When something here turns out to be wrong, it is corrected in the open. This page lists every meaningful correction: what changed, when, and on which page. Typo fixes and layout tweaks are not listed; changes to facts, translations, attributions, data and technical claims are.
Last updated
Correction log
Newest first. Each entry links to the page it corrected, and that page carries the same note in its Sources & Verification panel.
The homepage said both 1511 and 1527 interview questions, and 24 data models; the SQL playground said 985 questions in one place and 988 in another; the homepage counted 22 sacred texts. Every one of these is now derived from the data and stamped automatically: 1,527 questions, 991 of them SQL, 23 data-model scenarios and 23 sacred texts. A build check now fails if a hand-typed number drifts again.
About 9% of roles showed a location that does not exist (for example "San Francisco, NY"). This came from pairing the first city of a multi-location posting with the last region. A posting can name several places: the card now shows the first one, the rest are listed and searchable, and each city has one spelling.
The inherited English meanings were found to be misaligned with the Tamil for a substantial number of verses. Examples: verse 10's Tamil is "standing, sitting, lying, walking I think of You", but its meaning described a fortified city; verses 12, 24, 25, 54 and 100 were likewise wrong. The 17 interpreted verses now carry corrected renderings built from their word-by-word breakdowns. Every other verse is labelled "Working translation — under review" instead of being presented as settled.
The essay was rebuilt after publication, following a source-by-source review. Income is now census and SAIPE throughout, CPI uses the published July 2026 value, property tax is modelled at 1.1%, and the headline finding changed. The article carries its own dated corrections note with the full detail.
An inconsistency about Adaptive Query Execution was fixed in four places: "AQE cannot split what was never divided" contradicted the page's own skew-splitting material. The pages now say that coalescing merges partitions, that skew handling can split an eligible partition, and that neither rescues a uniformly under-partitioned stage. The drill count was also corrected to 39 (21 inline + 18 dedicated).
How sources are recorded
Each of the 23 sacred-text study apps has a Sources & Verification panel at the foot of the page. It says where the text was taken from, who prepared the page, how to cite a verse, and what has and has not been checked. Its status is one of four:
Status
What it means
Sources noted
The page's source file names where its text came from: a scripture location, a website, a commentary. It has not been independently re-checked against that source.
Source not recorded
The source file does not say where the text came from. We say so instead of guessing. A source should be added on the next review.
Under review
A known problem is being worked through. The panel says what it is.
Verified
The text itself has been compared against a named source on a stated date. No page carries this status yet, and a build check refuses it without that evidence.
The records live in data/provenance/ and follow a documented schema. Nothing in them is inferred. When a record says "not recorded", that is the honest state of the page, not an oversight in the panel.
Report an error
If you find a mistranslated verse, a wrong attribution, a broken citation, an outdated statistic or a technical claim that does not hold up, please send a note through the contact page and choose Article feedback. Include the page address and, if you can, a source. Corrections that change meaning are logged above, with a date.
