At 0.8+85.7i, epstein_completed at the capped dps=20 returns noise around 1e-33 to 1e-35 where the converged value is 1.6e-58, a factor of about 1e23 to 1.9e25 with no correct digit; dps=60 buys about 16 real digits, not 60, because cancellation costs roughly 44 of them.
hunts/dps_cap (Record 16 of 98 in chronological sequence)
Guiding Question
What does the min(dps, 20) cap in epstein.py's rival interfaces cost on a completed-Epstein evaluation high on the strip?
Method & Verification
Five concurrent 2026-08-14 runs evaluated abs(epstein_completed(0.8+85.7i, (2,1,3), dps=D)) on a precision ladder; they disagree on the exact dps=20 noise magnitude because parse precision moves the floor, and they agree the capped value has no correct digit.
Lineage & Relationships
gate5_p6_a ↗, gate5_p6_b ↗, gate5_p6_c ↗, support_e6241336 ↗
Primary Sources (at pin 8fa46e134)
Editorial Notes
Hunt #81 and Hunt #113 are duplicate case-log entries for the same directory, both Status: probe, complete, with the same measurement text. They are not two sequential outcomes. The directory landed 2026-08-21 as the unmerged union of five concurrent 2026-08-14 runs (live-b4528bd7, live-aeb4e5de, live-f12f9441, live-e72925d1, support/e6241336). Handbacks all say settled; the case log says probe, complete. Reconciled as one completed measurement. Concurrent runs disagree on the exact dps=20 magnitude (3.1e-33, 4.85e-35, 3.39e-34) because at capped precision the returned magnitude is not a function of the point; the headline keeps that spread rather than picking one noise value. The independent confirmation of the converged 1.6e-58 is the separate support hunt support_e6241336. The 'Renumbered twice' footnote that follows these case-log entries is about lambda_dh_bounds and prime_zeta_rightmost, not about dps_cap.
Date Provenance
commit e630e400731afbb5533a2bae4c3c6fd49da00e32, hunts/dps_cap/, author 2026-08-14T07:06:02-05:00