Identify the ordering of tied print events within receipt milliseconds

Establish the within-receipt-millisecond ordering of tied Polymarket trade-print rows and determine how that ordering affects subsequent trade-sign calculations.

Background

The trade-classification procedures use the archive’s receipt timestamps and exclude quotes stamped in the same millisecond as a print because the relative message order is unavailable. Print rows sharing an asset and timestamp are ordered using the cached row order, but that ordering is not known to reflect the venue’s actual event sequence.

Because tick-based signs carry forward the last nonzero sign, unresolved ordering can affect not only the classification of tied prints but also later signs. The paper therefore treats the issue as an unresolved data-ordering problem rather than claiming that removing tied rows solves it.

References

Removing those rows from scoring, while retaining their history, leaves 9.3~million common observations and raises Lee--Ready's equal-day balanced accuracy to 0.957. That is a substantial sample change, and cannot establish the unknown within-tie order or remove its effect on later signs.

— Measuring Trade Direction in a Prediction Market: Settlement Ground Truth and Trading-Cost Measurement on Polymarket  (2610.06412 - Dubach, 5 Oct 2026) in Section 4.2, “Timestamp ambiguity and quote timing”