newer Metaplex Core collection. Shipments also batch multiple cards per transaction, and CC pays gas on roughly a quarter of them, since a Privy wallet holding no $SOL can’t sign its own burn. Three separate blind spots, and the collection filter alone guaranteed the count would be wrong. The actual figures come from CC’s fulfillment data. 587 unique wallets withdrew 4,660 items worth $2.41M in insured value from May 31 to June 30, roughly 8x the wallets I reported. Their latest 30-day data, shared on our call, shows ~5,400 assets shipped to 634 unique users at $3.29M insured value. CC has also provided the full set of burn transaction signatures behind those shipments so the on-chain flow can be mapped correctly in the follow-up. Error 2. “Physical redemptions consumed 41.6% of May pre-burn net revenue” One error here plus a framing failure. The error is in the composition. A quarter of that number was never redemptions. The $3.76M May delta splits into $2.84M of redeemed inventory at cost and $0.92M of CC’s marketplace repurchase program, which is CC buying cards back into the vault, the opposite of inventory leaving. Redemption-only drag was 31.4% in May, falling to 18.0% in June. Relatedly, the article implied redemption pressure was climbing. True in dollars, but as a share of GMV redemptions have been stable at 2.5 to 3.0% since early 2025, with Q2 2026 the lowest in the series. The framing failure is that “consumed” presented redemptions as pure loss, and they are not. Each redemption earns CC more than a sellback. The full pack revenue stays, and what leaves is a card acquired below market. The CC team is right on the per-event economics. Higher redemption rates raise margins per event, not lower them. For scope, one thing this correction does not say. The article’s margin figures never assumed redeemed cards carry zero margin. The underlying data already nets redemptions at an assumed 85% acquisition cost, and my sensitivity model used the same base case. The failure was presentation. The numbers carried the redemption margin, the prose never showed it. The cost basis itself is a separate question the follow-up will address, and it likely cuts in CC’s favor. Error 3. Partner/API revenue I reported $1.83M all-time partner revenue, “mostly tied to Moonbirds,” with three near-zero quarters since. The attribution was wrong. Moonbirds was a one-day mint event, not ongoing API usage, and I’m assuming the Blockworks field I used does not capture partner machine revenue. Partner packs are recorded inside gacha revenue under partner memo tags. Per CC’s team, the Solflare integration alone generated ~$488K last month across machine profit and retained card value, and Magic Eden ran between $400K and $500K per month at its peak last fall. B2B revenue is materially larger and more recurring than I claimed. One assumption also refined, with CC’s input. My sensitivity model and the Blockworks dashboard both treat inventory acquisition cost as a flat 85% of value. CC describes a moving number instead. They buy cards at roughly 85% of market price at the time of purchase. Card prices have risen sharply since last year, so by the time a card is actually redeemed, its original cost sits closer to 70% of the market price on redemption day. And after heavy inventory accumulation at today’s prices in the past month, with the modern index flattening, they estimate current cost is back near 85%. For scale, their modern Pokemon index roughly doubled from its December low, their vintage index is up more than 130% since October, and vault inventory has grown from ~$3M at token launch to ~$23M today. For the period the article covered, this cuts in CC’s favor. Margins on redeemed cards ran better than a flat 85% assumption implies, and the follow-up will model the number instead of assuming it. CC also flagged the flip side themselves, that carrying heavy inventory works against them if card prices fall, offset by machine margins. What stands, and one further correction. The Blockworks-sourced volume and tier data stand. $728.9M all-time GMV through June 23, the migration into larger packs, Mythic capturing 36.7% of June GMV within 13 days of launch. But the article tied the margin decline to that migration, and re-deriving the series rules that out on timing. The decline was complete by December 2025, before the $1,000 and $2,500 machines existed, and margins did not compress further once those machines took over half the volume. The CC team was right to challenge that mechanism. The follow-up piece rebuilds CC’s margin picture from spin-level data in their public transaction API, with costs matched to the packs that produced them, and goes to their team for fact-checking before publication. The root cause was process. I mapped on-chain events to business processes without verifying the mapping with the operator. Tuomas said a 15-minute call would have prevented these errors. He was right. It took one hour, and they answered everything, with data. Going forward, @FourPillarsFP research on CC gets its data points and platform mechanics verified with their team before publishing. Conclusions stay ours, but facts get checked first. The original article now carries a permanent correction notice. The errors were mine alone. My thanks to Tuomas and the @Collector_Crypt team for engaging with data instead of anger. The walkthrough of their withdrawal architecture, pricing engine, and partner P&L is what every researcher should ask for before writing. I should have asked the first time.