SPX sessions closed in positive gamma, with the net-gamma percentile under 20, on expiration days

7 sessions · the archive filtered by the dealer book · all sessions · the calendar

7 of 1,106 sessions match. Of them 100% closed above the flip (archive 84%), 43% held the expected-hold band (archive 60%), mean open-to-close move 0.42% (archive 0.60%).

Every row is a finished SPX session and links to its page, where the same numbers stand with the day's events and a free minute-by-minute replay. The filters are the closing measurements those pages print: the zero-gamma flip against the close and the open, the net-gamma percentile, how often the flip was crossed, whether the expected-hold band held, and the calendar event the day carried. The shares above are counts over the matched days next to the same counts over the whole archive; a share that matches the archive is a base rate, not a pattern.

SessionCloseMoveFlipvs flipPctlCrossBandEvents
2026-09-187,648.27+0.09%7,574.3+74.001brokeOpex
2026-07-177,458.16+0.12%7,378.7+79.531heldHousing, UMich, Opex
2025-05-305,909.25+0.27%5,771.5+137.8160heldPCE, Opex
2024-05-315,281.85+0.70%5,180.6+101.300brokePCE, Opex
2023-12-154,715.45+0.05%4,673.6+41.801heldOpex
2023-04-214,133.90+0.10%4,133.8+0.153brokeOpex
2023-01-203,971.87+1.62%3,869.2+102.700brokeOpex

Turn a filter into a rule

A count is where a question starts. The Lab takes a rule in plain words ("sell the 0DTE straddle at 10:00 on days that opened below the flip with the percentile under 20"), compiles it against every archived session and reports the out-of-sample verdict with day-clustered intervals. An assistant with the MCP server can page through the same archive with list_sessions and get_session.

Measurements are model output derived from public trade and quote data; they are approximations and can be wrong. Nothing here is investment advice; see the Terms.