Splitrule

One venue, one screen: what it charges, what it changed, and what Splitrule has adjudicated on it.

VENUE

Polymarket US

QCX LLC, trading under the assumed name Polymarket US. Designated contract market, designated 07/09/2025. Access is a direct account with identity verification plus your own key, and the public catalogue answers without one. Splitrule supports it at Tier A, a full adapter.

Read 2026-08-04. The regulator's list records the assumed name against the QCX entity. Splitrule reads one venue in public, and this line is the whole of what it says about a venue's standing.

VENUE · WATCHES

What Splitrule checks on this venue every day

A watch reads one published page and compares it with the digest recorded here. It confirms that the page is unchanged, reports that it moved, or reports that it could not see. It keeps no copy of a page it does not retain, and it carries no URL onto this screen.

The registry holds 4 watches on this venue. The pin below covers 1 of them; the rest are read by the served lane alone and print no pinned row here.

VENUE · SCHEDULE

What does it cost to trade on Polymarket US?

These terms come from the venue's own published schedule, not from a summary of it. The fee is proportional to the price times one minus the price, so it peaks at 50¢ and falls toward the tails. Any flat per-contract quote is wrong on this venue by construction.

taker:  0.06 x C x P x (1-P)          1.50¢ per contract at 50¢
maker:  rebate of 0.0125 x C x P x (1-P)   0.31¢ per contract at 50¢

The zero-fee era on this venue ended 2026-07-01. The schedule names no settlement, membership or withdrawal fee. Splitrule reads that silence as unstated rather than as zero.

The forms, and where they were read

VENUE · FEE PER CONTRACT

How much edge does a trade on this venue need before it is worth doing?

An event contract pays a dollar or nothing, which makes the answer exact rather than approximate. A fee of one cent per contract is one percentage point of true probability you have to be right by, on top of the price. So the minimum edge a trade demands is not a matter of opinion or of market conditions. It is the fee schedule, read out loud.

Nothing below reads a market. The prices are a grid of whole cents chosen here, and each figure is a property of the published schedule this page names.

What does going in and coming out cost?

A taker who buys and later sells at the same price has paid the fee twice before being right about anything. That arithmetic decides whether a spread is an opportunity or a bill. The fee is proportional to the price times one minus the price, so it is largest at 50¢ and smallest at the edges.

PriceTaker, heldTaker round tripMaker rebate
0.28¢0.57¢0.06¢
10¢0.54¢1.08¢0.11¢
25¢1.12¢2.25¢0.23¢
50¢1.50¢3.00¢0.31¢
75¢1.12¢2.25¢0.23¢
90¢0.54¢1.08¢0.11¢
95¢0.28¢0.57¢0.06¢

The takeaway. A taker round trip at 50¢ costs 3.00 points of edge on this venue. The maker column is a rebate at every price, so the resting side is paid rather than charged. The two sides differ in structure, not by a few basis points.

Polymarket US taker: edge needed, hold at 50¢ 1.5000¢

inputs

  • P = $0.50
  • coefficient = 0.06
  • legs = 1 leg (hold to settlement)

formula

edge = coefficient x P x (1 - P) x legs x 100, in cents per contract

rounding

none. The frontier runs on the unrounded rate. Rounding is per fill, so a per-contract figure with a rounding step would describe one fill size, not a rate

provenance

  • docs.polymarket.us/fees, effective 12 AM ET, Wednesday July 1, 2026
  • schedule dated 2026-07-01; read 2026-08-04T20:54:37Z
  • checked daily; last run 2026-09-10, and the pmusFeeSchedule page still matched the digest we recorded, so this watch confirms it is unchanged and keeps no copy of it
  • capture pmus-fees-2026-08-04T20-54-37Z (machine-local), static capture of the published schedule; 99-row fee table and five worked examples in the bytes

bound

exact on the published terms. A $1 payout makes a cent of fee identical to a point of true-probability edge, so this is the edge the trade demands, not an approximation of it

§ permalink to this receipt
Polymarket US taker: edge needed, round trip at 50¢ 3.0000¢

inputs

  • P = $0.50
  • coefficient = 0.06
  • legs = 2 legs (in and out at the same price)

formula

edge = coefficient x P x (1 - P) x legs x 100, in cents per contract

rounding

none. The frontier runs on the unrounded rate. Rounding is per fill, so a per-contract figure with a rounding step would describe one fill size, not a rate

provenance

  • docs.polymarket.us/fees, effective 12 AM ET, Wednesday July 1, 2026
  • schedule dated 2026-07-01; read 2026-08-04T20:54:37Z
  • checked daily; last run 2026-09-10, and the pmusFeeSchedule page still matched the digest we recorded, so this watch confirms it is unchanged and keeps no copy of it
  • capture pmus-fees-2026-08-04T20-54-37Z (machine-local), static capture of the published schedule; 99-row fee table and five worked examples in the bytes

bound

exact on the published terms. A $1 payout makes a cent of fee identical to a point of true-probability edge, so this is the edge the trade demands, not an approximation of it

§ permalink to this receipt
Polymarket US maker: edge needed, hold at 50¢ -0.3125¢

inputs

  • P = $0.50
  • coefficient = -0.0125 (a rebate, signed)
  • legs = 1 leg (hold to settlement)

formula

edge = coefficient x P x (1 - P) x legs x 100, in cents per contract

rounding

none. The frontier runs on the unrounded rate. Rounding is per fill, so a per-contract figure with a rounding step would describe one fill size, not a rate

provenance

  • docs.polymarket.us/fees, effective 12 AM ET, Wednesday July 1, 2026
  • schedule dated 2026-07-01; read 2026-08-04T20:54:37Z
  • checked daily; last run 2026-09-10, and the pmusFeeSchedule page still matched the digest we recorded, so this watch confirms it is unchanged and keeps no copy of it
  • capture pmus-fees-2026-08-04T20-54-37Z (machine-local), static capture of the published schedule; 99-row fee table and five worked examples in the bytes

bound

exact on the published terms. A $1 payout makes a cent of fee identical to a point of true-probability edge, so this is the edge the trade demands, not an approximation of it

§ permalink to this receipt

The schedule panel above carries the terms these figures come from, and the rounding rule that decides who is right on a small fill. Run your own numbers on any market’s own page, under the filter.

RULES DRIFT · POLYMARKET US

What Polymarket US changed in its own rules text, over 40 intervals, 2026-07-31 to 2026-09-08

Venues publish the rules a market settles by, and they edit them. Nobody publishes the history. This is 40 intervals of that history for one venue, read field by field.

One venue. Every figure below is Polymarket US and says nothing about any other venue. Nothing on this screen is folded together with a reading from anywhere else.

The axis runs from 2026-07-31 to 2026-09-08, joined as 40 intervals from 41 usable captures. 5 captures are quarantined each as a failed walk (20260811 read 1 of a 10,807-market floor; 20260823 read 6,500 of a 10,807-market floor; 20260829 read 23,500 of a 10,807-market floor; 20260909 read 100,000 of a 10,807-market floor; 20260910 read 100,000 of a 10,807-market floor), so the axis steps over each and joins the captures either side into one wider interval; the widest interval on the axis is 48.01 hours.

The census is rebuilt daily and this page quotes ONE build of it, named in every receipt.

RULES DRIFT · WHAT MOVED

Deadlines moved. Resolution sources did not move.

Over 838,175 carried market-intervals, 550 carried a change of any kind. Deadlines moved, rules text changed length bracket and wording changed. Not one market changed the source its outcome is read from or changed its void language.

Deadlines moved, rules text changed length bracket and wording changed, and nothing else did Deadlines moved, rules text changed length bracket and wording changed, and nothing else did DEADLINE-SHIFT 321 LENGTH-CLASS-CHANGE 141 PROSE-EDIT-ONLY 88 321 DEADLINE-SHIFT 100.0% of the longest bar, 321 141 LENGTH-CLASS-CHANGE 43.9% of the longest bar, 321 88 PROSE-EDIT-ONLY 27.4% of the longest bar, 321
Every value in the chart above: Deadlines moved, rules text changed length bracket and wording changed, and nothing else did
ValueWhat it isShare
321DEADLINE-SHIFT100.0% of the longest bar, 321
141LENGTH-CLASS-CHANGE43.9% of the longest bar, 321
88PROSE-EDIT-ONLY27.4% of the longest bar, 321
Top class per changed market-interval, so the 3 bars are disjoint and sum to 550. The other 10 classes the extractor can fire are all zero here too.

The day-scale rate is 386 changes over 727,579 carried market-intervals, in the 32 intervals at least 20 hours wide and measured end to end by one instrument: 0.0531%. An all-interval rate would mix 0.01-hour and 48.01-hour intervals under one denominator.

One more interval sits outside that rate. The Splitrule capture instrument changed inside it, so its 144 changes over 15,693 carried market-intervals are counted apart from the rest. On that interval, a probe attributes them to the venue rather than to the instrument change.

Day-scale amendment rate 0.0531%

inputs

  • changed market-intervals = 386
  • carried market-intervals = 727,579
  • intervals at least 20h wide, single-era = 32 of 40
  • intervals at least 20h wide crossing an instrument boundary = 1

formula

386 / 727,579 = 0.0531%

rounding

four decimal places, display only; the counts are integers and are not rounded

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

a first reading over 40 intervals (2026-07-31 to 2026-09-08) on one venue, and the subset matters: the all-interval rate mixes 0.01-hour intervals with 48.01-hour ones under one denominator, so it is not quoted here. 1 day-scale interval carries 144 changes over 15,693 carried and is EXCLUDED from the numerator and the denominator above, because its two captures were taken by different instrument eras. Including it the rate is 0.0713%; the receipt beside this one says what it holds

§ permalink to this receipt
Changes across the instrument boundary 144 of 15,693

inputs

  • day-scale intervals crossing a boundary = 1
  • changed / carried across them = 144 / 15,693
  • instrument era 1 to 2 at 2026-08-12T23:11:16Z
  • probe verdict = VENUE-EDIT

formula

an interval is a boundary crossing when its two captures fall in different instrument eras; eras come from verification/rules-versions/instrument-boundaries.json, which records the commits that changed the walk or the axis (540d5d9, 6823258)

rounding

none — counts

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only
  • instrument era 2, 1 recorded for this venue
  • crossing interval(s): 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z

bound

the venue edited the rules text, so these are amendments and not an artifact of the instrument change. The rate above excludes this interval either way: a boundary crossing compares two captures taken by different instruments, and averaging it into a venue-behaviour rate answers neither question

§ permalink to this receipt
Carried market-intervals 838,175

inputs

  • intervals = 40
  • axis span = 2026-07-31 to 2026-09-08
  • widest / narrowest interval = 48.01h / 0.01h

formula

one carried market-interval is one market present in BOTH captures of one interval, summed over the 40 intervals; a market added or removed inside an interval cannot have changed across it

rounding

none — a count

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

markets recur across intervals by construction, so this is market-intervals and never markets: the same market carried through all 40 intervals contributes 40. Added 339,790 and removed 269,418 are counted separately and never netted

§ permalink to this receipt
Resolution sources changed 0 of 838,175

inputs

  • SOURCE-SWAP = 0
  • SOURCE-VANISHED = 0
  • SOURCE-APPEARED = 0
  • VOID-LANGUAGE-CHANGE = 0
  • carried market-intervals = 838,175

formula

SOURCE-SWAP fires when the set of source hostnames in the rules text moves with both sides non-empty; SOURCE-VANISHED and SOURCE-APPEARED read that same set emptying or filling from empty, each while the prose itself remains; VOID-LANGUAGE-CHANGE fires when the frozen void-marker set moves in either direction

rounding

none — counts

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

the extractor reads hostnames and frozen markers out of the prose, so a source that changed without its hostname changing would not fire. It is a floor on stability, not a proof of it

§ permalink to this receipt

Interval by interval, in capture order. 5 bars are wider than their neighbours: each is a joined interval stepping over one of the 5 quarantined walks.

2 of the 28 intervals carried no rules change at all 2 of the 28 intervals carried no rules change at all 08 Aug 14:30 · 24h 4 09 Aug 14:33 · 24.05h 4 10 Aug 14:30 · 23.95h 13 12 Aug 14:30 · 48h 50 13 Aug 14:30 · 24h 144 14 Aug 14:30 · 24h 1 15 Aug 14:32 · 24.04h 2 16 Aug 14:31 · 23.98h 9 17 Aug 14:30 · 23.99h 30 18 Aug 14:30 · 24h 18 19 Aug 14:30 · 24h 8 20 Aug 14:30 · 23.99h 5 21 Aug 14:30 · 24h 5 22 Aug 15:00 · 24.5h 2 24 Aug 14:30 · 47.5h 3 25 Aug 14:30 · 24h 61 26 Aug 14:30 · 24h 3 27 Aug 14:30 · 24.01h 1 28 Aug 15:01 · 24.5h 21 30 Aug 14:31 · 47.5h 0 31 Aug 15:47 · 25.28h 26 01 Sep 14:30 · 22.71h 22 02 Sep 14:31 · 24.02h 7 03 Sep 14:31 · 24h 5 04 Sep 14:31 · 23.99h 13 05 Sep 14:31 · 24h 8 07 Sep 14:32 · 48.01h 0 08 Sep 14:32 · 24.01h 12 144 13 Aug 14:30 · 24h 100.0% of the largest magnitude here, 144 61 25 Aug 14:30 · 24h 42.4% of the largest magnitude here, 144 50 12 Aug 14:30 · 48h 34.7% of the largest magnitude here, 144 30 17 Aug 14:30 · 23.99h 20.8% of the largest magnitude here, 144 26 31 Aug 15:47 · 25.28h 18.1% of the largest magnitude here, 144 22 01 Sep 14:30 · 22.71h 15.3% of the largest magnitude here, 144 21 28 Aug 15:01 · 24.5h 14.6% of the largest magnitude here, 144 18 18 Aug 14:30 · 24h 12.5% of the largest magnitude here, 144 13 10 Aug 14:30 · 23.95h 9.0% of the largest magnitude here, 144 13 04 Sep 14:31 · 23.99h 9.0% of the largest magnitude here, 144 12 08 Sep 14:32 · 24.01h 8.3% of the largest magnitude here, 144 9 16 Aug 14:31 · 23.98h 6.3% of the largest magnitude here, 144 8 19 Aug 14:30 · 24h 5.6% of the largest magnitude here, 144 8 05 Sep 14:31 · 24h 5.6% of the largest magnitude here, 144 7 02 Sep 14:31 · 24.02h 4.9% of the largest magnitude here, 144 5 20 Aug 14:30 · 23.99h 3.5% of the largest magnitude here, 144 5 21 Aug 14:30 · 24h 3.5% of the largest magnitude here, 144 5 03 Sep 14:31 · 24h 3.5% of the largest magnitude here, 144 4 08 Aug 14:30 · 24h 2.8% of the largest magnitude here, 144 4 09 Aug 14:33 · 24.05h 2.8% of the largest magnitude here, 144 3 24 Aug 14:30 · 47.5h 2.1% of the largest magnitude here, 144 3 26 Aug 14:30 · 24h 2.1% of the largest magnitude here, 144 2 15 Aug 14:32 · 24.04h 1.4% of the largest magnitude here, 144 2 22 Aug 15:00 · 24.5h 1.4% of the largest magnitude here, 144 1 14 Aug 14:30 · 24h 0.7% of the largest magnitude here, 144 1 27 Aug 14:30 · 24.01h 0.7% of the largest magnitude here, 144 0 30 Aug 14:31 · 47.5h 0.0% of the largest magnitude here, 144 0 07 Sep 14:32 · 48.01h 0.0% of the largest magnitude here, 144
Every value in the chart above: 2 of the 28 intervals carried no rules change at all
ValueWhat it isShare
408 Aug 14:30 · 24h2.8% of the largest magnitude here, 144
409 Aug 14:33 · 24.05h2.8% of the largest magnitude here, 144
1310 Aug 14:30 · 23.95h9.0% of the largest magnitude here, 144
5012 Aug 14:30 · 48h34.7% of the largest magnitude here, 144
14413 Aug 14:30 · 24h100.0% of the largest magnitude here, 144
114 Aug 14:30 · 24h0.7% of the largest magnitude here, 144
215 Aug 14:32 · 24.04h1.4% of the largest magnitude here, 144
916 Aug 14:31 · 23.98h6.3% of the largest magnitude here, 144
3017 Aug 14:30 · 23.99h20.8% of the largest magnitude here, 144
1818 Aug 14:30 · 24h12.5% of the largest magnitude here, 144
819 Aug 14:30 · 24h5.6% of the largest magnitude here, 144
520 Aug 14:30 · 23.99h3.5% of the largest magnitude here, 144
521 Aug 14:30 · 24h3.5% of the largest magnitude here, 144
222 Aug 15:00 · 24.5h1.4% of the largest magnitude here, 144
324 Aug 14:30 · 47.5h2.1% of the largest magnitude here, 144
6125 Aug 14:30 · 24h42.4% of the largest magnitude here, 144
326 Aug 14:30 · 24h2.1% of the largest magnitude here, 144
127 Aug 14:30 · 24.01h0.7% of the largest magnitude here, 144
2128 Aug 15:01 · 24.5h14.6% of the largest magnitude here, 144
030 Aug 14:31 · 47.5h0.0% of the largest magnitude here, 144
2631 Aug 15:47 · 25.28h18.1% of the largest magnitude here, 144
2201 Sep 14:30 · 22.71h15.3% of the largest magnitude here, 144
702 Sep 14:31 · 24.02h4.9% of the largest magnitude here, 144
503 Sep 14:31 · 24h3.5% of the largest magnitude here, 144
1304 Sep 14:31 · 23.99h9.0% of the largest magnitude here, 144
805 Sep 14:31 · 24h5.6% of the largest magnitude here, 144
007 Sep 14:32 · 48.01h0.0% of the largest magnitude here, 144
1208 Sep 14:32 · 24.01h8.3% of the largest magnitude here, 144
Changed market-intervals per interval, labelled with the interval's own width because a count over 0.01h and a count over 48.01h are not comparable. Each row is stamped with the moment its interval ended, never the moment it began: the last row, 08 Sep 14:32 · 24.01h, covers the 24.01 hours before that stamp. This chart draws the newest 28 of 40 intervals. The 12 older intervals before them, 2026-07-31 to 2026-08-07, carried 73 changed market-intervals.
Axis depth 40 intervals

inputs

  • captures on the machine = 46
  • viable / quarantined = 41 / 5
  • first capture 2026-07-31, last capture 2026-09-08

formula

2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, joined as 40 consecutive intervals

rounding

the interval count is an integer and the dates are the capture stamps, to the day; interval hours below carry two decimal places, display only

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

5 capture(s) are QUARANTINED, each a failed walk rather than a catalogue that emptied: 20260811T143013 read 1 distinct market(s) against a floor of 10,807; 20260823T145057 read 6,500 distinct market(s) against a floor of 10,807; 20260829T143216 read 23,500 distinct market(s) against a floor of 10,807; 20260909T143254 read 100,000 distinct market(s) against a floor of 10,807; 20260910T143316 read 100,000 distinct market(s) against a floor of 10,807. The axis steps over each and joins the captures either side of it as one wider interval; the widest interval on the axis is 48.01 hours

§ permalink to this receipt
Changes in sports 408 of 550

inputs

  • sports carried = 720,701
  • sports changed = 408
  • culture carried = 23,244
  • culture changed = 0

formula

sports rate 0.0566% over its own carried population, not over the venue's

rounding

four decimal places, display only

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

changes landed in 2 of the 10 categories (sports 408, politics 142), and the sports share is nearly uninformative on its own because sports is 85.9845% of the carried population. What survives is narrower: culture carried 23,244 market-intervals and changed none

§ permalink to this receipt

RULES DRIFT · WHAT THIS AXIS CANNOT SEE

42.3196% of the markets this store has ever seen were never carried

A change is only visible on a market present in BOTH captures of an interval. Of the 322,588 distinct markets across 41 captures, 136,518 appear in exactly one, and that one is neither the first capture nor the last. A market seen once is never carried, so no edit to its rules text can fire here. A further 18,022 markets are seen once at one end of the window, where a single appearance says more about the window than about the market, and they are excluded from that count.

Per interval the same bound reads differently, over a different population: 339,790 of 1,177,965 rows in a later capture had no earlier row to compare against, 28.8455%. The first share counts distinct markets across the whole axis. The second counts rows inside 40 intervals. They are two measurements and they do not combine.

Where this venue edits. Every one of the 550 changed market-intervals sits in 2 of the 10 categories the venue labels (sports 408, politics 142). The other 8 carried 35,892 market-intervals between them across the same 40 intervals and changed none. The category label is the venue's own. Splitrule derives no taxonomy of its own from the text.

A changed count does not say what changed. The two movers moved different things: sports — 321 where deadlines moved, 87 where wording changed; politics — 141 where rules text changed length bracket, 1 where wording changed. A deadline shift replaces one date or clock token in text that otherwise holds. A length-bracket change is prose that crossed a size boundary while every tracked field held. The count column reads both as one changed market-interval, and the class column is what separates them.

Category (venue label)Carried market-intervalsChangedTop class per change
sports720,701408DEADLINE-SHIFT 321, PROSE-EDIT-ONLY 87
politics81,582142LENGTH-CLASS-CHANGE 141, PROSE-EDIT-ONLY 1
culture23,2440
macro3,0160
finance2,9260
technology2,7050
crypto1,9750
climate1,1610
science4700
geopolitics3950

When the changes happened. The 550 changed market-intervals are not spread evenly across the 40 intervals: sports changed in 34 of them, politics in 1. All 142 politics changes fall in one interval, from 2026-08-12 to 2026-08-13. In the other 39 intervals politics changed none.

Which changes moved a term of the contract. A change counts here when it moved one of four tracked token sets. The four are a source host, a source reference, a deadline or clock token, and a void marker. Sports: 321 of 720,701 carried market-intervals. Politics: 0 of 81,582 carried market-intervals.

A zero there is not a finding that a category held still. Politics carries 142 of the 550 changes in this corpus, and none of them moved a term this instrument reads as a term. Of those, 141 crossed a text-size bracket while all four term sets stayed byte-identical. Text that names no date, host, URL or void marker is consistent with boilerplate. It is equally consistent with an operative sentence this extractor cannot read. The corpus cannot tell the two apart. This is a lower bound on what changed, never an estimate of it.

In 40 capture intervals of Polymarket US rules text — 838,175 carried market-intervals between 2026-07-31 and 2026-09-08 — Splitrule recorded zero resolution-source swaps, zero sources removed or added, and zero changes to void language. Every term-moving edit it did record was a deadline shift, and all 321 of them sat in sports markets.

Changes that moved a term 321 of 838,175

inputs

  • changed market-intervals that moved a tracked term = 321
  • changed market-intervals of any kind = 550
  • carried market-intervals = 838,175
  • by category = sports 321
  • changed and moved no tracked term = 229

formula

a change is TERM-MOVING when it moved one of four tracked token sets: a source host, a source reference, a deadline or clock token, or a void marker. Summed over the 40 intervals, 321 of the 550 changed market-intervals did, over 838,175 carried

rounding

none — counts

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

markets recur across intervals by construction, so this is market-intervals and never markets: a market that moved a deadline in two intervals is counted twice, and the distinct-market count is not this number and is not in the census. It is also a LOWER BOUND on materiality and never an estimate of it: a criterion rewritten in prose that names no new date, host, URL or void marker does not fire, and 141 of the 550 changes are exactly that shape: they crossed a text-size bracket with all four token sets byte-identical. That text is consistent with boilerplate and equally consistent with an operative sentence this extractor cannot read, and the corpus cannot tell the two apart

§ permalink to this receipt

Where the Splitrule instrument moved. The day-scale rate above counts 386 changes over 727,579 carried market-intervals, and every one of them is sports. It contains no politics change. The census counts an interval apart from that rate when its two captures were taken by different instrument eras. The interval holding those changes is the only such interval on this axis.

The Splitrule capture instrument changed inside the same interval that carried the largest cluster of changes in this corpus. On that interval, a probe attributes them to the venue rather than to the instrument change. Its changes are counted apart from the quoted rate either way.

Markets never carried 136,518 of 322,588

inputs

  • distinct markets over the whole store = 322,588
  • seen in exactly one capture = 154,540
  • ...of those, in neither the first capture nor the last = 136,518
  • captures joined = 41

formula

a market is CARRIED when it is present in both captures of an interval, and a drift class fires only on a carried market; a market that appears in exactly one capture is therefore never carried and no edit to its rules text can ever fire: 42.3196%

rounding

four decimal places on the share, display only; the counts are exact

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

a market seen once in the FIRST or the LAST capture is an edge effect of where the window stops rather than a market that lived and died between two walks, so the 18,022 of those are counted and excluded from the value. This bounds the instrument, not the venue: the walk reads every open market the catalogue returns, and the same store carries rotating categories whose markets ARE caught when the venue lists them across two walks

§ permalink to this receipt
Rows with nothing to compare against 339,790 of 1,177,965

inputs

  • later-capture rows = carried 838,175 + added 339,790 = 1,177,965
  • rows with no prior row = 339,790
  • intervals = 40

formula

summed over every interval: rows in the later capture of the interval whose market was absent from the earlier one, over all rows in the later capture: 28.8455%

rounding

four decimal places on the share, display only; the counts are exact

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json, extractor x2
  • built 2026-09-10T15:11:32.787Z by vps, sha256 b9380fef743e9e1c…
  • captures 2026-07-31T22:03:53.939Z to 2026-09-08T14:32:29.646Z, Polymarket US only

bound

this is a share of ROWS INSIDE INTERVALS and the never-carried share is a share of DISTINCT MARKETS over the whole axis. The two shares have different numerators, different denominators and different questions. Neither is a restatement of the other

§ permalink to this receipt

Both shares are Splitrule's measurement of its own reach, not a statement about the venue.

VENUE · ADJUDICATED EVENTS

Every event Splitrule adjudicated, newest first

An event is one thing that happened at one venue, over one named interval, with the cause adjudicated rather than asserted. Each one has its own page, its own permanent address, and a derivation under every number on it.

This feed is a record, not an alert. It says what Splitrule saw and adjudicated. It does not promise to catch the next one, and it monitors no market for any reader. Subscribe at feed.xml.

Events are rare by construction. Splitrule publishes one when a probe adjudicates a cause, and never to fill a page.

VENUE · SPREAD BY HOUR

How wide has the book been, hour by hour?

Splitrule keeps every scheduled walk of the open catalogue instead of throwing it away. The band below is the spread by hour of day, as quantiles, and every hour nobody measured is drawn as a gap rather than closed.

The history charts need JavaScript. The numbers behind them are served as plain JSON at /api/history.json, which needs none.

VENUE · CROSS-LISTED

How many questions on this venue Splitrule has confirmed elsewhere

A confirmed mark says the same question was found at another door. The count is what this panel prints, and the hover carries the denominator, the emitted row bound and the capture the count came from. No venue is named, no free text from the artifact is printed, and no price appears here.

Cross-listed not read [SERVED · DAILY]
source
D6 · served · Splitrule’s own verdicts, counts only.
cadence
Daily at 13:20Z · ceiling 72 h.
coverage
Counts only. A live read adds the denominator, the rows emitted and the capture id.

VENUE · DISPUTES

Resolution disputes on record

These incidents come from an oracle regime this build does not cover. The panel is collapsed because it is context rather than a reading about this venue, and every entry says which of the two it is.

Open the dispute record

The dispute record needs JavaScript. It is editorial content with public citations, shipped as a static file with this build.