The 24 degrees is what makes the thread bug expensive instead of obvious. Nakshatra segments are 13 degrees 20 minutes wide [6], so divide: 24 / 13.333 = 1.8 [24]. A chart computed on a cold thread lands 1.8 segments over, inside a real nakshatra with a real ruling planet, and that seeds a full 120-year dasha cycle [7][9]. Nothing raises [8]. Every number in the response agrees with every other number in the response.
That is why the fixture list carries more weight than it first appears. Look at which of those tests would actually have caught thread B. The sum-to-120-years invariant [12] would not: a cycle seeded from the wrong nakshatra still sums to 120 years. The Ashtakavarga check would [15], because it compares a computed total against a value published outside the codebase. That is the difference between a self-consistency test and an oracle, and only the second kind detects a constant offset applied to every longitude.
The 28/36 fixture is the neatest of the set [13]. Two identical charts share a nadi, which is a dosha, so a perfect compatibility score is a defect rather than a nice result. The developer offers it as an interview question for astrology-app devs [13], and it is a fair one.
The sunrise formatter broke the same way, two orders of magnitude smaller. 07:59:34 rendered as 07:00, because 59.57 seconds rounded to 60 without carrying into the hour [17]. The displayed value is off by 59 minutes and 34 seconds [25], and it still looks like a sunrise.
The 429 is a middleware ordering problem with an unusually wide blast radius. The site's API key had exceeded its monthly quota, and the quota middleware was registered outside the CORS middleware, so its short-circuited response never passed through the layer that attaches Access-Control-Allow-Origin [20]. To a browser, a response missing that header is indistinguishable from a CORS violation [21], which is why the console named CORS and the CORS config had not been touched [18][19]. The 429 body itself was informative, with reset dates, in both languages, and no client ever saw it [21]. The fix is to attach the headers on each early-return path in the outer layer [22].
Two of these lessons transfer without argument. Grepping C bindings for a mode setter and assuming thread-local until proven otherwise [11] costs nothing and applies to any sync endpoint sitting on a threadpool [3]; re-asserting the mode inside every function that touches the library [10] is the same kind of cheap. The third one has a precondition. The standing rule that a failing accuracy test means the code is wrong and never the test [16] holds here because the spec was frozen centuries before the software, and because the user's grandmother has a printed panchang on the wall and will check it [27]. Where the spec is a ticket from last quarter, that rule mostly freezes your bugs into fixtures.
