Moira 6.8.2 - KernelPool Native Planetary Evaluator & Ephemeris Graph Routing

Release date: 2026-09-20. Upgrade path: 6.8.0 to 6.8.2.

Moira 6.8.2 delivers critical engine performance and architectural optimizations when operating with large sharded small-body catalogs. It restores high-speed compiled C++ planetary reduction under pooled kernels and optimizes SPK graph traversal from linear search to $O(1)$ indexed lookup.

Key Changes

Native Planetary Evaluator Admission (planets.py)

• Admitted KernelPool into NativePlanetaryEvaluator reduction planning.
• Added _resolve_primary_spk_reader() to extract the authoritative planetary JPL ephemeris (de441.bsp) from a heterogeneous reader pool.
• Restores 0.53 ms 10-planet chart reduction speed in pooled environments (a 50,000× speedup over pure-Python fallback under large shard sets).

Indexed Graph Routing & Candidate Pruning (spk_reader.py)

• Implemented $O(1)$ direct edge lookup (_find_direct_edge_tdb) using pre-indexed pair mappings (_pair_readers).
• Added candidate-pruned BFS routing (_find_candidate_route_tdb) with strict hub pruning against the planetary clock spine ((0, 3), (3, 399), (0, 10), (3, 301)), preventing minor planet shards from acting as graph hubs.
• Added bounded LRU route caching (_route_cache) with exact epoch validity intervals, reducing repeated small-body lookups to 0.07 ms.
• Optimized clock route evaluation in _ephemeris_kernel_identity_at_tdb to query only planetary spine readers, reducing segment checks from 2,355 down to 5 per epoch conversion.

Small-Body Catalog Expansion & Release Memoization

• Includes official sealed 449-shard 2026.09.18.1 minor planet catalog expansion (11,223 numbered bodies, 214 Centaurs, 1,048 TNOs) with 100% outer solar system coverage.
• Added thread-safe memoization (_VERIFIED_RELEASE_CACHE) in small_body_catalog_release.py.

Native Quadrant House Systems & Nutation Epoch Caching (houses.hpp, nutation.cpp)

• Accelerated calculate_houses from 128.6 µs to 18.8 µs ($6.85\times$, >53,000 charts/sec).
• Implemented compiled C++ solvers for Placidus (P), Koch (K), Regiomontanus (R), Campanus (C), Porphyry (O), Equal (E), and Whole Sign (W) in src/native/include/houses.hpp, running in 0.45–1.9 µs natively with sub-picodegree parity ($\le 10^{-11 \circ}$).
• Implemented thread-local circular epoch cache in native IAU 2000A nutation (nutation_2000r06), dropping warm-epoch angle reduction from 45 µs to 0.78 µs ($57\times$ faster) and nutation lookup to 0.19 µs ($197\times$ faster).
• Verified against Python reference over 10,000+ historical epochs and configurations spanning $-3000 \to +3000$ CE.

Verification Scope

• Planetary architecture invariants (tests/unit/test_planets_architecture_invariants.py): 20 passed in 54.3s, confirming C++ native evaluator admission under KernelPool and parity with pure-Python reference reductions to within $10^{-12}$ degrees.
• Live production benchmark: chart reduction HTTP 200 in 1.50 ms, liveCards figure tRPC query in 14.12 ms, calculator compute mutation in 21.35 ms.

Install

pip install moira-astro==6.8.2