MISSION, `dps_cap`: what the precision cap in the rival interfaces costs

A bounded measurement run. One quantity, two precisions.

Scope

zeta/epstein.py hands the counterexample battery three rival interfaces, and each one silently caps the precision it was asked for:

line	caller	cap
1073	`epstein_interface.zeros_on_line`	`Z_epstein(..., dps=min(dps, 15))`
1092	`epstein_interface.count_zeros_box`	`count_zeros_box(..., dps=min(dps, 20))` and `epstein_completed(..., dps=min(dps, 20))`
1126	`zeta_interface.count_zeros_box`	`dps=min(dps, 20)`, `xi(..., dps=min(dps, 20))`
1142	`dh_interface.count_zeros_box`	`dps=min(dps, 20)`

The caller's dps is accepted and then discarded. Nothing in the returned dict records that it was. This run measures the size of that discard at one point, and does not touch zeta/.

The measurement

abs(epstein_completed(mpc('0.8','85.7'), (2,1,3), dps=D)) at D = 20 (what the cap delivers) and D = 60 (what the caller asked for), plus a convergence ladder to establish which of the two is the number and which is noise.

What this run does not establish

One point, one form, one function. It bounds nothing about the heights at which the battery actually calls count_zeros_box, and it is not a claim that any published battery verdict is wrong. It measures a cost, not a defect in a result.

id: dps_cap
question: What does the min(dps, 20) cap in epstein.py's rival interfaces cost on a completed-Epstein evaluation high on the strip?
frontier: uncharacterised, the caps are undocumented at their call sites and no test pins the accuracy the capped path delivers
proposed_attack: evaluate the same quantity at the capped precision and at the requested precision, and separate signal from roundoff with a convergence ladder
dead_routes:
  - reading the returned dict for a precision warning, the interfaces record no provenance for the discarded dps
required_oracles:
  - internal convergence of the same routine at dps 80 and 100, agreeing to all digits printed
  - an independent recomputation by a separate process, reported as agreement or disagreement in order of magnitude
kill_conditions:
  - the two precisions agree to within an order of magnitude, making the cap free at this point
  - the dps=60 value fails to stabilise against dps=80 and dps=100, making it noise too
  - the quantity is shown to be outside any range the battery evaluates, making the cost unreachable
agents_may:
  - measure
  - code
  - attack
  - report a candidate for the ledger
agents_may_not:
  - declare novelty
  - declare theorem status
  - promote their own claim
  - edit zeta/epstein.py

id: dps_cap-2026-08-14-run1
hunt: dps_cap
started: 2026-08-14T00:00-05:00
finished: 2026-08-14T00:00-05:00
ran:
  - python hunts/dps_cap/measure.py
  - python -c convergence ladder at dps 30, 40, 80, 100
outcome: the capped precision returns a magnitude 24 orders too large at this point, with no correct digits; the requested precision converges
artifacts:
  - hunts/dps_cap/results.json
  - hunts/dps_cap/HANDBACK.json