Splitrule

One adjudicated event, with its interval, its denominator and its receipts.

ADJUDICATED EVENT · POLYMARKET US · 13 Aug 2026

A Venue Changed the Rules Text of 141 Politics Markets in One Day

Scope. One venue, one 24-hour interval, and the whole open catalogue that venue carried across it. A mechanical probe adjudicated the cause as a venue edit, and the tracked instrument registry carries the verdict.

Resolution text is the contract. It decides what a yes means, which source settles the market, and what happens in the case nobody priced. Splitrule keeps a versioned copy of the resolution rules text of every market. It takes a new copy once a day. It then compares each copy with the one before it, field by field.

On 13 Aug 2026 that comparison returned a number it had not returned before. The figures section below states how many markets changed and out of how many. The interval also crossed a change in the Splitrule capture instrument, so two opposite explanations were live at once: the venue edited its markets, or the instrument change moved what the extractor reads.

A probe separated the two mechanically. It ruled the interval a venue edit, and the verdict sits in the tracked registry that records instrument eras. The figures below are the figures of that interval. The per-market findings of the probe are not on this page.

Every figure on this page is scoped to the one interval named below. This page quotes no rate, no span figure and no current census reading. A figure from another frame does not belong beside these, and it is never differenced against them. The axis this interval sits in is the rules census on the venue screen.

ADJUDICATED EVENT · THE FIGURES

What the interval measured, with the derivation under each number

The interval ran 24 hours, from the capture at 2026-08-12T14:30:28.312Z to the capture at 2026-08-13T14:30:27.981Z. Of the 15,693 markets carried through both captures, 144 changed their rules text: 0.9176%. The probe publishes counts, digests and a verdict, and those are the figures here. Expand any figure for its inputs, its formula and its census.

Markets that changed rules text 144 of 15,693

inputs

  • changed market-intervals = 144
  • carried market-intervals = 15,693
  • interval width = 24 hours

formula

144 / 15,693 = 0.9176%

rounding

four decimal places on the percentage, display only. The counts are integers and are not rounded

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json built 2026-08-18T14:31:58.075Z
  • census digest 16a8add6035c883eac2932ac36abc2fd8c179c196e18690def13256d8d8f74bd
  • interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z, from the artifact's own intervals list
  • instrument eras from verification/rules-versions/instrument-boundaries.json

bound

inside ONE interval a market is carried at most once, so these carried market-intervals are markets and the count is 144 markets. The denominator is markets present in BOTH captures: 5,836 were added inside the interval and 5,893 were removed, and neither population can carry a change across it. This is a closed interval between two captures that already happened, so this row does not move: the same values were read from four census builds five days apart, and the page test re-derives them from the tracked artifact on every run

§ permalink to this receipt
Carried politics markets in this interval 1,047 of 15,693

inputs

  • politics carried = 1,047
  • carried market-intervals = 15,693
  • categories carried in this interval = 10
  • largest category = sports at 13,890

formula

1,047 / 15,693 = 6.6718%

rounding

four decimal places, display only

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json built 2026-08-18T14:31:58.075Z
  • census digest 16a8add6035c883eac2932ac36abc2fd8c179c196e18690def13256d8d8f74bd
  • interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z, from the artifact's own intervals list
  • instrument eras from verification/rules-versions/instrument-boundaries.json

bound

this is the SIZE of the politics population inside the interval, and it is not a count of what changed. The census records the category of every CARRIED market per interval, and the class of every CHANGED market only across the whole axis. The probe artifact splits the ADJUDICATED class by category, and that figure is below. No committed artifact splits ALL the changed rows of this interval by category. The label is the venue's own. a closed interval between two captures that already happened, so this row does not move: the same values were read from four census builds five days apart, and the page test re-derives them from the tracked artifact on every run

§ permalink to this receipt
The interval 24 hours

inputs

  • from capture 2026-08-12T14:30:28.312Z
  • to capture 2026-08-13T14:30:27.981Z
  • instrument era 1 to 2
  • crosses an instrument boundary = true

formula

one interval is the join of two consecutive captures of the same venue. An interval is a boundary crossing when its two captures fall in different instrument eras, and eras come from the tracked registry (cause commits 540d5d9, 6823258 at 2026-08-12T23:11:16Z)

rounding

none — the capture stamps are exact and the width is their difference

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json built 2026-08-18T14:31:58.075Z
  • census digest 16a8add6035c883eac2932ac36abc2fd8c179c196e18690def13256d8d8f74bd
  • interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z, from the artifact's own intervals list
  • instrument eras from verification/rules-versions/instrument-boundaries.json

bound

this interval CROSSES a change in the Splitrule instrument, so it was probed rather than published as a rate. The drift page excludes it from the day-scale rate for the same reason, in both directions: a boundary crossing compares two captures taken by different instruments

§ permalink to this receipt
Adjudicated cause VENUE-EDIT

inputs

  • probe verdict = VENUE-EDIT
  • boundary recorded at 2026-08-12T23:11:16Z
  • instrument version 1 to 2
  • cause commits = 540d5d9, 6823258

formula

tools/drift-boundary-probe.ts separates an extraction change from a venue edit by signature: a truncation that stopped leaves the old text as a prefix of the new one and clusters at a few exact lengths, while a template rollout does neither and concentrates in the category vocabulary of the venue itself. The rule: at least 20 firings or no ruling; truncation tell = top-5 from-side lengths cover >= 0.5 of firings AND every firing is containment; template tell = one span recurs (in 2+ firings) across >= 0.5 of firings OR >= 0.9 of firings in one venue category at >= 3x enrichment against its share of the carried population

rounding

none — a verdict, not a number

provenance

  • verification/rules-versions/instrument-boundaries.json, the tracked instrument-era registry
  • boundary row: polymarket-us 2026-08-12T23:11:16Z, era 1 to 2
  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

the venue edited the rules text, so these are amendments and not an artifact of the instrument change. UNPROBED and UNDECIDED are different states and neither licenses calling a change drift. The rule read: top-5 from-side lengths cover 76/141 = 53.9% (distinct from-side lengths 20); containment 2/141 (strict prefix 2, reverse 0); most-recurring single span 3/141 = 2.1%; top venue category 141/141 of firings = 100.0%, 1047/15693 of carried = 6.7%, enrichment 14.99x. Truncation tell false, template tell true. Each number in that line is a figure on this page. The probe reads a machine-local corpus and publishes the counts, so the verdict is citable here and its per-market working is not

§ permalink to this receipt
Structural classes that did not fire 10 classes at zero

inputs

  • classes at zero across the whole census read = 10
  • source and void classes among them = 5 (SOURCE-VANISHED, SOURCE-APPEARED, SOURCE-SWAP, VOID-LANGUAGE-CHANGE, SOURCE-LOCATION-MOVED)
  • census read = the build stamped 2026-08-18T14:31:58.075Z, identified by its digest rather than by its width

formula

a class with zero firings across the whole census read has zero firings inside any interval of it, so an axis-wide zero bounds this interval. The direction is one-way and only this way: an axis-wide count says nothing about which interval a NONZERO count came from

rounding

none — counts

provenance

  • census artifact verification/rules-versions/drift-census.polymarket-us.json built 2026-08-18T14:31:58.075Z
  • census digest 16a8add6035c883eac2932ac36abc2fd8c179c196e18690def13256d8d8f74bd
  • interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z, from the artifact's own intervals list
  • instrument eras from verification/rules-versions/instrument-boundaries.json

bound

the extractor reads source hostnames and frozen void markers out of the prose, so a source that changed without its hostname changing would not fire. This is a floor on stability rather than a proof of it. The zero is frozen as what THAT census build read, so it cannot go stale. A later census that fires one of these classes does not change this reading. It makes a new one

§ permalink to this receipt
Markets that fired LENGTH-CLASS-CHANGE 141 of 15,693

inputs

  • firings of LENGTH-CLASS-CHANGE = 141
  • firings whose text both captures hold = 141
  • markets carried through both captures = 15,693
  • markets that changed rules text in this interval = 144

formula

141 / 15,693 = 0.8985%

rounding

four decimal places on the percentage, display only. The counts are integers

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries
  • census artifact verification/rules-versions/drift-census.polymarket-us.json built 2026-08-18T14:31:58.075Z
  • census digest 16a8add6035c883eac2932ac36abc2fd8c179c196e18690def13256d8d8f74bd
  • interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z, from the artifact's own intervals list
  • instrument eras from verification/rules-versions/instrument-boundaries.json

bound

these 141 are a subset of the 144 markets that changed rules text here: a market fires a class only if it changed. Both counts come from the same two captures and the same axis build, so 3 of the changed markets changed some other way. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt
Firings in politics 141 of 1,047

inputs

  • politics firings = 141
  • politics markets carried through both captures = 1,047
  • all firings = 141
  • all markets carried = 15,693
  • venue categories that fired at all = 1

formula

141 / 1,047 = 13.4670%. Share of firings 141 / 141 = 100.0000%. Share of the carried population 1,047 / 15,693 = 6.6718%. Enrichment = 100.0000% / 6.6718% = 14.99 times

rounding

four decimal places on each percentage and two on the enrichment, display only

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

the denominator is the population this numerator can come from, and it is not the whole catalogue. Enrichment reads one when firings fall where the catalogue already is, so it is the figure that survives a category holding most of the markets. The label is the venue's own published one. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt
Direction and size of the change 141 grew, 0 shrank

inputs

  • grew = 141, shrank = 0, unchanged in length = 0
  • magnitude buckets: 100-999 characters: 141
  • length-class moves: lenXS to lenS: 141
  • measurable firings = 141

formula

for each firing, the character count of the new text minus the character count of the old one, after the same whitespace normalisation the extractor applies. The sign gives the direction and the absolute value gives the bucket. A length class is the Splitrule name for a band of character counts

rounding

none. Character counts and firing counts are both integers

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

the buckets partition the measurable firings exactly once each, so they sum to that population and not to anything wider. A length class is a Splitrule term, not the venue's. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt
New text that contains the old text 2 of 141

inputs

  • old text is a strict PREFIX of the new text = 2
  • old text is a strict SUBSTRING of the new text = 2
  • new text is a strict substring of the old text = 0
  • measurable firings = 141

formula

(2 + 0) / 141 = 1.4184%

rounding

four decimal places on the percentage, display only

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

containment is not evidence of a venue edit on its own, and it is not evidence against one. A venue that appends a paragraph produces containment too. It matters here because an extraction that stopped truncating produces containment on EVERY firing, so a low count rules that explanation out. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt
Clustering of the old text lengths 76 of 141

inputs

  • distinct old-text lengths across the firings = 20
  • the 5 most common lengths: 169 characters: 22; 165 characters: 18; 168 characters: 16; 171 characters: 10; 167 characters: 10
  • firings those 5 cover = 76
  • measurable firings = 141

formula

76 / 141 = 53.9007%

rounding

four decimal places on the percentage, display only

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

a hard character cap in the Splitrule extraction shows here as a spike, because independently written documents do not share exact character counts. Clustering alone does not carry a verdict: short formulaic text clusters for its own reasons. A length is the size of one document and it names no market. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt
The same markets one interval earlier 82 of 141 carried, all stable

inputs

  • firings this interval = 141
  • of those, carried through the earlier interval = 82
  • of those, no class fired there = 82
  • of those, already firing there = 0
  • not carried through the earlier interval = 59
  • the earlier interval = 20260810T143027 to 20260812T143028

formula

82 + 59 = 141, so the split covers every firing exactly once

rounding

none. Every value here is a count of markets

provenance

  • probe artifact verification/rules-versions/boundary-probes.json, written once per interval
  • probed at 2026-08-19T00:09:52.059Z by tools/drift-boundary-probe.ts, extractor x1
  • class LENGTH-CLASS-CHANGE, interval 2026-08-12T14:30:28.312Z to 2026-08-13T14:30:27.981Z
  • capture 20260812T143028: 15,112,543 bytes, sha256 d5da96e4c7edd07836b1709f45be085b47292fc44b97a2f8825d9bfa0c1520a0
  • capture 20260813T143027: 14,998,641 bytes, sha256 c753642fd5850324f16112643ded6ff46ae1f2da1528cabda97988cb16f9f0d5
  • capture 20260810T143027: 11,568,693 bytes, sha256 f178fdf047fd8eb2e6f43b388123873426b7be759def5e28924af2a312d52509
  • the captures are machine-local by ruling. The counts above are what the tracked artifact carries

bound

EVERY NUMBER HERE COUNTS THIS INTERVAL'S OWN FIRINGS. None of them is a count of what happened in the earlier interval, and nothing here is subtracted from anything on another page. The earlier interval is named as an identifier, the way a capture stamp is. A market not carried there was absent from one of its two captures, which is a fact about the catalogue. the probe reads each document once and emits the total. The artifact holds counts alone, so no figure here can be taken back to a single market. The count covers this interval and nothing wider

§ permalink to this receipt