I like that my coding agents refer to me as the founder for each project even though none of the projects is a an actual startup. #codingAgent #genAI #buildInPublic
For networks too spread out to walk, we built per-segment risk tiers into GroundPulse: ground motion from radar, its uncertainty, pipe strain vs allowable, and what's nearby. Missing input? It says which. Reading only the corridor cut two years of radar from 160 GB to 13 GB. #InSAR #Sentinel1
Drunken Man Makes Interesting Point About Society https://theonion.com/drunken-man-makes-interesting-point-about-society-1819563812/
Two days on our satellite engine, all of it on the error bars. We measure ground movement in millimeters per year. Every number needs a second one saying how much to trust it. That's what we worked on. Nothing's calibrated until the test comes back. #InSAR #SAR #Geodesy #RemoteSensing #Rust
Pros And Cons Of Wearing Shorts In The Office https://theonion.com/pros-and-cons-of-wearing-shorts-in-the-office/
Legacy statistical systems. The business rules lived in the old code and nowhere else, and no subject-matter experts were available. We recovered the rules from source and proved parity by running the same data through old and new. Platform retired. The client runs it now without us.
I've been using scheduled agents to find performance work on GroundPulse. It turns satellite radar into millimeter measurements, so I wrote down what wasn't allowed to change before asking for suggestions. Fifteen made it through. 4.67% less time, and the published files came out identical.
I run scheduled loops against my R package. Each invokes a named skill: idea-scout files candidates, backlog-pipeline opens one PR, cran-check maps every NOTE to a fix. One run filed 4 issues; all 4 shipped in 1.4.0. None of them merges or releases. That stays with me.
Half a pixel sounds like nothing. At 30 m posting it's 15 meters — the difference between "the shoulder is settling" and "the field next to it is." The expensive failure in InSAR isn't a wrong millimeter. It's a right reading attached to the wrong asset.
myIO - that’s me! Big update -take a look for Shiny apps, MCP, Quarto, and WebR #tstats #shiny #mcp #quarto #webR #oss #cran #release
CRAN Updates@cranupdates.bsky.social · 3mo agoUpdates on CRAN: archive (1.1.14), bayesTFR (7.4-5), CovCorTest (1.2.0), crew (1.3.2), deepgp (1.2.3), ggsci (5.2.0), ip2locationio (1.1.1), ip2whois (1.0.1), margEVT (0.1.1), munsellinterpol (3.5-1), myIO (1.3.0), nametagger (0.1.8)
The last mile of analysis is usually PowerPoint. You find something real, then spend an afternoon exporting stills into slides so someone else can follow it. myIO 1.3.0, on CRAN today, adds named keyframes with play/step controls: the walkthrough is the chart, not a rebuild. #rstats #dataviz
Satellite radar measures ground movement across an entire pipeline corridor to the millimeter. The hard part isn't the measurement — it's knowing which ones are worth a truck roll. dolphinRust now separates measurement precision from solution quality, and says not_evaluable when it can't tell.
GRAND JUNCTION, CO—Pausing as he lifted a forkful of lettuce and baby spinach toward his mouth, local man Logan Barger reportedly took a moment Monday to appreciate what a good boy he was being as he ate his salad.
Satellite radar hands you one "coherence" number to decide whether a measurement is trustworthy. It's doing two jobs: how good the data was, and how well the model fit it. A strong fit over weak data clears the threshold anyway. dolphinRust now emits both, separately.
Tourist Disappointed To Leave Scotland Without Seeing Single Fat Bastard https://theonion.com/tourist-disappointed-to-leave-scotland-without-seeing-single-fat-bastard/
A crop forecast that's a single number is a guess wearing a lab coat. Ours gives a range — the band we're 90% sure the real yield lands in. And we check it: on held-out years the truth falls inside it 90.6% of the time for corn. Plan against the low end. An honest range beats a confident guess.
U.S. Military To Test Troops’ Testosterone Levels https://theonion.com/u-s-military-to-test-troops-testosterone-levels/
Want to learn geocomputation with R? The materials from our tutorial at UseR! 2026 are now available: slides, hands-on exercises, datasets, and solutions, covering CRS, vector and raster data, and mapping with **tmap**. github.com/geocompx/use... #rstats #geocompx #rspatial #UseR2026
httr2 1.3.0 is out now! Includes a bunch of improvements to OAuth token caching and summarises some of the biggest changes in the last few patch releases — opensource.posit.co/blog/2026-07.... #rstats
httr2 1.3.0
httr2 1.3.0 changes how OAuth tokens are cached on disk — you'll need to re-authenticate once after upgrading — and rolls up faster streaming, OAuth server metadata discovery, OpenTelemetry tracing, a...
opensource.posit.co
Satellite monitoring at scale is two questions: "where should we look?" and "what exactly is happening there?" The second is expensive and always will be. New research (arXiv 2606.07374) hints the first can become nearly free — and cheap triage is what makes portfolio-wide monitoring a real product.
FIFA Admits It Has No Idea Why Soccer Players Walk Out On Field With Little Kids https://theonion.com/fifa-admits-it-has-no-idea-why-soccer-players-walk-out-on-field-with-little-kids/
Shipped a Python client for TerraYield's yield-prediction service. Analysts get pandas DataFrames with conformal 90% intervals and model provenance attached — because a point estimate without uncertainty and lineage isn't a number you can defend. The notebook is the real UI, not the dashboard.
Historians Theorize Declaration Of Independence Started As ‘Twilight’ Fan Fiction https://theonion.com/historians-theorize-declaration-of-independence-started-as-twilight-fan-fiction/
Built an InSAR pipeline, then found two silent bugs by reading its actual output instead of trusting green tests: one served 2021 data as current, the other quietly zeroed out whole datasets. Neither threw an error. Check the output, not just the tests.
Rebuilding scientific software in-house follows a repeatable shape: validate against the original pixel for pixel, A/B on real data, optimize last. We did this to NASA's OPERA dolphin in Rust. Our own unwrapper beat the outsourced one (~9% fewer errors), and phase-linking ran 2.38x faster.