Audit: pmxt/v1
What this page is for: the problems with this corpus and the evidence behind them. The files are on the download page, and what the columns mean is in the format notes.
What is wrong with this data
Treat v1 as a partial sample of its era: useful for presence and shape, never for completeness claims.
- About a third of traded markets are missing. Against on-chain fills, v1's coverage ratio is 0.6500 across its six pre-registered audit hours, ranging 0.3224 to 0.9994 hour by hour. Independently, on the hours v1 shares with v2, v1 misses 41.93% of the market universe of that moment while v2 misses 2.28% of v1's. Two methods with no shared input, the same answer. audit instrument against on-chain fills, and recon 58954ca6, 2026-08-31
- Roughly half of v1's fills are phantom, and that is the venue's behaviour rather than a fault in this archive. 46% to 48% in every measured v1 hour. Polymarket's relayer reported fills that the on-chain contracts then rejected; traders called these ghost fills, and Polymarket's own exchange upgrade of 2026-04-28, which brought new contracts, a rewritten order book, and fees charged at match time instead of order time, was aimed squarely at that class of failure. PMXT recorded what the venue reported, which is what an archive is supposed to do. The phenomenon ends at 2026-06-19: markets from that date onward carry none. And no market is entirely phantom, since every traded market has at least one real fill, so this affects fill COUNTS and never which markets appear. Polymarket's note on the upgrade. audit instrument for the percentages, Polymarket upgrade notes for the cause, 2026-08-31
- The real data ends earlier than the listing suggests. v1's last hour with data is 2026-04-15T08. The 21 hours listed after it, 2026-04-15T09 to 2026-04-16T05, are zero-row placeholder objects of 214 bytes each. They are listed because they exist, not because they hold anything. hour census against the published v1 SHA256SUMS.txt, with object sizes, 2026-08-31
- No transaction hashes, so row-level reconciliation is impossible. v1 records no trade events and carries no transaction_hash column, so nothing in this era can be reconciled against chain fills by any method. read from the files, 2026-08-28
Measured against the chain
Every figure here is measured against on-chain fills for the same hour, by an instrument that shares no input with this archive. Six hours were drawn, and which six was written down before anything was measured. Six hours is a floor, not a sample large enough to describe a corpus, so nothing here is a population rate.
| Hour | Traded markets | Ratio | Missing |
|---|---|---|---|
2026-02-23/23 |
3,478 | 0.9994 | 2 |
2026-03-12/17 |
5,215 | 0.7168 | 1,477 |
2026-03-12/18 |
4,768 | 0.6804 | 1,524 |
2026-03-20/00 |
4,159 | 0.3224 | 2,818 |
2026-03-21/09 |
4,334 | 0.4742 | 2,279 |
2026-04-15/01 |
3,826 | 0.7585 | 924 |
| All 6 | 25,780 | 0.6500 | 9,024 |
Against our own capture, which scored 1.0000 on the same instrument: this corpus is at 0.6500, so roughly 35 in every 100 markets that demonstrably traded have no book here at all. That is what a low coverage number means: not that the data is imprecise, but that the market is missing, and nothing downstream can recover a book that was never recorded. See all four side by side.
Pre-registration 5ae4e8765d601145, draw
02a94091dea9c11f, measured 2026-08-31.
Markets held, against markets traded
A corpus holds a book for many markets that nobody traded in a given hour. That is not waste: an untraded market with a book is breadth, and it is the difference between an archive you can ask questions of and one that only answers about the busy markets. This counts both sides for each drawn hour.
This corpus does not have one number. Across the drawn hours it runs from 3.1x to 8.2x, and in date order the direction is plain, so a median of 3.9x would describe a state it was never in for long. It is shown by hour, oldest first.
| Hour | Markets held | Markets traded | Held per traded |
|---|---|---|---|
2026-02-23/23 |
28,367 | 3,478 | 8.2x |
2026-03-12/17 |
19,244 | 5,215 | 3.7x |
2026-03-12/18 |
19,532 | 4,768 | 4.1x |
2026-03-20/00 |
12,762 | 4,159 | 3.1x |
2026-03-21/09 |
16,356 | 4,334 | 3.8x |
2026-04-15/01 |
30,097 | 3,826 | 7.9x |
Our own capture holds about 52 books for every market that traded; this one holds about 4, roughly 13.2 times fewer. What it holds can still be trusted: this is a statement about how much of the market it watched, not about whether what it recorded is right. See all four side by side.
Breadth is not completeness and a bigger number is not automatically better: this says nothing about whether the markets that did trade have books, which is what the coverage ratio above measures. The two share no arithmetic, and they rank the four corpora in the same order.