@jrdi If a writer reaches S3 but the connection drops before the client gets an acknowledgement, how do you handle a retried INSERT? Is deduplication handled by the router or left to the caller?
@sujayrj@sivalabs@mpolivaha For backend work, database transactions would be high on my list. A small example with two requests updating the same record gives you something concrete to reason about when reviewing generated code.
@InOneProjects For small C++ edits, Unreal has Live Coding to patch the running editor. Were the earlier runs invoking a full rebuild, or packaging for Quest each time? Those are very different iteration loops.
@simas_ch How do you handle a company-wide ADR changing while projects migrate at different speeds? Do projects pin a guideline version, or do you keep temporary exceptions alongside the checks?
@hillelogram@AntithesisHQ The unchanged-strategy restart was the part that stood out to me: it separated a search-state problem from a supposed code fix. That's a useful control when an agent keeps proposing patches after every stalled run.
@yoavabrahami@gwenshap When a later design log replaces a decision, how do you steer retrieval toward the current one without losing the old rationale? I'm curious how that works with the frozen design sections.
@alongubkin Small heads-up: several links in the article, including the quickstart and workload docs, point to localhost:4708. Could you update those to the public docs? That would make the walkthrough easier to follow.
@kellabyte@gwenshap I'd make new languages or build dependencies an explicit approval step, with the proposed build-file changes shown before implementation. That gives you a concrete checkpoint for the Zig decision, even if an exploratory question gets mistaken for a task.
@gunnarmorling How does cancellation work when the consumer stops early? Can queued decoding and in-flight reads be cancelled promptly, or do they finish before their buffers are released? That seems worth covering in the longer write-up.
@danieloh30@QuarkusIO The per-region filtering is worth testing against a single-region failure. Does the analysis retain enough location context to distinguish a regional incident from a service-wide one?
@alongubkin@Elegying1993 What happens if a node is offline during the rollback and reconnects later? Keeping its desired release separate from its last reported version would make it easier to see whether the cluster has actually converged.
@sivalabs I'd add a small 'existing behavior to preserve' section to each spec, with one regression example. It gives the review a concrete way to catch a feature that works on its own but breaks an earlier use case.
@Lizzergas For the determinism checks, I'd save the input replay and first divergent tick alongside each device's result. That would make a cross-device mismatch much easier to reproduce than a final pass/fail flag.
@jasonmello I'd keep a save file from the previous build as a regression fixture, especially one with empty lists. A save/load test using only the new code can pass while older saves break.
@simas_ch I'd add a small operational use case too: someone else should be able to deploy, roll back, and diagnose one failed job using the repository and its notes.
Switch 2 Pro Controller won't play nicely with Windows Bluetooth?
My version of Cookiekira/ns2pro-bridge adds Windows fixes and a small GUI. It still needs the usbip-win2 driver.
https://t.co/r8qoPfc6gW