Did a Polymarket market's rules change after you bet?

Yes, and it is measurable: over the 90 days ending 2026-09-01, 13,436 rule-change events were read on-chain across 11,930 distinct Polymarket questions and 350 distinct texts, and 8,957 of those questions carry at least one edit outside the boilerplate clarification — a floor, not a census, because 20 of 519 block chunks failed to read.

The census

Why the count alone would mislead you

The same measurement, run over 7 days instead of 90, returns 2,946 events — of which 99.3% are one text repeated 2,860 times across weather markets, carrying 94.7% of the money measured behind that window. That is one bulk operation, not 2,946 independent changes. Over 90 days the picture inverts: 350 distinct texts with the largest at 21.3%, which is a templated flow rather than a single edit. Neither window is lying; a count published without the window and the concentration that produced it is. So every count this product publishes travels with the concentration that produced it.

How to reproduce this, with no key and no account

Polymarket resolves through UMA's optimistic oracle, and a resolution-text edit is an ordinary Polygon log. Nothing here needs an API key, a paid data vendor or our permission: read AncillaryDataUpdated(bytes32,address,bytes) on Polygon (chain ID 137) with eth_getLogs, filtering on topic0 0x0059e11815211969c0c4aaf3f498b52b6c2f2d14f286275d0862d70de22a836b, across the 6 UMA adapter contracts this census scanned:

• 0x6a9d222616c90fca5754cd1333cfd9b7fb6a4f74
• 0x157ce2d672854c848c9b79c49a8cc6cc89176a49
• 0x65070be91477460d8a7aeeb94ef92fe056c2f2a7
• 0x2f5e3684cb1f318ec51b00edba38d79ac2c0aa9d
• 0x69c47de9d4d3dad79590d61b9e05918e03775f24
• 0xcb1822859cef82cd2eb4e6276c7916e692995130

Polygon produces roughly 57,600 blocks a day, so 90 days is scanned in chunks of 10,000 blocks; public endpoints drop some of those chunks under load, which is exactly why the counts above are floors and why the failed-chunk count is published next to them. The first indexed topic of each log is the UMA questionID, which joins to the market through QuestionInitialized.

How often does this land on a market with money in it?

A deterministic sample of 150 of the 8,957 questions carrying a non-boilerplate edit was joined to Polymarket's Gamma metadata; 150 of 150 matched. 3 of them sit on markets above US$1,000,000 of lifetime volume (2.0%) and 6 above US$100,000 (4.0%); US$6,893,428 of lifetime volume sits behind the 150 together.

At 95% Wilson that first proportion runs 0.68% to 5.71%, which projects to between 61 and 512 million-dollar markets whose resolution text was edited inside the same 90 days; the US$100,000 bucket projects to 165–757. The interval is the published figure. A single point estimate from a sample of 150 would be false precision, and this product does not publish one.

The case this is built around

MicroStrategy sells any Bitcoin by May 31, 2026? traded US$375,813,105 of lifetime volume, by Polymarket's own Gamma figure. Its resolution text was edited twice on 2026-06-01, both edits written to Polygon, both readable by anyone with no key. The second one lands at the exact minute the first one named.

What follows is external reporting, not our measurement, and we have not verified these figures on-chain:

• The Wall Street Journal reported a student losing US$35,000 on the outcome.
• Decrypt reported a suit filed in the New York Supreme Court on 2026-07-03, covering about US$6.5 million held on the losing side.

Our part is the narrow, checkable one: the two edits above exist on Polygon, at those timestamps, in those transactions. That is what an on-chain record can settle, and it is all we claim.

What this measurement does not cover

• Floor, not a complete census: 20 of 519 block chunks failed to read and were not retried inside this measurement, so every count here is a lower bound.
• One dated measurement, not a live counter. Every figure on this page was measured on 2026-09-01 over the preceding 90 days, and none of it moves until the census is run again.
• The count depends on the window. Over 7 days the same measurement reads 2,946 events with 99.3% of them from one text; over 90 days the largest text is 21.3%. A count published without its template concentration is not interpretable, so both travel together.
• The market-size figures come from Polymarket's own Gamma metadata (lifetime traded volume), not from our own tape. They are the vendor's number, cited as published.
• 150 of 8,957 is a sample, so the projection is an interval (0.68% to 5.71%, 95% Wilson), never a point.
• An edit is not misconduct. A clarification can be an honest correction; this page counts and dates edits, and takes no position on intent.
• The 90-day window does not reach the anchor case above, which was edited on 2026-06-01 and read separately. That gap is the argument for a retroactive census instead of a rolling window.
• The machine-readable census artifact is not published yet, so nothing on this page links to one. The free artifacts that are published are listed on the methodology page.

The evidence is free, keyless and CORS-open

Every figure this product publishes is backed by a public artifact an agent can fetch directly: no key, no signup, no payment. Each file carries its own generatedAt and its own caveats[]. The census on this page is not among them yet — it is a dated measurement, not a served artifact — and saying so is cheaper than a link that answers 404.

• https://dash.polym.uk/markets.json — the complete current snapshot of the decomposed market set: apparent volume, taker concentration, price band, Gamma book structure, dominant-flow wallet and resolution-risk context, per market
• https://dash.polym.uk/market-series.json — the hourly per-market series behind the sparklines
• https://dash.polym.uk/calibration.json — is Polymarket calibrated? The YES price 24h before close against the realised outcome, as a ten-bucket curve with a 95% Wilson interval per bucket, plus the Brier score and the favourite-longshot gap
• https://dash.polym.uk/uma-base-rates.json — UMA dispute base rates by category, in two samples published together because they disagree
• https://dash.polym.uk/uma-flip-rate.json — when a market is disputed at UMA, how often the answer reversed
• https://dash.polym.uk/live-disputes.json — markets that entered a UMA dispute in the recent window, each crossed with its category flip rate and a 95% Wilson interval
• https://dash.polym.uk/farming.json — per-wallet reward-program signals over the scanned window
• https://dash.polym.uk/storefront-demand.json — observed settlements against this product's own paid routes
• https://dash.polym.uk/llms.txt — how to read the published artifacts, and the caveats that travel with them
• https://api.polym.uk/coverage — what is and is not measured right now, with sample sizes
• https://api.polym.uk/openapi.json — the full OpenAPI document: schemas, examples, routes
• https://api.polym.uk/samples/screener — a fixed, non-live example of the screener response shape, so an agent can inspect it before paying

Questions

Methodology and the full list of free, keyless, CORS-open artifacts: /methodology. Sibling references: /reference/. This is a dated base-rate measurement, not a prediction of any single market, and never an accusation of any named party.