# Pilot temporal-thinning method

This retrospective sensitivity uses all 24 city-days, sizes 10/50 and horizons 5/15/60/180 minutes. The existing article's ten-share, sixty-minute comparison was fixed before this sensitivity calculation. Prior research findings were already known; this is not preregistered or unseen evaluation.

Partition entries by event, numeric outcome token, size and horizon. Order by UTC decision time, then candidate ID and composite entry identity. Retain the first entry and the next only at or after the selected entry's full planned horizon. An entry with a missing exit reserves the same horizon; excluded entries do not extend it. Selection does not inspect future exit availability, prices, payouts or return values. Forecast and observation updates share each token's reservation.

The original 3,354 entry rows expand to 13,416 planned comparisons: 8,144 selected and 5,272 excluded. Missing exits remain unknown. Every city-day appears in both populations and all size/horizon groups. All eight original groups reproduce exactly, and an independent implementation matches every selected composite identity. Candidate IDs alone are not unique across outcomes and sizes.

Positive means greater than 1e-9 USD, preserving the original diagnostic. Fee coefficients 0.05 and 0.10 use the inherited exponent-one curve and displayed-level rounding scenario. They are not flat notional percentages or certified historical fees. Temporal thinning leaves dependence across sibling markets, weather updates, sizes and horizons. No portfolio return, confidence interval or cross-venue survival rate follows.

See the [method and source-hash summary](../evidence/pilot/summary.json), [all 592 groups](../evidence/pilot/groups.csv) and [independent follow-up validation](../evidence/pilot-followup.json). The full selection ledger and original scenario rows remain in the repository; their hashes are retained in the summary and reviews.

Source method: docs/ASSYMETRIX-PILOT-ROBUSTNESS-METHOD.md. Original SHA256: `022c7934ca64f56298bfd704e61f6570f6d1c4213a816be4a5ad9e040604daf1`. This portable note is an explicit condensed editorial derivative, not the complete internal method document.