Same raw samples, different metadata: introducing DNG Compare
Compare DNG raw samples and metadata, then preserve exact recoverable file versions with Python, Rust and OCaml.
Interoperable Cloud Services at First Pair
How AI helps to build exactly the things you need that no existing SaaS supports. For many years I used Bitly with a free plan, and managed to tack on a custom domain, chief.sc, in there. A decade of shortcuts followed. I prefer free plans for durable resources to
Making Lightroom usable again after years of stalled imports
After years of stalled imports, a large Lightroom catalog handled an ordinary October card quickly. The reusable repairs preserve existing edits and separate indexed performance, previews, settings and Sync outcomes.
Two rivers, one road: Rust and OxCaml race to count meanders
We ran the verified meander-counting machine as fast as two languages could take it. OxCaml took the lead, our first explanation was wrong, and one experiment let Rust catch up and count 50 crossings.
How many ways can a river cross a road? Counting Arnold's meanders with proofs
Arnold asked in how many ways a river can cross a road n times. Nobody has a formula, so we counted the rivers with programs proved correct in Lean 4: every value up to 32 crossings, a paper on First Pair, and a film.
TPC-DOOM: how fast does your SQL database run Doom?
TPC-DOOM measures the speed of the fastest databases
Sail plays Doom: SQLDoom's own client at 32 tics a second on Spark SQL
Best Spark jobs. Best SQL Doom. On Rust.
Anthropology Math
EIgen Times, Anthropology, and their math form a new teaching method, First Pair.
Sail runs Doom on DataFusion with a single query
Sail is a Spark-compatible compute engine written in Rust on Apache DataFusion: PySpark and Spark SQL clients connect to it over Spark Connect, and DataFusion executes what they send. This week Sail played Doom. Freedoom's first level, 1,200 tics of it, ran as one SQL query:
Nutmeg 0.1.0: graph algorithms where the data already is
Grust's graph kernels inside Sail's Spark Connect server: stage a graph from a DataFrame, call an algorithm, get a DataFrame back.
Grust Langoustine: narrower graphs, cheaper accounting, and scores you can ask for in f32
Grust's latest backend-neutral property-graph release.
What one PageRank sweep costs: five implementations on one host
What one sweep of one PageRank kernel costs across five implementations on one host, with every distance carrying its margin.