01
Independent layers
Each output category is validated against the best reference for that domain, not a single catch-all source.
Validation Methodology
Every DevDarsha release must pass a multi-layer accuracy gate before it reaches production. This page explains the principles behind how we validate Panchang output, and why we believe this approach produces trustworthy results.
Prefer a human-readable companion? Read How DevDarsha Calculates and Verifies Panchang Data for the public guide version of this methodology.
Panchang: production/live
Computed from astronomical positions, calendar rules, and release-gated validation layers before publication.
Horoscope / Kundli: live
Birth data, chart, dasha, dosha, and matching outputs are live and remain deterministic.
Numerology: live
Name/date systems are explicit and reproducible; all eight endpoints are live, with access to advanced endpoints gated by plan.
Validation Philosophy
A single reference source is never enough. Panchang output spans multiple domains: astronomical calculations, traditional calendar logic, regional observance rules, and muhurat formulas. No single source is authoritative across all of them.
Our approach assigns each domain to the reference method that is most reliable for that specific type of data. A failure in one layer cannot hide behind success in another. Each layer is scored independently, and every release must pass all active layers before shipping.
02
Agreement over single oracles
For calendar-based outputs, we require agreement across multiple independent references before accepting a date as ground truth.
03
No hidden failures
Every test case either passes or fails. We do not round, exclude, or suppress results to make a release look better.
04
Strict tolerances
Tolerance thresholds are set to catch real algorithmic errors while accounting for documented sources of natural variation.
The Five Verification Layers
Panchang output is tested in five distinct groups. Passing the astronomical layer does not substitute for passing the festival layer. Each group has its own reference, its own tolerance, and its own pass gate.
Astronomical
Sunrise, sunset, and moon rise/set times validated against authoritative ephemeris-grade references. Strict sub-minute tolerance. Longitude fields verified against planetary computation standards.
Panchang Elements
Tithi, Nakshatra, Yoga, and Karana derived from raw astronomical positions and cross-checked against curated Panchang references, with special attention to sandhya and gandanta boundary cases.
Vrat & Lunar Events
Amavasya, Purnima, Ekadashi, Sankashti, and Parana dates validated via multi-source calendar agreement so no single source can define ground truth alone.
Festival Dates
Key festival dates tested against an agreement set of multiple independent curated calendars. Cases where references disagree are tracked separately, not silently excluded.
Muhurat Timings
Rahu Kaal, Gulika Kaal, Yamaganda, Abhijit, and Choghadiya are formula-derived outputs. Validation uses internal formula consistency tests and boundary checks.
How Tolerances Are Set
Every verification layer has a defined tolerance: the maximum acceptable deviation between API output and reference. Tolerances are chosen to match the natural precision boundaries of each domain, not to maximize pass rates.
Astronomical timings have strict sub-minute tolerances because atmospheric refraction and observer elevation are well-characterized and bounded.
Calendar element boundaries such as tithi and nakshatra have wider tolerances measured in ghatikas, reflecting the inherent fuzziness of boundary transitions when lunar speed varies.
Formula-derived outputs such as muhurat timings have narrow tolerances because there is no observer uncertainty; deviations are purely algorithmic.
Why We Use an Agreement Set for Festivals
Festival dates in the Indian calendar tradition vary by region, sect, and interpretation. Testing the engine against a single calendar would mean inheriting that calendar's specific variant as ground truth, which is not appropriate for a pan-India API.
The agreement-set model: for each festival and year, multiple independently curated Panchang sources are consulted. A date enters the test denominator only when all sources agree.
This makes engine accuracy a defensible measurement: it compares our output against the consensus of independent experts, not a single authority.
The Release Gate
No version of the engine ships without passing the full regression suite. This includes:
- All previously passing test cases so a new change cannot silently break an existing pass.
- All active verification layers while layers in progress are tracked separately but do not block shipping.
- Any new test cases added in that cycle so new coverage enters the same release gate.
Accuracy coverage expands with every release. The test suite grows continuously, and historical results are preserved so regressions are immediately visible.
Enterprise accuracy documentation
Detailed validation reports, per-layer test data, and custom accuracy certification for specific festivals or regions are available for enterprise customers.
Contact Sales