Vedaksha v10.0.0: accuracy, speed, and a much larger Vedic surface

ELP/MPP02 by angle addition

Each term's phase is a sum of multiples of a few fundamental arguments. The series now evaluates the sine and cosine of those fundamentals once and builds every term's from them with the addition formulas, through a prefix tree shared between terms, instead of one sine and cosine per term. The answer moves by at most 7.5e-7″ over 1610 to 2390, against the series' own arcsecond-level fit.

VSOP87A by angle addition

The same idea over the frequencies the terms share. Earth has 624 distinct frequencies among its 2,202 terms, so it needs one sine and cosine per frequency. Output moves by at most 5.7e-8″ over 1900 to 2100.

The Moon's light time

The light-time iteration returns the evaluation it already has when the next retarded instant is the same double, and takes its first step from the distance series alone where the provider offers one (EphemerisProvider::moon_distance). This part changes no output bit, checked against a verbatim copy of the old iteration at 4,000 instants. ecliptic_position(Moon) now makes 1.003 full series evaluations where it made 2.756.

Interpolated walks for long scans

The Moon's and the Sun's apparent place is fitted with Chebyshev polynomials over short segments, the first time a time inside one is asked for, within 5e-5″ for the Moon and 1e-5″ for the Sun. Long scans then make a tenth to a twentieth of the ephemeris calls they did. Planets are not walked, because deflection switches off where a ray crosses the solar disc and no polynomial follows that step.

ChebyshevProvider, opt-in

The analytical series fitted per segment, within 0.1 m for the Moon and 3 m for a planet over the whole range. AnalyticalProvider is unchanged and remains the exact series.

Fused multiply-add, measured and not adopted

It gave no gain outside the noise, and it would have made output depend on the target. Without it the lunar series uses only IEEE multiply and add and pure-Rust libm. The series also no longer calls the SIMD trig kernel whose rounding differs between architectures, so its two bit digests are now one value on aarch64 and x86_64.

Delta T is read from the measurements and the published predictions

Before 1962 it is the cubic spline of Stephenson, Morrison and Hohenkerk (2016, Table S15). From 1962 to 2026-10 it is the USNO rows, then USNO's own predictions to 2033.75, and after that the Espenak and Meeus (2006) expressions shifted to meet the last prediction, so there is no step. Delta T is also continuous through each month now; it was constant within a month and jumped at its end. Before 1900 the change is largest: at 1700, 1650 and 1620 Delta T is 14.6, 43.7 and 67.1 s, where it was 8.8, 50.2 and 95.4 s. After 2033.75 Delta T is a prediction, and the new one differs from the old by up to about a minute by 2099, which moves a 2099 Moon by up to 33″. Sources, retrieval dates and checksums are in DATA_PROVENANCE.md.

Nutation matches the series it names

30 of the 77 IAU 2000B luni-solar rows held coefficients that are not in the series, and 17 had the sign of the main longitude term reversed. They moved Delta psi by up to 6.4 mas. The table now matches the IAU 2000A series row for row, and what remains is the model's own truncation, up to 2.6 mas.

Light deflection

The apparent place of every planet now includes the Sun's gravitational deflection: 4 mas at 90° of elongation, 0.47″ at 1°, and zero for a ray that passes inside the solar disc. The Sun and the Moon are not deflected.

The VSOP87A frame

Planet and Sun vectors are rotated into the equatorial frame by the matrix the VSOP87 authors publish for VSOP87A, in place of a rotation by the IAU 1976 obliquity alone. The analytical Sun, Mercury, Venus and Mars go from about 0.18″ to about 0.09″ mean.

Crossings are refined to the resolution of a Julian Day

muhurta::refine_crossing times tithi, nakshatra, karana and yoga ends, sign ingresses, lunar months, sankrantis and the syzygies that eclipses are searched from. It ran Newton's method until a step was under 1e-6°; it now runs until the time stops changing. Answers move toward the true root by up to 7.7 ms for the Moon, 77 ms for a Sun ingress, 85 ms for Mars, 128 ms for Mercury and 16 ms along a chain of eclipses.

The complete shadbala of BPHS 27

shadbala::compute_shadbala_complete computes it part by part, with options where the text is silent: the scope of moolatrikona, equal or unequal horas, ayana bala by the verse's table or by declination, the yuddha winner, and the recension of the special aspects. The parts are also functions, among them dig_bala_at, drik_bala_sputa, saptavargaja_bala, tribhaga_bala, paksha_bala, nathonnata_bala and cheshta_bala_at. shadbala_chart::shadbala_for_chart computes it for a computed chart and a place, with the birth moment taken from the engine's own sunrise and sunset. The MCP and WASM surfaces serve this form.

Bhava bala, vimshopaka, sputa drishti, four upagrahas

Bhava bala (bhava_bala::compute_bhava_bala, BPHS 27.26 to 31), vimshopaka (vimshopaka, BPHS 7.17 to 27, all four schemes), sputa drishti (drishti::sputa_drishti) and the upagrahas Kala, Mrityu, Ardhaprahara and Yamaghantaka (BPHS 3.66 to 70).

Eight dasha systems on MCP and WASM

compute_dasha serves Vimshottari, Ashtottari, Yogini, Chara, Narayana, and now Sthira, Niryana Shoola and Maheshwara, on a shared Jaimini strength (rashi_strength). The new three come with the inputs lagna_sign, graha_longitudes and karaka_scheme. Yogini goes beyond its first round (rounds and until), and Chara gains antardashas. A chart with no Brahma graha returns "no_brahma": true with a message rather than an error; the text gives no fallback for such a chart. An option that does not apply to the system asked for is refused.

Eclipse, ingress, station and heliacal searches

For eclipses there are khagola::grahana_with, ShadowEnlargement and SolarHorizon, and the penumbral_magnitude and horizon fields on Grahana. graha_ingresses returns every sign ingress in a range, and stations every retrograde station. next_udaya, next_asta and kalamsha_deg give heliacal rising and setting by the Surya Siddhanta's kalamsha (IX.2 to 9, X.1) or by a critical altitude. kala_mana gains lunar_month_span and year_turnover.

Chara Dasha, the two-lord signs

For Scorpio and Aquarius with both lords outside the sign, the stronger lord is now judged in the order of the verses (BPHS 46.158 to 166): companions first, then the nature of the rashi, then exaltation, then the larger count last. The count used to be compared first, and since two signs always give different counts the other tests never ran. Over every placement of the grahas, Scorpio's length changes in 14.16% and Aquarius's in 11.42%. Narayana, an alias, changes identically. The other reading of where exaltation falls is ExaltationRank::Decisive. DATA_PROVENANCE.md, Fix 14, has the record.

Lunar eclipses

Earth's shadow is now enlarged by Danjon's rule, as NASA's catalogue does; the old radii enlarged the umbra by 2% and the penumbra not at all. Against all 228 lunar eclipses of NASA's 2001 to 2100 catalogue, the module finds every one and no other, with the same type, both magnitudes within 0.0009 and greatest eclipse within 2.2 s. Umbral magnitudes fall by about 0.008 and penumbral contacts widen by 64 to 123 s. Three shallow penumbral eclipses missed before are found (2013-05-25, 2027-07-18, 2096-06-06), and 2042-09-29 is penumbral, not partial. ShadowEnlargement::Chauvenet keeps the old enlargement.

Solar eclipses

An eclipse with the Sun below the horizon throughout is no longer reported, and contacts outside daylight are clipped to sunrise and sunset.

Purnimanta months

A full-moon-to-full-moon month now takes the name, adhika and kshaya of the amanta month whose shukla paksha closes it (Calendar Reform Committee report, chapter V, section 5.7). It was named by the sankranti it contains, a month early on some shukla days. Amanta answers are unchanged.

Shadbala on MCP and WASM

It is the complete form now (see section 5). Its dig bala is continuous in the arc from the cusp (BPHS 27.7) and its drik bala is a quarter of the sputa drishti, with Mercury's and Jupiter's aspects whole (27.19). shadbala::compute_shadbala_full and its answers are unchanged.

Yogini on MCP

Each period carries "round": 1. Values are unchanged.

Search tools read the Moon and the Sun from a walk

MCP search_muhurta and search_transits read the Moon and the Sun from an interpolated walk. Measured against the direct path on 1,220 muhurta windows and 928 transit events, every field but the times is identical, and the times move by at most 0.08 ms and 0.64 ms.

Moon speed

A caller that takes the Moon's speed from moon_apparent_position instead of apparent_position sees it shift by up to about 1e-2 °/day, toward the true rate.

MCP compute_shadbala

It takes a moment and a place: julian_day, latitude, longitude and ayanamsha (all required), and optional house_system, elevation_m and tz_offset_minutes. It casts the chart itself and returns every part of BPHS 27 for the seven grahas, the birth moment it used, the four angular cusps, polar_fallback (true where the house system is undefined at the latitude and the chart's fallback cusps are the ones Dig Bala reads) and each graha's motion state. The earlier planets array and day and paksha flags are refused by name, and PlanetEntry and parse_planet are removed from vedaksha-mcp. A Tropical ayanamsha is refused, since shadbala reads the sidereal signs. Where the Sun does not both rise and set (polar day and night) the tool returns an error naming that, because there is no Hindu day to divide.

WASM compute_shadbala(input_json)

It takes the MCP tool's arguments key for key, the same Julian Day range, and returns the same result. It refuses what the tool refuses, in the same order and words.

WASM compute_dasha(input_json)

It takes the MCP compute_dasha arguments as one JSON object and serves all eight systems; it took three positional numbers and served Vimshottari only. {"moon_longitude": m, "birth_jd": j, "levels": l} reproduces the old call. A request with levels outside 1 to 5, birth_jd outside 1800 to 2400 CE or moon_longitude outside [0, 360) is refused, with the MCP tool's messages.

MCP and WASM compute_dasha, Yogini

Both refuse a request for more than 50,000 periods (rounds, or until far ahead, at deep levels), counting the rounds an until reaches as the engine does. The longitude inputs declare exclusiveMaximum: 360, as the handler always refused 360.

Rust

Code that builds a khagola::Grahana with a struct literal must add its two new fields.