Fifty Years of Cards. Ten Days of Ledger.
Cemini23 ·
My father and I have collected baseball and sports cards our whole lives. Fifty years of them, now catalogued into 1,068 rows, and only 387 of those rows have a numeric comp. 614 carry the literal text NO_EVIDENCE in the price cell. The letter explains the two mechanisms that came out of the work: a sold comp is a claim about a different card that shares a name, and a grading submission is an option with a computable breakeven probability.
What does the pricing run check?
The 11 November 2026 issue Fifty Years of Cards. Ten Days of Ledger. is the canonical write-up. It runs over the operator's own card inventory and reports four things: how many cards have a real comp, what the comped cards net after fees, which cards are worth submitting to a grader, and which cards cannot be identified well enough to spend money on. Every figure is computed from the master CSV on the date shown. No estimate in the piece is hand typed.
How much of a collection actually has a price?
| Measure | Result, 11 November 2026 |
|---|---|
| Cards in the collection | 1,068 |
| Cards with a numeric comp | 387, or 36 percent |
| Cards with no price at all | 681, or 64 percent |
| Cards marked NO_EVIDENCE in the price cell | 614 |
| Comp total, gross | $9,947.13 |
| Comp total, after eBay fees | $8,474.34 |
| Reference floor from notes on the uncomped rows | $7,149.39 across 128 rows |
| Rows blocked by the spend gate | 226 |
Why is a comp not a price?
A sold-listing comp records one sale of one card. It does not record a rate for your card. The clearest case in the collection is a 2024 Bowman Draft Chrome Prospect Autograph of Josh Caron, Blue Refractor, serial 071/150. The base page gives a comp of $2.25, the median of five recent raw sales. The blue refractor page gives $19.99. Same card, same photograph, same collector, same day, 8.9 times apart. The base page was not lying. It was answering a different question.
The worst case is a Reggie Crawford Lumber Signatures bat relic auto whose serial is unreadable through glare. Four sales might apply: $17.99 and $0.50 on the base page nineteen days apart, $3.80 on Holo Silver, and $9.75 on Holo Gold. Fifty cents to seventeen ninety-nine for a card the operator cannot positively identify. The row stays NO_EVIDENCE until a straight-on photograph exists.
How does the grading model work?
A submission is an option. The fee is the premium, paid now. The grade is the payoff, received later, and the submitter does not control it. The model therefore returns the one number that decides the trade: the probability of a PSA 10 at which the submission breaks even.
- Breakeven probability. The tier cost plus shipping, minus the net gain at a PSA 9, divided by the gap between the net gain at a 10 and the net gain at a 9.
- Yes band. A breakeven at 45 percent or below is a submission, provided the expected gain clears $35.
- No band. A breakeven above 70 percent is a hard no.
- Middle band. Between 45 and 70 percent, a person decides. The tool does not.
What is a 10 worth?
For a raw 1955 Topps Jackie Robinson in the collection, the guide puts the raw card near $455, the PSA 8 near $10,424, and the PSA 10 near $278,160. The PSA 8 is 23 times the raw card. The PSA 10 is 611 times. Those are guide prices, not sales, and the tool labels them that way on the row. The shape is still the point. On vintage cards the grade is not a quality measurement. It is the asset.
Seven cards clear the bar, and all seven are vintage: Jackie Robinson, Yogi Berra, and five Mickey Mantles. Three are already in BVG slabs, so those are crossover decisions rather than fresh submissions. No modern base card makes the list.
Is a 10 always a 10?
No. The collection holds sixteen graded cards, and the labels tell three different stories. A GMA Gem MT 10 on a 2019 Zion Williamson comps at $1.25. A GMA 10 on a 2015 Aaron Judge comps at $7.61. A BCCG 10 on a 1992 Mariano Rivera comps at $52.25. The number on the label is identical. The market is not. The tool treats CSA, BCCG, GMA, BVG, and SGC as crossover candidates, not as graded value.
Why does the gate never say yes?
The spend gate answers one question: may this row go to a person for a decision? It never answers yes on its own. It holds no dollar threshold and no grade cutoff, so it cannot move either one. A row clears only when it has no audit error, no open identity question, and no unresolved photo citation. A clear row still needs a human to approve the spend, the slab route, or the consignment. The 226 blocked rows are not a valuation. They are a list of the things the operator does not know.
What does the tool refuse to do?
- It does not write a comp from a reference. A dollar figure in a note is a reference, not a sale, and the price cell stays empty.
- It does not match a serial card to a base page. A serial Bowman, Topps, Chrome, Prizm, Select, Donruss, or Gypsy Queen card no longer resolves against an unnumbered base slug.
- It does not trust a single sale. When two sales of a supposedly identical card differ by more than three times, the tool takes the median. A three-times spread is two different cards wearing one name.
- It does not clear an identity on a guess. A field loses its question mark only on a one-to-one match at 0.90 confidence or higher, and it never invents a year.
What happens after the collection sells?
The plan is to finish the ledger, sell with a full record of every card and every sale, and then publish the tools and the rules. That means the intake pipeline, the parallel-matching rule, the three-times spread rule, the grading probability model, the spend gate, the pull sheet that hides the internal ids, and the consignment manifest that photographs every serial before a box leaves the house. The work took ten days and a long list of mistakes, and each mistake is now a rule in the code.
What else has Outlier Weekly covered?
Atto is the August letter on genealogy paperwork. Historical Atlas is the October letter on a free place lookup. Swarmer is the October letter on drone-swarm revenue. Rearmament sets the war theme. Voyager, Tanbreez, and tungsten are the September and October names. SPCX / Anthropic and Anthropic IPO are the listing letters, and the parlay letter is the operator stack.
Subscribe at outlierweekly.substack.com/subscribe. The letter is free to read.