Games
5,700. That is where an all-pairs nested loop stops fitting in a 16.67 ms frame on an M3 in Node 23. A uniform grid holds to 158,400 and a spatial hash to 123,200 — but the quadtree everyone reaches for breaks at 21,200, and updating it incrementally instead of rebuilding it made it nine times slowe
Integration
Accepting an Idempotency-Key header is the easy part. The SELECT-then-INSERT check almost everyone writes lets 969 duplicate charges through 1,000 concurrent requests. Here is what actually holds, measured on Postgres 16.
Integration
You stream it — and on Node 23 you have no choice, because fs.readFile refuses any file of 2,147,483,648 bytes or more with ERR_FS_FILE_TOO_LARGE. Below that ceiling the buffered version works fine on your laptop and dies under load: five concurrent 1 GB downloads cost 5,166 MB of RAM buffered again
Integration
Node 23's fetch waits 301 seconds for a response and http.request waits forever — measured. Long before that, a slow vendor took a healthy database endpoint from 100% to 7.7% success, while the endpoint that touched nothing stayed at 100%.
AI & Data
A typical regex redactor fully removed 296 of 900 labelled identifiers on a synthetic corpus — recall 0.3289. It caught 100% of US phone numbers written (555) 019-2837 and 0% of the same number written +84 904 879 900. Names it cannot catch at all. And even at recall 1.0, 100% of the records were st
AI & Data
Because something in the chain stored the text in a charset that has no byte for ầ. Measured on MySQL 8.4.11, a latin1 column turns Trần Thị Hương into Tr?n Th? H??ng — 14 bytes where UTF-8 needs 20, and the missing bytes are gone for good. The à mojibake is the recoverable one; question marks are n
Games
On disk, 1.17x — almost nothing. In texture memory, 1.54x against sprites padded to a power of two, and 16x against a naive fixed-cell grid. Measured over 340 generated sprites with three packing algorithms, four sort orders and four padding widths. Shelf packing matched MaxRects at 1/1500th of the
Integration
Of eighteen DDL statements measured on a 10-million-row Postgres 16.15 table, three rewrote it and froze every read and write for about 8 seconds. The other fifteen finished in under a second — and one of the cheap ones still stopped reads for 18 seconds.
AI & Data
Keyword search and vector search each fail almost completely on one predictable class of query, and the aggregate score hides it. Measured on 60 labelled queries: BM25 put the right document first for 100% of exact-identifier searches and 13.3% of paraphrases; the embedding model managed 40.0% and 7
AI & Data
Because the tokenizer was trained mostly on English. The same support notice is 163 tokens in English and 237 in Vietnamese — 1.45x for identical meaning — so a Vietnamese user pays 45% more to say the same thing. Measured with o200k_base and cl100k_base across seven languages, plus the accent test
Integration
Measured on Postgres 16 with a 497 MB, 5,000,000-row CSV. COPY FROM STDIN ran at 1,265,000 rows a second - 120x a loop of single-row INSERTs - and six parallel connections reached 3,571,000. The multipliers everyone repeats are not equal: dropping indexes was worth 2.3x, UNLOGGED only 1.2x, and sync
Infrastructure
We benchmarked the same Node workload natively and in a container on an M3 Mac. CPU, memory and container-to-container networking were free. A bind mount cost 30x against a named volume. Almost none of that is Docker's fault — it is the Linux VM macOS runs it in, and on Linux most of this article do