ICANN gTLD Application Window Recap

NawChart has now successfully sent a synthetic professional claim through its own frozen Snapshot → PayloadArtifact → SubmissionAttempt → Test trabsport…accepted it through its edit

The whole idea of Frensense was due to the introduction of AI agent to developers workflow. AI(Artificial intelligence) is probabilistic so we need something deterministic for code review. Semgrep is good but rules are not easy to write. Solution: Created an engine that doesn't need rules, we bake the code has rules in it.

64 DeepSeek agents solved 70.7% of SWE-bench Pro in under a day. We ran 46,784 attempts across all 731 problems on a single 8× B300 node. On the Doubleword stack, every agent completed the full benchmark in 20h 23m, delivering ~30× the throughput of throughput-oriented SGLang. That extra throughput mattered: a single agent solved 51.2% on average, while accepting a solution from any of the 64 pushed the score to 70.7%. Rushil breaks down the results and what’s driving that ~30× throughput gain in the latest Doubleword Inference Lab blog. Link in the comments 👇

Reproducibility isn't a given. It's something you build, one run at a time. A run involves dozens of small decisions, like when a reagent was added, how long an incubation ran, which equipment was used, and what an operator adjusted on the fly. Every one of those details is part of the story behind the result. CellPort runs alongside the work itself, capturing conditions, timing, materials, equipment, and operator actions as they happen and connecting each one directly to the result it produced, so the full story builds with every run. That's what CellPort builds: a process you can explain, run after run, whether it's the same scientist repeating it or a new team picking it up for the first time. See how it works → https://lnkd.in/gQ-XsPer

Harness is the next battle ground and here is a simple test! I asked Claude to select 3 test problems of its choice and execute them. Ran the exact queries on CellCog and asked Claude to be the Judge. PS: the PDF generated by Claude was just too simple, had to ask CellCog to regenerate it! Which kind of became the 4th test where CellCog won!

Opensourcing labgrid-tui, a terminal dashboard for labgrid. In one screen you see live what exists in the lab, what each target can do, and which command to run to interact with it. Custom command packs also let you keep your team's own recipes right in the TUI. It's read-only and copy-only, so it's safe to explore and to keep open as your everyday lab view and command generator. Try it in under 1 min without an actual lab: uvx labgrid-tui tour Repo in the first comment.

New Document News! Reducto's r-1 got a new document parser! I tested it in a 77-page lab. It led Docling and Surya in a pi harness on extraction-only questions, and it preserved a draft watermark and legal redline that a lot of other parsers missed. It didn't solve all issues (no surprise there), for example r-1 and Chandra both produced incorrect flowchart connections on one example, and detailed hard to read maps and equations can still be tricky. This is a tiny task-specific example look, not a general leaderboard, but still cool to see! Source: https://lnkd.in/eeb9--W4 https://lnkd.in/e2B8TsWq

Some R&D happens in a lab. Some of it happens in a feedlot. Bovision is building fixed, non-invasive monitoring and alarm infrastructure for cattle feedlots. Getting that right in real conditions takes constant testing and tuning, which means the evidence trail grows as fast as the build does. On a small engineering team, every hour spent catching the documentation up to the work is an hour taken off the work itself... Enter RAD, it captures the evidence while the work happens, keeps the trail of what was tried and what it proved up to date without anyone chasing it, and writes the record ready for review when claim time rolls around. Be like George. Get R&D off your plate. Get RAD. Pain free R&D That's RAD!

A twosome and a foursome can book the same Saturday 8am. The course has the same cost either way. At $89 per player, a foursome yields $356. At most courses, a twosome in the same slot yields $196. That's $160 short on a booking that felt almost full. When the peak window fills with twosomes, that gap compounds. A managed tee sheet doesn't just track it. It prices against it, with a per-player offset that adjusts based on group size and demand timing. That's what Part 3 is about. Part 3 of the TeeMetrics series is live. https://lnkd.in/g_Pf3YWV