zeta/epstein.py caps working precision in one place inside the rival interface it hands the battery: epstein_interface(...)["count_zeros_box"] passes dps=min(_d, 20) both to count_zeros_box and to the epstein_completed closure it winds (zeta/epstein.py:1091-1093). Every other entry in that dict passes the caller's dps through untouched.
This hunt measures one number at two precisions and reports both magnitudes. It changes nothing in zeta/.
The quantity:
abs(epstein_completed(mpc('0.8','85.7'), (2, 1, 3), dps=D)) for D in {20, 60}
It is deliberately off the critical line and high up the strip, which is where the incomplete-gamma split in epstein_completed is at its most cancellation-prone: Lambda_Q decays like a gamma factor in |t|, so the true magnitude at t = 85.7 sits far below anything 20 digits of working precision can resolve against the O(1) terms that build it.
Scope. One number, two precisions, one comparison. Not a study of the cap's effect on count_zeros_box verdicts, not a fix, not a claim about whether the cap is wrong for the job it was written for. Those are threads, recorded as such in HANDBACK.json, not pursued here.
id: dps_cap
question: What magnitude does abs(epstein_completed(0.8+85.7i, (2,1,3))) take at dps=20 versus dps=60, and does the dps=20 value carry any correct digits?
frontier: unmeasured; the cap at zeta/epstein.py:1091-1093 has no recorded cost
proposed_attack: evaluate the same quantity at both precisions and compare magnitudes
dead_routes:
- reading the cap's cost off the docstrings, which record defects only at dps 30-70 and never above t = 20
required_oracles:
- independent evaluation at higher working precision
- self-consistency of the value under further precision increase
kill_conditions:
- the two magnitudes agree to within an order of magnitude, so the cap costs nothing here
- the dps=60 value fails to stabilise under a further precision increase, so the reference itself is noise
agents_may:
- measure
- code
- report both magnitudes with their working precisions named
agents_may_not:
- edit zeta/epstein.py
- generalise from one point to the cap's effect on any battery verdict
- promote their own claim