Quarterly-model auction extension - November 7, 2026
Recorded before calculating the quarterly auction combinations. The annual
auction results and quarterly-versus-annual comparison have already been seen.
The user selected quarterly refitting for prospective use after that comparison
and then requested testing auction additions against the quarterly baseline.

Use precisely the original auction_study/PROTOCOL.txt data gate, four event
flags, two vote directions, original-tenor classification, annualized metrics,
matched 20-year period, timing tests and reporting. Refit every 13 weeks rather
than 52 for BOTH the baseline and each augmented model, using expanding history
and the January 6, 2006 anchor. C=0.1; unit positions; no new thresholds, sizes,
interactions, calendar definitions or auction fields. Eight combined variants.

Choose by highest early gross calendar-week HAC t, with at least 20 early
events; ties by candidate then sign. Save the early choice before computing
later comparisons. Early outcomes end before November 1, 2016. Later sample
and 2026/last-52 windows are unchanged. Nominal 20-year remains ineligible for
early selection; show both signs from May 15, 2020 against the matched baseline.

Disclose every variant and paired gross/net differences against the quarterly
baseline and constant higher-yield benchmark. Original annual results remain
separate. Event-only results are unchanged by model-refit frequency and must
not be represented as a new independent confirmation.

This is a second exploratory auction round informed by earlier results, not
an untouched test or search-adjusted significance calculation. Do not silently
adopt an auction flag into the selected quarterly production specification.
