How we verified this engine.

Arcsecond-level astronomy with bounded cross-checks

Esha's chart positions are computed by a clean-room Skyfield + JPL DE440s engine in the Lahiri sidereal system, Mean Node stated explicitly. Three bounded checks establish the calculation scope:

• 25,000 deterministic charts vs. the pinned real Swiss Ephemeris compressed backend — the maximum observed planet delta was 2.0137 arcseconds and the maximum observed ascendant delta was 7.8155 arcseconds under the shared stated Lahiri and Mean-Node convention. DE440s and INPOP21a are separate integrations fitted to the same modern observational basis and share ICRF/relativistic conventions. Horizons is JPL-family, not independent of DE440s. This is corroborated against two independent planetary theories; no observationally-independent verification is claimed. Every oracle call required the returned Swiss backend flag; a fixed output fixture and a live Moshier negative control proved that the requested backend actually supplied the values. Report: swisseph_25k/STEP1_RESULTS.
• Raw ICRF integration checks, before ayanāṁśa — INPOP21a at 2,000 deterministic dates had a maximum ecliptic-longitude delta of 0.025635 arcseconds; Horizons/DE441 at 500 dates had 0.002938 arcseconds. Each date covered the Sun, Moon, and five classical planets over 1900–2050.
• 100 charts vs. a separate, independent reference implementation — ascendant, all planets, every divisional chart, all vimśottarī daśā periods, and aṣṭakavarga, byte-for-byte compared. Report: ENGINE_CLASSICAL_ACCURACY_AUDIT_2026-06-24.

The INPOP/Horizons figures bound integration error, not errors common to the observations or frame. They are raw ICRF checks before ayanāṁśa. The Swiss result includes the shared stated Lahiri realization and is a configured-pipeline consistency check, not independent validation of ayanāṁśa. It does not establish that realization as unique or observationally validated.

Classical arithmetic, module by module

A 7-domain-expert review (Foundation/Pañcāṅga, Divisional charts, Daśā timing, Strength/Bala, Yogas, Jaimini/KP, Predictive/Synthesis) read the actual engine code module by module — 94 modules total — and, wherever feasible, verified each computation against both the running engine and an external oracle. Specific results from that audit:

• All 16 divisional-chart (varga) mappings exact — zero mismatches over 20,000 random test longitudes each.
• The full aṣṭakavarga bindu tables exact (sarvāṣṭakavarga total = 337, verified place by place).
• All six ṣaḍbala (six-fold planetary strength) components fall within their classical [0, 60] range.
• Vimśottarī daśā verified: the full 120-year cycle, Moon-nakṣatra seed, proportional split, running balance.
• Aṣṭakūṭa (kundli-matching) point maxima confirmed exactly 36, exhaustively across all 27×27 nakṣatra pairs.
• Varṣaphala (solar-return) accuracy: the Sun returns to its natal longitude within a fraction of an arcsecond (0.03–0.6″ across the verified runs).

Clean-room, provably

No copyleft or proprietary astrology library runs at deploy time — the engine is built on Skyfield and the same public JPL ephemeris data professional astronomers use, not a wrapped third-party library. This is checked two ways every time the codebase changes: a static scan of the shipped source for forbidden imports, and a runtime contamination check. Both must pass before any change can ship — it is a standing gate, not a one-time claim.

Every chart is deterministic: the same birth details always produce the same chart, and every accuracy fix is locked behind a regression test, so a later change cannot silently reintroduce a bug that was already found and corrected.

What this does not claim

This page verifies calculation — where the planets were, and what the classical formulas say about that placement. Astrological interpretation itself carries no such validation claim here, and Esha never states or implies that a reading predicts real-world outcomes. See Methodology for how interpretation is sourced and where the honest limits are.

Ask for the specifics

The reports named above are internal engineering artifacts, not public documents — but the specific numbers on this page are drawn from them directly, not summarised from memory. If you're writing about astrology-app accuracy and want to verify a claim on this page, reach out via Helpand we'll point you to the exact evidence.