Skip to content

paran_performance_validation_2026 07 12

Jump to bottom
TheDaniel166 edited this page Oct 15, 2026 · 3 revisions

Paran Performance Validation

Date: 2026-07-12

Status: Python fast path admitted; the subsequent native heliacal parity repair is documented separately and is now admitted

1. Governing Object

The optimized product is unchanged: exact rise, set, upper transit, and lower transit events within one 24-hour window, combined into paran pairs under the existing time-orb doctrine. Analytic geometry and Newton iteration provide brackets or estimates; the exact altitude/hour-angle signals remain the event authority.

2. Baseline

Measured in the project .venv on the local Windows reference checkout after kernel/catalog warm-up where applicable:

Operation Before
_lst 0.076 ms
stellar transit 14.4 ms
stellar find_phenomena 100.9 ms
stellar _crossing_times 160.0 ms
Mars _crossing_times 298.5 ms
67-latitude field column approximately 45 s in the supplied audit

The focused pre-edit truth slice passed 95 tests.

3. Admitted Changes

  1. Reuse the one nutation evaluation when constructing apparent sidereal time.
  2. Seed meridian transit from the forward hour-angle displacement, perform at most eight Newton refinements, verify the exact residual and 24-hour window, then fall back to the original scan.
  3. For fixed stars, derive daily altitude extrema and hour-angle brackets from center-window RA/Dec, then refine the exact altitude signal.
  4. For planets, use center-window RA/Dec only to estimate brackets; exact topocentric altitude remains authoritative and failed estimates fall back to scanning.
  5. Split paran horizon and meridian collection so latitude-independent stellar MC/IC events can be reused in a field column.
  6. Add explicit request/field-scoped crossing reuse. There is no global cache; planetary keys retain the active reader identity.
  7. Remove the prior duplicate MC/IC computation inside _crossing_times.

4. Parity Evidence

  • Newton-first transits are compared with scan/bisection truth for all ten planets plus Regulus and Acrux, upper and lower transits, at five latitudes from -66 to +66 degrees; tolerance is 0.5 seconds.
  • Fixed-star horizon events are compared with scan/bisection truth for five stars over the full -66 to +66 degree sweep in two-degree increments; event-set equality is exact and timing tolerance is 0.5 seconds.
  • Planet horizon estimates are pinned against scan truth for all ten planets at -66, 0, and +66 degrees in the normal suite. A full two-degree sweep is part of the implementation acceptance receipt.
  • The existing offline Horizons rise/set and paran fixtures remain required.
  • Circumpolar diagnostic states are compared against the prior sampled states, including a grazing fixed-star geometry case.

5. Performance Receipt

Run:

$env:MOIRA_NO_DOWNLOAD = "1"
$env:MOIRA_PERF_ASSERT = "1"
.\.venv\Scripts\python.exe tests\benchmark_paran_performance.py

Observed after implementation:

Operation After
_lst 0.043 ms
stellar transit 0.53 ms
stellar _crossing_times 11.1 ms
warmed Mars _crossing_times 22.2 ms
67-latitude field column 1.09 s
field sample average 16.3 ms
53-available-star packet core 0.51 s

These are local performance evidence, not astronomical validation. The script's absolute gates are opt-in because timing varies by host.

6. Native Heliacal Boundary

This document originally gated native heliacal dispatch after a controlled Sirius custom-policy probe demonstrated divergent setting results. That gate was correct at this stage. The later repair replaced the defective nutation model, aligned the observables and complete policy, added independent Python/native differential tests, and admitted native search by default. See native_heliacal_parity_validation_2026-07-12.md for the separate evidence.

Clone this wiki locally