Source-linked AI summary
From Public Evidence to Contractual Outcome: First and Stable Decidability on Kalshi
Maksym Nechepurenko
TL;DR
Public evidence may settle a listed contract before venue determination, but identifying when requires the applicable rules, evidence objects, and unfinished conditions. This paper defines first and stable clocks and tests historical recoverability while building prospective capture infrastructure; the historical binding was not recoverable, and the shakedown established observation capability rather than clock results.
Problem
Venue timestamps combine source delay, contractual waiting, ambiguity, review, and unobserved procedure, leaving a need for a clock upstream of venue action.
Source: §1 · p. 2Method
Results
The historical test found the required rule and release objects unobservable in the available public record, while the prospective shakedown established observation capability but reported no clock values.
Source: §12 · p. 11Takeaways & Limitations
A clock may be economically well-defined yet retrospectively unobservable, and endpoint counts or venue timestamps cannot substitute for a reproducible rule–evidence proof.
Source: §12 · p. 11Public sources may be revised or overwritten, venues may use evidence invisible to researchers, and sample-weighted claims do not automatically generalize exchange-wide under uneven coverage or nonresponse.
Source: §11 · p. 10
Abstract
from arXiv · showhide
Public evidence can become sufficient to settle a prediction-market contract before the venue records its first determination, but the relevant boundary depends on the applicable rule version, exact release object, source hierarchy, correction history, and unfinished contract conditions. This paper defines two Kalshi clocks: first decidability, the earliest contemporaneous singleton in the rule-evidence mapping, and stable decidability, the retrospective earliest time after which the same singleton remains unchanged through finalization. A completed retrospective identification test establishes a narrow feasibility result. In a frozen blind pilot, all 25 identities and blinding checks passed and current rule text was recovered for all 25; no exact or bounded historical rule version and no exact or bounded official source-release object was recovered. The full historical recovery covered 152,694 ordinary tickers, 11,530 exact event identities, and 6,540 read-only official requests, with zero historically eligible events and zero historically eligible tickers. This is an observability result, not a claim that no market was decidable or that public evidence never existed. A completed prospective infrastructure shakedown established observation capability for three source programmes across 25 markets, with integrity revalidation of 781,266 lifecycle frames, 22 closed lower-bounded reconnect receipts, no unresolved reconnect gap, no due-but-missed official release, and an inactive price layer. Production evidence enrollment is active. The prospective sample is constructed only at enrollment close from prospectively frozen identities and pre-outcome fields; its frozen target size is selected mechanically under the registered full, reduced, exploratory, or no-go support disposition. No contractual-decidability clock, human-adjudication, price, or cross-venue result is reported here.
1 Introduction
The paper separates public contractual sufficiency from venue action, defining first and stable clocks for when admissible evidence implies one settlement value. Its rule–evidence model makes the applicable rules, evidence objects, and remaining contract conditions part of that determination.
2 Rule–evidence model
- First decidability is the earliest contemporaneous time when the settlement mapping is a singleton and no contract-required future condition remains; stable decidability is retrospective.Later source corrections can reopen an early singleton, so stable decidability is a robustness clock rather than traders’ real-time information set.Source: §1 · p. 2, §2 · p. 3
1 Introduction
- Because manual adjudication is costly, the paper uses a reproducible probability sample with explicit weights, blind review, and measured attrition rather than promising exchange-wide clocks.An unidentifiable clock is treated as a public-governance observability property, not a value to replace with the venue timestamp.Source: §1 · p. 2
- The target is conditional on a listed contract: the question is when its contemporaneous rule graph and public evidence first and stably imply one admissible settlement value.The paper distinguishes this boundary from settlement legibility, which concerns which uncertainties become contracts at all.Source: §1 · p. 2
2 Rule–evidence model
- A rule–evidence graph combines the applicable rule graph and admissible public evidence to determine whether the settlement-value set is a singleton.Its recorded components include exact source declarations, thresholds, time zones, early-close clauses, and dated details.Source: §2 · pp. 2–3, Figure 1 · p. 3
- A named source and publication timestamp alone do not identify contractual decidability, because contract terms or pending conditions can change the admissible outcome set.Different strikes, exceptions, time windows, source hierarchies, and future conditions can map the same release to different outcomes.Source: §2 · p. 3
- Interval evidence identifies latency bounds, but not a point estimate without an additional model locating the clock within the interval.Venue determination and final outcomes cannot define the blind upstream clock without introducing hindsight.Source: §2 · p. 3
3 Retrospective identification test
The completed historical test found that the available public record did not support reconstruction of the version-bound rule and official release objects required for the registered clock.
3 Retrospective identification test
- 0 historically eligible events and 0 historically eligible tickers were identified across a recovery covering 152,694 ordinary tickers, 11,530 exact event identities, and 6,540 read-only official requests.The historical frame is a venue-side feasibility denominator, not a source of the missing version-bound rule and official release objects.Source: §3 · p. 4
- 25/25 pilot identities passed auditable-identity and blinding checks, and current rule text was recovered for 25/25, but no exact or bounded historical rule version or official release object was recovered.Historical rule-version recovery was 0/25, as was recovery of exact or bounded official source-release objects.Source: §3 · p. 4
- The result is retrospective unobservability at the accepted evidence grade, not evidence that contracts were undecidable or that public evidence never existed.The required historical binding could not be reconstructed from the available public record.Source: §3 · p. 4
4 Prospective evidence-capture amendment
The prospective shakedown established source-observation capability and infrastructure integrity, while prospective enrollment follows registered timing and support rules.
4 Prospective evidence-capture amendment
- Observation capability was established for the BLS CPI, FOMC target-rate, and Federal Register executive-order programmes across three events and 25 markets.These programmes demonstrate observation capability only, not an empirical cohort; the Federal Register subset was conformance-only.Source: §4 · p. 4
- 781,266 lifecycle frames passed full integrity revalidation, 22 reconnect receipts were closed lower bounds, and no unresolved reconnect gap or due-but-missed official release was found.The price layer remained empty, and the shakedown established observation capability rather than a contractual-decidability result.Source: §4 · p. 4
- Prospective enrollment starts at TPROD and uses a 30-day primary interval, with a registered 42-day total ceiling and hard follow-up through 2026-11-03.Only postenrollment receipts may support the prospective study; shakedown observations are excluded by default unless a separately frozen no-outcome-leakage rule passes.Source: §4 · pp. 4–5
5 Sample design and prospective frame
The prospective frame is sampled at enrollment close from frozen identities using only pre-outcome fields. A registered support disposition sets the target size, and deterministic stratified allocation and independent review govern selection.
5 Sample design and prospective frame
- The frame is built at enrollment close from prospectively frozen identities and pre-outcome fields, excluding venue outcomes and later price paths from sampling and initial review.After the initial review is preserved, later linkage reveals venue clocks and final values for consistency checks and latency construction.Source: §5 · p. 5
- The frozen target is 500, 150, 50, or 0 markets under full, reduced, exploratory, or no-go support, respectively, using prespecified support thresholds.For a simple random full-support sample, 500 yields an approximate 95% margin of error of 0.044 for a proportion; design features and attrition can widen it.Source: §5 · p. 5
5.1 Pre-outcome strata and deterministic allocation
- Strata use pre-outcome source class, market form, and expected horizon, with sparse cells merged deterministically and allocation based on square-root stratum size.Eligible non-empty strata receive at least two observations, capped at stratum size, with deterministic largest-remainder adjustment and ordering.Source: §5.1 · pp. 5–6
5.2 Pilot and independent second review
- The completed 25-market historical pilot is separate from the prospective frame, and independent second review targets at least 20% of the sample.The 100-packet minimum is a target bounded by the realized sample; a conflict-enriched audit remains separate from the probability-based reliability subset.Source: §5.2 · p. 6
6 Population and lifecycle authorities for sampling
The paper distinguishes venue-side population frames from the evidence needed to construct contractual-decidability clocks. Sampling and initial adjudication remain blind to venue outcomes, while archived rule and source objects and later venue linkage serve separate roles.
6 Population and lifecycle authorities for sampling
- The MVE population contains 7,611,594 exact tickers and 5,262,526 exact event keys, but supports dependence diagnostics and a small exploratory stratum, not unrestricted child-ticker adjudication.The historical venue-side frame is a feasibility denominator, not a historically admissible clock cohort.Source: §6 · p. 6
6.1 Blinding boundary
- Initial review excludes venue results, determination and finalization times, endpoint class, and later price paths; later linkage occurs only after that review record is preserved.Contradictions create a new adjudication version rather than overwriting the initial record.Source: §6.1 · p. 7
6.2 No broad rules crawl
- Packet construction relies on prospectively archived versioned rules, official source objects, contemporaneous receipts, and revision or supersession graphs for the selected sample.Later retrieval can verify current availability but cannot create missing valid-time evidence; the 7.6-million-object MVE population is not a rules-crawl denominator.Source: §6.2 · p. 7
7 Blind adjudication protocol
- Blind adjudication first selects the applicable release object and evaluates rule predicates, then venue clocks and final values are linked after the initial record is preserved.At least 20% of the sample, targeting 100 eligible packets when available, receives independent second review; agreement and clock-boundary measures are reported separately.Source: §7 · p. 8
8 Estimands and partial identification
The estimands distinguish first from stable decidability and treat interval-valued clocks through bounds rather than midpoint estimates. The analysis also separates conflict cases and reports results by stratum with sampling weights.
8 Estimands and partial identification
- If stable decidability occurs after venue determination, a negative stable-clock latency is classified as a conflict, not as negative delay.The conflict indicates venue action before the publicly archived rule–evidence sequence became retrospectively stable.Source: §8 · p. 9
- The bounded share of close-to-determination time occurring after first decidability replaces an unbounded ratio that could exceed one when decidability preceded close.Source: §8 · p. 9
- Exact and interval clocks are reported separately, and interval observations bound the share determined within a threshold rather than using midpoints.The lower and upper indicators supply the bounds.Source: §8 · p. 9
- Estimates are reported by stratum with sampling weights, using source release as the primary cluster-bootstrap dependence unit and event or series sensitivity checks.Source: §8 · p. 9
9 Registered empirical questions
The registered empirical questions cover identification yield, residual latency, revisions, heterogeneity, and reviewer reproducibility. The questions specify descriptive and reliability measures without presuming their outcomes.
Q1 — Identification yield
- Q1 asks what weighted share of sampled markets yields an exact first clock, interval first clock, stable clock, or no admissible clock.Source: §9 · p. 10
Q2 — Positive residual latency
- Q2 asks for the identified share with LFD exceeding timestamp or interval precision tolerance, reported as exact and bounded shares.Source: §9 · p. 10
Q3 — Revision sensitivity
- Q3 asks how often first and stable decidability differ and which source or rule mechanisms account for that gap.Source: §9 · p. 10
Q4 — Heterogeneity
- Q4 describes how identification yield and LFD vary by source class, market form, horizon, and pre-outcome rule features.Multiplicity correction or hierarchical shrinkage is specified if many contrasts are retained.Source: §9 · p. 10
Q5 — Reviewer reproducibility
- Q5 asks for raw agreement, κ, clock-interval overlap, and adjudicated boundary differences on the random audit subset.Source: §9 · p. 10
Registered controls
- Pseudo-source releases and generic homepage timestamps serve as negative controls, while venue close and determination-as-decidability are specified as substitutions or pathologies.Source: §9 · p. 10
10 Validity and reproducibility
The design requires reproducible sampling and evidence review without post-outcome timing fields. Markets lacking required timing evidence are excluded from timing analysis but remain in observability and attrition accounting.
10 Validity and reproducibility
- The design requires reproducible sampling without post-outcome timing fields, recorded rule versions, exact source-object identity or explicit failure classification, blind initial adjudication, and independent review.A market without the required timing evidence is excluded from the corresponding timing analysis but remains visible in observability and attrition accounting.Source: §10 · p. 10
- Markets lacking required timing evidence are excluded from the corresponding timing analysis while remaining visible in observability and attrition accounting.Source: §10 · p. 10
- The analysis distinguishes sample and weights, source-object attrition, clock grades, reviewer reliability, latency bounds, heterogeneity, revisions, conflicts, and negative controls.Source: §10 · p. 10
11 Limitations
The paper distinguishes public contractual sufficiency from venue action and reports a historical observability boundary, while prospective enrollment remains short of clock results. Its classifications retain interval, ambiguous, conflicting, and unmeasured cases.
11 Limitations
- Contractual sufficiency may be ambiguous, sources may be revised or overwritten, and venue evidence may be unavailable to researchers; stable decidability is retrospective, not traders’ real-time knowledge.Judgment remains despite blinding and replication; sample-weighted claims do not automatically generalize exchange-wide under uneven coverage or nonresponse, and long LFD is not a causal estimate of venue inefficiency.Source: §11 · p. 10
12 Conclusion
- Historical recovery could not reconstruct the version-bound rule and official release objects needed for the contractual-evidence clock, without showing that markets were undecidable or public evidence never existed.Source: §12 · p. 11
- Prospective observation capability and active enrollment are reported, but the semantic measurement task and clock results remain pending, conditional on registered support thresholds at enrollment close.No human adjudication, clock value, latency estimate, price study, or accepted cross-venue pair is reported.Source: §12 · p. 11
- Population scale, venue finality, and public contractual sufficiency are distinct layers; endpoint counts, determination timestamps, and final results cannot substitute for a reproducible rule–evidence proof.Source: §12 · p. 11
A Adjudication classifications
- Adjudication classifications retain exact or interval clocks, rule/source ambiguity, multiple plausible releases, conflicting primary evidence, and unmeasured cases; reconciliation preserves initial judgments.Source: A Adjudication classifications · p. 11
Data and Code Availability
The registered KMVE population and hierarchy are available, but raw active enrollment capture and empirical production-enrollment results are not distributed. The paper also records its AI assistance, funding, and competing-interest disclosures.
Data and Code Availability
- The registered MVE population and hierarchy are available as the KMVE dataset, but raw active production-enrollment capture and empirical enrollment results are not distributed.No sampled ticker identities, contractual-decidability clocks, adjudication outcomes, price results, or accepted cross-venue pairs are released with the preprint.Source: Data and Code Availability · p. 11
Generative AI Disclosure
- ChatGPT and Codex provided editorial and technical assistance; the author made substantive research decisions, reviewed the manuscript, and assumes responsibility for its contents.Source: Generative AI Disclosure · p. 12
Funding
- The research received no external funding.Source: Funding · p. 12
Competing Interests
- The author reports no external sponsor influence on research decisions and states that the article evaluates no commercial product and makes no investment recommendations.Source: Competing Interests · p. 12