India's morning begins with one sheet.

Quick chart + Moon sign (Rasi), Nakshatra and a short reading. Birth date, time of birth and birth city are all mandatory — Lagna changes with time and place, so no horoscope is generated without them.

Fields marked * are mandatory — for both Quick and Detailed. No horoscope is generated without date of birth, time of birth and birth city: Lagna (Ascendant) is calculated from the exact time and place of birth.

Positions are approximate sidereal (Lahiri) calculations made in your browser, in the same spirit as this site's panchang — for an official kundli for marriage matching or major decisions, have it cast by your family astrologer with exact ephemeris tables.

My charts — saved on this device

Positions are approximate sidereal (Lahiri) calculations made in your browser, in the same spirit as this site's panchang — for an official kundli for marriage matching or major decisions, have it cast by your family astrologer with exact ephemeris tables.

Baby Name Starting Syllables

R1 · Wave-1 defect regressions — 11/11 PASS. Ekadasi Parana computed (Delhi 21 Nov 2026: 12:11 PM – 4:58 AM, 22 Nov, the past-midnight date shown); kshaya Ekadasi found for Delhi (Fri 20 Nov) while Chennai keeps 21 Nov; Panchak windows render (13 of 60 days); matching shows the real partial count with no “{q}” fragment and a live 18.5/36 score; matching city suggestions work; My Charts save → reopen → delete cycle works; nazhigai appears only in Malayalam and disappears on switching away; no doubled “✓ ✓” badges; the phone month calendar fits 390px with zero page overflow; Punjabi, Odia and Assamese dates render in their own scripts; a markup payload typed into name fields renders as inert text in the horoscope, the matching result and My Charts.

R2 · Wave-2 defect regressions — 5/5 PASS. PDF generation provably waits for the Indic webfonts (measured 701 ms against a delayed font load) before painting; kundli PDFs regenerated in Tamil, Malayalam and Bengali are valid 2-page files whose rendered pages were visually inspected — glyphs shape correctly, the Lagna summary and the Sade Sati line are complete, and none of the 397 drawn text strings contains a doubled “::”. Listen falls back to the default voice with the correct language tag when a voice object is rejected, and the spoken text equals the sheet's own text in English, Tamil and Hindi. The printable year calendar shows festival names in Tamil, Hindi and Kannada (28 festival cells, 12 month tables).

R3 · Wave-3 defect regressions — 9/9 PASS (one defect found and fixed in this wave; the whole group was then re-run). “Today” follows the selected city, not the device: with the device in Toronto at an instant when the two cities are on different dates, the sheet, Today button state, month highlight, festival countdown and the Choghadiya now-row all show Chennai's 14 November, and an open sheet rolls over at Chennai's midnight with no reload. Pongal 2027 is Friday 15 January in the festival list and Thai 1 on the sheet; Diwali 2027 is 29 October — fixed in this wave: the “Tamil Deepavali is kept a day earlier, on 28 Oct, in Tamil Nadu” note existed in the data but was shadowed by the short-description dictionary, so it never displayed; it is now appended to the displayed description in all 12 languages and verified in the rendered cards. Holi 2027 and the other added festivals are present (26 entries for 2027). Marriage muhurtham accepts exactly the published-list dates (26 Nov; 3, 4 Dec; 15, 20 Jan; 11 Feb — precision 6/6), Krishna-paksha tithis are accepted (17, 25, 26 among them), and Chaturmas candidates are rejected by the named Chaturmas rule. City aliases resolve (Bangalore, Bombay, Calcutta, Madras, Trivandrum, Cochin, Benares, München), “Poona” resolves to Pune — not Poonamalle — and “Baroda” still blocks as ambiguous with an explanatory message. The About text discloses the Google Fonts loading in all 12 languages, and the site's factual claims re-verified: 12 languages, 18,582 cities, 244 countries, 8 muhurtham ceremonies, 36 guna.

R4 · Wave-4 defect regressions — 13/13 PASS. Matching errors render as text, never raw dictionary keys, and an unknown city blocks gracefully with no crash; hour 13 and minute 99 are blocked in both forms; a failed matching validation clears the previous result; a future birth date is blocked with a message in both forms (checked in English, Tamil and Hindi); the festival list survives tradition chips, search and language switches with countdowns and descriptions live; festival descriptions match their own festival — the English set matches its source 48/54 verbatim (6 are deliberate rewordings of the same festival, read individually) and a digit-alignment scan across all 12 languages found 0 misalignments, so no rotation remains; duplicate reminders are deduplicated; the mobile menu closes on outside tap and on Escape; anchor landings leave section headings clear of the sticky header on phone and desktop; the navigation distinguishes Rasi Palan (labelled “Horoscope”) from Kundli; a past Dasasandhi wedding date is ignored with an explanatory note; the year calendar's weekday header is localised; and My Charts reports honestly when storage rejects a save (“Could not save — this device's storage is full or blocked. Your existing saved charts are unchanged.”).

R5 · Golden accuracy anchors — 6/7 PASS. Delhi's sheet for 13 Nov 2026 reads Chaturthi · Mula · Dhriti · Vanija with end-times 8, 6, 4 and 8 minutes from published almanac values — inside the known Moon-tolerance band. Karthigai 1 = 17 Nov 2026 and Aippasi 27 falls on 13 Nov. Chennai has exactly 24 Ekadasi tithis in 2026 by an independent elongation count; the finder lists 25 sunrise-days only because the 25 May tithi is vriddhi, prevailing at the sunrises of both 26 and 27 May. The 2027 eclipse list matches the published eclipse list of five (6 Feb annular solar, 21 Feb total lunar, 18 Jul penumbral lunar, 2 Aug total solar, 17 Aug penumbral lunar). Sunrise and sunset reproduce an independent solar-position implementation to 0.00 minutes on five city/date pairs including Kathmandu (+5:45). Item 39 — FAIL against the recorded anchor, page verified correct: for the known-answer birth (Chennai, 15 Aug 1990, 10:30 AM) the page gives Lagna Thulam 4.4°, Moon Rishabam / Rohini pada 3, Sun Kadagam, Rahu Magaram and Jupiter Mahadasha running — all matching the anchor — but Saturn in Dhanusu (266.1°) where the anchor says Magaram, and a Moon-dasha balance of 3.28 years where the anchor says 4.62. Two independent ephemerides (a published lunar series; independent Keplerian elements, geocentric) agree with the page — Moon Δ0.03°, Saturn Δ0.13° — so the anchor's two values, not the engine, are wrong; the engine was deliberately left unchanged, and Saturn/Rahu and all other bodies sit within 0.13° of the independent check on that chart and in the correct signs across five births (1950–2024).

R6 · Full sweeps — 6/6 PASS. Rendered read-backs in all 12 languages with a junk scan (raw keys, “undefined”, “NaN”, leftover placeholders, English diacritics, nazhigai rule): zero hits. Dictionary parity holds for every keyed set (214 + 145 + 62 + 33 + 26 + 17 + 15 + 7 + 4 keys × 12 languages) and every name array (signs 12, nakshatras 27, tithis 16, traits 12, planets 10 × 11 non-English languages with the designed English fallback); the 8 Kerala festival detail entries ship in English/Tamil/Malayalam with an English fallback in the other nine. The 365-day Chennai sweep found zero disagreements between the sheet, month calendar, vratham finder, festival list and Listen text. A feature smoke pass exercised sheet and month navigation, festivals with detail cards, all 8 muhurtham types, vratham and eclipses, daily/weekly/monthly palan, Thirukkural, horoscope generation, three-system matching, Tarabalam/Chandrabalam, baby-name syllables, peyarchi, reminders, the three tools, share buttons, both PNG downloads, a deep-link round-trip (?lang=ta&date=2026-11-21&city=Delhi lands exactly), the offline snapshot and A+ text size. All 14 in-page anchors resolve. Hostile inputs pass: a 5,000-character city paste, a 200-character name, years 0001 and 9999 (all values finite), and a storage-disabled profile still renders and computes. The clean-profile full journey produced zero page errors and zero unhandled exceptions; the only console entries were the test sandbox's own proxy refusing the Google Fonts fetch, which loads normally on the open web.

Out of scope — not tested here, by environment, not by omission: real-device speech audio quality (the test browser ships no voices; the spoken text and language tags are verified), the hosting-only offline service worker (it activates only on the real domain), and non-Chromium browser engines. “100% working” is claimed only for the 51 items above as scored.

Engine × result matrix. Chromium (Blink) desktop — full smoke PASS, 0 errors (the baseline for all value comparisons). Firefox (Gecko) desktop — full smoke PASS, 0 errors: the sheet for Chennai, 13 Nov 2026 is value-identical to Chromium (Chaturthi · Mula · Dhriti · Vanija, sunrise 6:07 AM, sunset 5:40 PM, to full float precision); language switches (Tamil, Hindi, Malayalam) clean; the voice-input unsupported note shows correctly (Firefox has no speech recognition); horoscope and matching generate; the kundli PDF downloads as a valid PDF file; storage persists across reload. Safari's engine (WebKit, AppleWebKit 605) desktop — full smoke PASS, 0 errors, after its OS libraries were installed into the sandbox (the first launch attempts failed on 42 missing system libraries; they were installed and the engine then ran — documented, not skipped): sheet values again identical to Chromium; all three language switches clean; the voice unsupported note shows (this engine build has no speech recognition); it is the only test engine with real speech voices installed (4), and Listen spoke with them without error; horoscope, matching and a valid kundli PDF all pass; every date-string format the code uses parses correctly in JavaScriptCore — no Invalid Date anywhere; storage persists across reload.

Engine watchlist findings. The Indian-calendar normaliser holds on all three engines — Gecko and JavaScriptCore both render “Kartika 22, 1948 Shaka” with no diacritics in the English view. City-time “today” (regression item 17) matches the Intl Asia/Kolkata date on all three engines. The horoscope anchor (item 39, corrected) is bit-identical across all three engines — Saturn 266.09° in Dhanusu, Moon-dasha balance 3.28 years, Lagna Thulam 4.4°. One genuine engine difference was found: SpiderMonkey's legacy date parser accepts strings the other engines reject (it parsed a garbage string as a real, wrong date), so the date tools' guards were hardened to a strict YYYY-MM-DD shape check on every engine. Storage completely blocked (every read/write throwing) still renders and computes on Chromium and Firefox with zero page errors. Back after a deep-link landing returns to the linked sheet (Tamil, Delhi, 21 Nov) intact.

Defects found and fixed in this wave (5). (1) The date picker could poison the whole page: a non-ISO value — what a text-entry browser such as desktop Safari hands back — set the sheet's date to Invalid Date and threw “RangeError: Invalid time value”, breaking the render. Reproduced live; the picker now validates format and round-trip exactly like the deep-link path and restores the current date on bad input, and the Age and Days-Between tools got the same strict guard (they could previously print NaN, or on Firefox's lenient parser a silently wrong count). Re-tested clean on Chromium and Firefox. (2) Install guidance named the wrong platform: the Install button showed iOS “tap Share → Add to Home Screen” instructions on every platform and never used Android Chrome's real install prompt. It now fires the captured native prompt where one exists (verified: the prompt is called and the box hides on accept), shows the Share-sheet steps only on iOS/iPadOS (verified with an iPhone user agent), and shows new browser-menu wording — added to all 12 languages — everywhere else. (3) Landscape phones: the sticky header measured 47–60% of the 320–412px-tall landscape viewport across the device matrix; on short landscape screens the bar now scrolls away with the page (verified scrolled fully off-screen) while portrait keeps the sticky bar. (4) Older-Safari hardening: the header's translucent background now has a solid fallback before the color-mix value, the blur gains its -webkit- form, page and print-view padding respect notch safe-areas (no visual change where insets are zero), and Listen warms up the voice list at load (Safari and Firefox return it empty until it changes) and resumes a paused speech engine before speaking. (5) Safari-engine width metric at 320px: WebKit reported the page 33px wider than the screen — its scroll-width counted the swipeable nav row's contents even though the row itself clips and sideways scrolling was impossible (scroll position stayed locked at 0). An html-level overflow clip, alongside the existing body rule, makes all three engines agree the page is exactly viewport-width, sticky header unaffected — verified at 320px on Chromium, Firefox and WebKit.

Device matrix (emulated). Pixel 7, Galaxy S-class and a 360px budget Android (Chromium engine, Android user agents, touch): the sheet leads the first screen everywhere; sticky header 192px (23–26% of portrait height); every main control at least 44px tall — no sub-44px target found; the swipeable nav row scrolls; the menu opens and closes by touch; the month calendar has zero overflow; horoscope city suggestions are tappable and the AM/PM toggle works; A+ raises text from 17px to 19.5px; zero errors. iPhone SE (320px), iPhone 15 and iPad Mini (WebKit engine, iOS user agents, touch): the same checks pass — header 29–34% on the phones, 19% on the iPad; suggestions, AM/PM, menu and A+ all work by tap; zero errors. One emulation limit, stated plainly: in the WebKit emulator, taps on non-interactive empty space do not synthesise the document click that closes the mobile menu — taps on any real control outside the bar close it, a programmatic click closes it, Escape closes it, and the outside-tap close passes on Chromium touch devices; the dead-space tap is the one path reserved for a real-iPhone check after hosting.

Load timing. Unthrottled local load to first sheet render: ~1.2–1.6s on the Android profiles, ~1.5s Chromium desktop, ~2.3s Firefox desktop; WebKit measured 3.6–4.1s in this sandbox on cold font caches (a sandbox figure, not a Safari benchmark). On a throttled Slow-4G profile (1.6 Mbps, 150 ms RTT, 4× CPU slowdown) the first sheet render measured ~15.0s on Pixel-7, iPhone-15 and 360px viewports alike — the page-weight trade documented in wave 3 (the 12 dictionaries and the 18,582-city database the offline design depends on), reported here as a number, not passed over.

Regression guard after the engine fixes — re-run on Chromium and Firefox (and WebKit where the item applies): item 17 (city-time today) PASS on all three engines; item 39-corrected anchor PASS, bit-identical on all three; item 46 (12-language junk scan) zero hits in visible content (apparent “{city}” hits were the page's own script text, not rendered content); item 48 (365-day Chennai sweep) 365 days checked, 0 invariant failures and 0 sheet-vs-engine disagreements on Chromium and Firefox; Tamil kundli PDFs re-downloaded and verified as valid 2-page files on Chromium and Firefox, and the English kundli PDF on all three engines. No wave 1–5 fix regressed.

Still genuinely untestable here: real iOS/Android hardware behaviour (true Dynamic Type and system font scaling beyond the A+ toggle, actual Samsung Internet and UC Browser quirks, OEM speech voices and real-device voice audio), the dead-space menu tap on a physical iPhone noted above, Safari's true storage-quota error shape on physical hardware (the blocked-storage path is verified with throwing storage on two engines), and the hosting-only offline service worker. WebKit here is Playwright's Safari-engine build, not a physical iPhone — real-device checks still belong after hosting.