habr is an institution. it's like the "runet hn", minus wild west vc ecosystem, plus integrated blog posting like lj ogs intended to. probably helps a lot with original work like TFA getting traction. more power to that!

runet sites of that era are often born out of the hacker's characteristic contrarian attitude "because we can". attempts to monetize them in more recent years are bound to accomplish little more than fuck up the content quality and/or the "owner cashes out and opens cafe" thing.

nevertheless, to this day, when i think habrahabr, i think way higher bar for technical competence than hn. it's all in the attitude.

It went downhill when they allowed getting an invitation via single blog post, requiring just one person to like it enough to give an invitation. Which wasn't hard to write - just translate something popular from hackernews before anyone else does it.

Shortly after, it became hilariously easy to farm and manipulate karma balances across the entire site. With 50 accounts (mults or real people all the same) you could create a new account a day.

Monetization started when it was already in a death spiral.

It was also notoriously politics-free, until something happened.

Agree, if your updates have time then datetime it should be. It's just in my work everything is date only. E.g. employment starts on a date, not datetime, there's no data loss.

It can be useful in some cases but can also be a hindrance. First because identifiers are more useful when they allow to actualy identify the thing; and also because now they can change from instance to instance, from customer to customer etc...

In the bitemporal model, both system and valid times are half-open intervals, and both the preceding and following interval can either have a different value or no value. Using only start times means that while records can be updated in either time stream, they cannot be logically deleted (in transaction time) or invalidated (in valid time) once they exist. There are databases where this assumption is valid, but in general it is problematic.

This is a fairly common problem. Data is often transferred between information systems in denormalized form (tables with hundreds of columns - attributes). In the data warehouse, they are normalized (data duplication in tables is excluded by using references to reference tables) to make it easier to perform complex analytical queries to the data. Usually, they are normalized to 3NF and very rarely to 6NF, since there is still no convenient tool for 6NF (see my DSL: https://medium.com/@sergeyprokhorenko777/dsl-for-bitemporal-... ). And then the data is again denormalized in data marts to generate reports for external users. All these cycles of normalization - denormalization - normalization - denormalization are very expensive for IT departments. Therefore, I had an idea to transfer data between information systems directly in normalized form, so that nothing else would have to be normalized. The prototypes were the Anchor Modeling and (to a much lesser extent) Data Vault methodologies.