/tech/ - Technical SEO

im staring at hundreds of thousands of lines of java and thinking about that converter tool. totally wish i'd moved to data classes earlier so we could scrap all this boilerplate. if you dont do it when the codebase is small its basically impossible once the monolith takes over.

article: https://dzone.com/articles/switching-to-kotlin

jetoctopus is trying to sell this idea that you can just link gsc, ga4, and logs all on one view so clients don't get lost in status codes. it's basically saying stop sending them deep crawl depth charts and instead give them a dashboard they can poke at whenever.
>Every report type below already exists inside JetOctopus
it sounds nice but i've seen these "on demand" dashboards turn into another ticket for the dev team when something breaks. plus, if you don't include that one-line takeaway like it says, you're just sending them a bunch of meaningless charts again. building this once is fine, but managing the commentary every month sounds like a dream more manual work. does anyone actually use a live dashboard for clients without them asking a thousand questions?

found this here: https://jetoctopus.com/how-to-build-a-technical-seo/

Been moving most of my heavy site audits over to Screaming Frog lately because Xenu just chokes on anything over a few thousand URLs. It's much better at handling the modern JS-heavy stuff without me having to manually configure every single crawler setting. If you're dealing with massive subdirectories, stick to the paid version so you don't hit that memory ceiling.
spoaderXenu is basically a relic at this point/spoiler

The idea of using a contract pipeline to handle JSON validation at the boundary sounds fine on paper, but I'm not convinced it solves the actual messiness. Seeing that they track an explicit state like 'observed' or 'unknown' instead of just forcing a value is interesting, but how do you prevent the schema from becoming a nightmare to maintain? It's easy to talk about preventing silent search defects with field-level rejection rates, but I wonder about the latency hit on long context windows. If you're running this for something like 'catalog. v3', I'd need to see how much the repair step actually drags down the throughput. Also, how long did they run this test for? It's hard to tell if these low first-pass output rates are a temporary fluke or just the cost of doing business with inconsistent prose. Hard to know if this scales without seeing some actual error distribution data over a larger sample.

article: https://dev.to/daltonreed1289/nodejs-chatbot-api-compare-long-context-good-quality-via-schema-drift-controls-2ed6

Managing those huge folders of V142. sql files is such a nightmare during merges. It's always some mess of conflicting scripts when two people branch off main at the same time. I saw this library SchemaSynchronizer that tries to do it declaratively like Terraform does for infra. You just keep a single schema-definition.json file and let it figure out the diff.

{  "tables": {    "work_items": {      "columns": [        {"name": "id", "definition": "UUID PRIMARY KEY"}      ]    }  }}

It looks clean on paper, but I'm worried about what happens during a complex refactor. If it just reconciles the state, it might try to drop and recreate things by mistake. Never trust an automated tool to handle heavy data migrations without checking the plan first.

more here: https://dzone.com/articles/declarative-database-schema-management

another thing to manage in my dev workflow. finally got fx inside the T3 desktop and web clients with that auto mode enabled.
>The built-in fx provider supports Supervised and Auto modes
it's probably going to be a nightmare once it starts hallucinating changes in my control plane settings. anyone actually using the mobile version for this?

found this here: https://vercel.com/changelog/fx-is-now-available-in-t3-code

it's wild that argon managed to drop package import time by 21% just by running for eleven hours without any supervision. i'm skeptical it could actually trace all those cross-module references without getting lost in the noise of a real production repo.

more here: https://www.infoq.com/news/2026/10/google-argon-optimizes-adk/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global

found another client site where they're wrapping every single breadcrumbitem inside its own nested list instead of just using one flat array. it's a total mess for the crawler and makes the lcp look spoilergarbage/spover.

jetlag looks pretty decent for catching those stupid toISOString() bugs where everything breaks at 9pm in la. it basically wraps your test command so you don't have to manually mess with locales or offsets. caught some nasty stuff like:
> Code that assumes offsets are whole hours, or rounds getTimezoneOffset() / 60, is 15 to 45 minutes off.
it even checks for weird stuff like leap years and daylight savings shifts in places like kathmandu. tbh i'm skeptical about how much it'll slow down my CI pipeline if i run ten different destinations every time. still better than finding out a billing module is broken because of a 45 minute offset during a production deployment. wonder if it handles decimal separators for different locales properly though.

https://hackernoon.com/how-to-test-your-code-across-time-zones-and-dates-with-jetlag?source=rss

Using byte range requests to pull parts of compressed VCF files directly from S3 sounds pretty efficient for avoiding massive transfers. I wonder if bcftools can handle the overhead once you scale up the number of concurrent Lambda functions hitting the same objects.

article: https://www.infoq.com/news/2026/10/csiro-serverless-genomic-variant/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global