Paper contents
I · Research question
What can public labels and polling establish without inventing synchronization?
An announcement about a delisting, network upgrade, or transfer suspension has at least two times: when it was published and when its text says an operational change should occur. Independent polling supplies a third time: when a public interface was first detected in a corresponding state. None is automatically a direct observation of the real operational flip.
The paper therefore asks a descriptive question rather than a return-prediction question: what can a carefully documented announcement label set and a short public-status polling study establish about coverage, timing, and observability? Missing coverage is coded as a gap, not a zero. An unmatched transition is not proof that no announcement existed anywhere.
Research 01 and 02 measure and frame net carry. Research 07 adds the operational-event, announcement-time, and transfer-status observation layer; it does not validate their return estimates.
Research 01 · Measuring Executable Net Carry Research 02 · Framing Net Carry
II · Data & method
Separate the record, the event, the key, and the interval.
Dataset A covers 2026-04-23 through 2026-10-20 UTC: 180 elapsed days, or 181 calendar dates when both endpoints are included. Its 20-venue target map contains 996 labels from ten venues with nonzero events, an observable zero at Pacifica, and nine venues with full-window or historical gaps. Publication and effective times remain separate; 253 null effective times are not imputed.
Dataset B contains 502 polling rounds at Bitget, KuCoin, Gate, and LBank over a 169.1-hour log span. A known gap of approximately 13.2 hours and other observational limits remain explicit. Q1, Q2, Q3, and duration use different denominators and are never pooled as if they were interchangeable.
Source and coverage map
Semantic event deduplication, rescheduling links, separate timestamp granularities, and an explicit distinction between observable zero and unavailable history.
Q1 · unmatched detected transition
A key-by-transition record has no semantically corresponding contemporaneous announcement in the saved auditable snapshot range. It is not an independent-event count.
Q2 and Q3 · strict evidence and timing
Q2 requires observable keys, due execution, sufficient polling, and a contrary state. Q3 compares detection with publication on unique matched announcement events.
Gap-aware duration
Continuous closures, gap-crossing endpoint pairs, right-censored records, and openings without an observed stop remain separate partitions.
A negative Q3 lag is not a proven flip order.
It means that the recorded detection preceded the announcement publication timestamp. The actual operational flip is usually latent between polls, and nominal source time is not always proof of first public visibility.
III · Results
Coverage, timing, and batch dependence—shown at record level.
Scroll horizontally or use the arrows. Select a figure to inspect its details.
IV · Robustness & evaluation boundary
Robustness means exposing composition—not manufacturing a confidence interval.
Fixed time subsamples, venue leave-one-out rules, lower and upper detection bounds, extreme-retrospective exclusions, and AMPL, GRAM, ERC20, and large-batch exclusions test whether descriptive summaries depend on a particular record mix. They do not convert venue-clustered, batched, nonrandom observations into independent draws.
The 720-pair median is 46.98 hours. Early and late medians are 45.70 and 47.83 hours; venue exclusions range from 41.48 to 58.00 hours in the reported leave-one-out set.
The pooled 54.14% is highly concentrated: LBank supplies 398 of 503 Q1 records. Removing Bitget, KuCoin, or LBank changes the denominator as well as the numerator.
The recorded median is +5.81 minutes. Lower- and upper-bound substitutions give a deterministic median envelope of +5.43 to +14.89 minutes—not a 95% confidence interval.
Excluding LBank ERC20 changes the median from 0.648 to 0.915 hours; excluding all same-start batches of eight or more leaves 75 intervals and a 5.028-hour median.
There is no forecast target, return model, trading rule, or model search. Diebold–Mariano tests, out-of-sample forecast evaluation, full-cost economic significance, and multiple-model corrections are therefore not applicable.
V · Evidence boundaries
No claim of execution, causality, complete coverage, or trading performance.
A publication timestamp is not an execution timestamp, and a detected status change is not proof of the latent operational flip or its cause.
Missing announcements remain gaps. Q1 means no contemporaneous semantic match in the audited surface—not that no announcement existed anywhere.
Q1 uses key-transition records, Q3 uses announcement events, and duration uses observed close–open intervals. Their denominators cannot be converted into a universal venue rate.
The paper reports no return forecast, realized tradability, executable profit, market impact, venue ranking, or investment recommendation.
VI · Limitations & ethics
Unknown remains unknown.
- Coverage is incomplete. Nine target venues have full-window or historical gaps. A gap is not zero, and Gate is not assigned a Q1/Q2 rate.
- Annotation is single-person. An 88-item single-person review does not establish inter-annotator agreement or an independent error rate.
- Polling detects rather than directly observes flips. Long gaps, round ordering, processing bounds, field oscillations, and many-to-many key mappings restrict interpretation.
- Replication is bounded. Released data can regenerate all reported tables and figures, but cannot reconstruct unavailable raw historical logs or the earlier 90-day baseline.
- Redistribution is intentionally limited. The public release contains labels, public URLs, and derived audit data—not announcement text, private notes, user data, nonpublic interfaces, or access-control circumvention.
Not investment advice. This paper is for research and methodological information only. It does not provide investment, legal, accounting, or tax advice; does not recommend any trade or leverage; and does not claim execution, profitability, or market-impact effects.
VII · Conclusion
An operational information layer with visible seams.
The release contributes a documented announcement dataset, an observation-first pairing framework, and a four-venue polling study whose gaps and counting units remain inspectable. Its central finding is methodological: publication time, text-stated effectiveness, detected status, and actual execution cannot be collapsed into one timestamp.
The data can support future research on operational information and market access. They cannot, by themselves, establish causal price effects, realized tradability, profitable execution, or universal venue rankings. Future predictive or trading studies must supply their own outcome definitions, temporal controls, cost models, and evaluation designs.