{

"scraped_date": "2026-02-24",

"source": "hacker_news",

"total_scraped": 136,

"nontech_count": 38,

"posts": [

{

"id": "47127396",

"title": "Binance fired employees who found $1.7B in crypto was sent to Iran",

"link": "https://www.nytimes.com/2026/02/23/technology/binance-employees-iran-firings.html",

"domain": "www.nytimes.com",

"author": "boplicity",

"score": 425,

"comment_count": 190,

"created_ts": 1771874610,

"is_internal": false,

"post_text": "",

"is_ask_hn": false,

"matched_keywords": [

"fired"

],

"comments": [

{

"top": "Isn't this like the #1 use case for crypto?\nEveryone wants an untrackable unblockable currency that is out of government control until the day it is used for things they don't like, then suddenly \"government please control this!\"",

"author": "paxys",

"replies": [

{

"text": "I thought the #1 use case for crypto was ransomware, followed by shitcoin rug-pulls, and the ability to commit theft without recourse.\nSending money to Iran is just a minor edge case.",

"author": "chihuahua",

"depth": 1

},

{

"text": "That's a rather narrow view of crypto's uses. What about subverting democracy by bribing the President?",

"author": "rwmj",

"depth": 2

},

{

"text": "Has the lack of crypto ever stopped this from happening? Look up cases of gold bars being found in senators houses, those are actually MUCH less tracable.\nShitcoins and Shitstocks(some SPACs) do allow of a legal way to \"give\" others money through the transfer of value in a way that is technically legal. This again is not crypto specific though.",

"author": "carshodev",

"depth": 3

},

{

"text": "So the best part about being bribed with crypto is if one flees to another country to escape the law, one still has the aforementioned bribes. That plus some measure of anonymity.",

"author": "greesil",

"depth": 4

},

{

"text": "Anonymity is the one thing cryptocurrency does not do well. It's much harder to sieze but much easier to trace.",

"author": "eucyclos",

"depth": 5

}

]

},

{

"top": "Contrary to a lot of comments here, the only way to use bitcoin (or any cryptocurrency) without tracking is to mine it yourself, and even then...\nWhere did you get it? Purchased/transferred? Where did they get it? What else did the person with that wallet do?\nIf the answer is \"mined\", even then, you have to actually do something with it, right? Buy something? Where is that something shipped? At worst you'll have to pay customs on it, and have it actually get through customs. At best, your address is in a database now.\nHave it shipped somewhere obscure? Video cameras are everywhere.\nHave it shipped to someone else's house and steal it off their porch? Again, cameras everywhere.\nNot have a physical item? Just a service? That's pretty much the closest you'll get to anonymous money transfer and full usage (along with whatever VPN you prefer).\nCool that was a fun mental exercise. Now everyone tell me why I'm wrong!",

"author": "lacoolj",

"replies": [

{

"text": "It's great for transactions where moving a large amount of money would be impossible due to KYC laws and smuggling the cash even more so. As long as you can find someone willing to sell you bitcoin and someone else willing to accept it, you're still in business.",

"author": "wildzzz",

"depth": 1

},

{

"text": "I mean, I can meet you in an ally, transfer some satoshis from my wallet to yours, you hand me a wad of cash/jewels/MtG/collector funkos and you might not even know my name.",

"author": "VirusNewbie",

"depth": 1

},

{

"text": "Hmm, doesn't this work equally well with a wad of $10 and $20 notes? I mean, yes, notes could be clandestinely marked. But aren't bitcoins also traceable after the first transaction?",

"author": "nine_k",

"depth": 2

},

{

"text": "yes but harder to move 10M in cash around from country to country.\nI'm assuming I'm purchasing/selling a \nlot\n of MtG/Funko here in this example.",

"author": "VirusNewbie",

"depth": 3

},

{

"text": "There’s a large industry for cleaning cash which then makes moving a clean 10M or even 10B in clean cash nearly trivial.\n10M might not be as noticeable but crypto being nominally in a country on its own isn’t that useful as you still want to be able to spend it at the end of the day.",

"author": "Retric",

"depth": 4

}

]

},

{

"top": "If one of two options can't be regulated or tracked, that is the option that will predominantly be used by actors who have outsized interest in being regulation or being tracked.",

"author": "jstummbillig",

"replies": []

},

{

"top": "Is Iran supposed supposed to be banned on Binance?",

"author": "BenGosub",

"replies": [

{

"text": "Every US company/citizen is not allowed to do trade with Iran due to the ITSR laws except under highly specific situations.\nIt gets more complex if a company is multinational though.\nA citizen can travel to Iran but even if they buy something there on holiday if they bring it back to the US they need to go through complex customs procedures to make sure its legally brought back in.",

"author": "carshodev",

"depth": 1

},

{

"text": "> Every US company/citizen [...] if they bring it back to the US\nIs that relevant here?\n> Binance Holdings Ltd., branded Binance, [...] was founded in 2017 by Changpeng Zhao. Binance was initially based in China, then moved to Japan, subsequently left Japan for Malta, and currently has no official company headquarters.\nThe founder seems to have been born in China and is Canadian.\nI still also don't understand if Iran is supposed to be banned on Binance or not.",

"author": "embedding-shape",

"depth": 2

},

{

"text": "\"flowed from two Binance accounts to Iranian entities with links to terrorist groups, a possible violation of global sanctions.\"",

"author": "giarc",

"depth": 3

},

{

"text": "It's a US-sanctioned country so allied nations play along with the sanctions and Binance is located within that US sphere of influence so Iran is supposed to be currently banned, yes.",

"author": "arjie",

"depth": 1

},

{

"text": "It's not like Western(-ish) nations have much of a choice here. As soon as your banks and financial system depend on the USD in any way, it comes with a mandatory dose of US imperialism and extraterritorial jurisdiction.",

"author": "hocuspocus",

"depth": 2

}

]

},

{

"top": "Remember that the CEO of Binance was pardoned by Trump after pleading guilty to financial fraud.",

"author": "LunaSea",

"replies": [

{

"text": "It's more than just that.\n> President Trump granted a pardon to Binance’s founder, Changpeng Zhao, who had spent four months in federal prison in 2024 for his role in the firm’s crimes. The Trump family’s crypto start-up, World Liberty Financial, has forged close business ties with Binance, and Mr. Zhao was a guest this month at a conference at Mar-a-Lago, Mr. Trump’s club in Palm Beach, Fla.",

"author": "paxys",

"depth": 1

},

{

"text": "it's more than just that:\nhttps://www.forbes.com/sites/zacheverson/2026/02/09/trump-st...\n> Binance holds about 87% of USD1, the stablecoin issued by a Trump family crypto venture—a greater concentration than any other major stablecoin has at a single exchange, roughly $4.7 billion of the $5.4 billion total supply.",

"author": "lavezzi",

"depth": 2

},

{

"text": "Iran obviously missed the memo. All they have to do is setup a wealth fund and invest heavily in a Trump venture; then they can become a most favored nation and forego all this conflict.",

"author": "michaelteter",

"depth": 1

},

{

"text": "They'd have to invest more than adelson, ~250M.",

"author": "throwaway85825",

"depth": 2

},

{

"text": "I wonder if the pardon bribe is less if your crime is something near and dear to the Orange King's heart.",

"author": "ourmandave",

"depth": 1

}

]

}

]

},

{

"id": "47123631",

"title": "Show HN: PgDog – Scale Postgres without changing the app",

"link": "https://github.com/pgdogdev/pgdog",

"domain": "github.com",

"author": "levkk",

"score": 194,

"comment_count": 44,

"created_ts": 1771860804,

"is_internal": false,

"post_text": "Hey HN! Lev and Justin here, authors of PgDog (<a href=\"https:&#x2F;&#x2F;pgdog.dev&#x2F;\">https:&#x2F;&#x2F;pgdog.dev&#x2F;</a>), a connection pooler, load balancer and database sharder for PostgreSQL. If you build apps with a lot of traffic, you know the first thing to break is the database. We are solving this with a network proxy that works without requiring application code changes or database migrations.<p>Our post from last year: <a href=\"https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44099187\">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44099187</a><p>The most important update: we are in production. Sharding is used a lot, with direct-to-shard queries (one shard per query) working pretty much all the time. Cross-shard (or multi-database) queries are still a work in progress, but we are making headway.<p>Aggregate functions like count(), min(), max(), avg(), stddev() and variance() are working, without refactoring the app. PgDog calculates the aggregate in-transit, while transparently rewriting queries to fetch any missing info. For example, multi-database average calculation requires a total count of rows to calculate the original sum. PgDog will add count() to the query, if it’s not there already, and remove it from the rows sent to the app.<p>Sorting and grouping works, including DISTINCT, if the columns(s) are referenced in the result. Over 10 data types are supported, like, timestamp(tz), all integers, varchar, etc.<p>Cross-shard writes, including schema changes (CREATE&#x2F;DROP&#x2F;ALTER), are now atomic and synchronized between all shards with two-phase commit. PgDog keeps track of the transaction state internally and will rollback the transaction if the first phase fails. You don’t need to monkeypatch your ORM to use this: PgDog will intercept the COMMIT statement and execute PREPARE TRANSACTION and COMMIT PREPARED instead.<p>Omnisharded tables, a.k.a replicated or mirrored (identical on all shards), support atomic reads and writes. That’s important because most databases can’t be completely sharded and will have some common data on all databases that has to be kept in-sync.<p>Multi-tuple inserts, e.g., INSERT INTO table_x VALUES ($1, $2), ($3, $4), are split by our query rewriter and distributed to their respective shards automatically. They are used by ORMs like Prisma, Sequelize, and others, so those now work without code changes too.<p>Sharding keys can be mutated. PgDog will intercept and rewrite the update statement into 3 queries, SELECT, INSERT, and DELETE, moving the row between shards. If you’re using Citus (for everyone else, Citus is a Postgres extension for sharding databases), this might be worth a look.<p>If you’re like us and prefer integers to UUIDs for your primary keys, we built a cross-shard unique sequence, directly inside PgDog. It uses the system clock (and a couple other inputs), can be called like a Postgres function, and will automatically inject values into queries, so ORMs like ActiveRecord will continue to work out of the box. It’s monotonically increasing, just like a real Postgres sequence, and can generate up to 4 million numbers per second with a range of 69.73 years, so no need to migrate to UUIDv7 just yet.<p><pre><code> INSERT INTO my_table (id, created_at) VALUES (pgdog.unique_id(), now());\n</code></pre>\nResharding is now built-in. We can move gigabytes of tables per second, by parallelizing logical replication streams across replicas. This is really cool! Last time we tried this at Instacart, it took over two weeks to move 10 TB between two machines. Now, we can do this in just a few hours, in big part thanks to the work of the core team that added support for logical replication slots to streaming replicas in Postgres 16.<p>Sharding hardly works without a good load balancer. PgDog can monitor replicas and move write traffic to a promoted primary during a failover. This works with managed Postgres, like RDS (incl. Aurora), Azure Pg, GCP Cloud SQL, etc., because it just polls each instance with “SELECT pg_is_in_recovery()”. Primary election is not supported yet, so if you’re self-hosting with Patroni, you should keep it around for now, but you don’t need to run HAProxy in front of the DBs anymore.<p>The load balancer is getting pretty smart and can handle edge cases like SELECT FOR UPDATE and CTEs with INSERT&#x2F;UPDATE statements, but if you still prefer to handle your read&#x2F;write separation in code, you can do that too with manual routing. This works by giving PgDog a hint at runtime: a connection parameter (-c pgdog.role=primary), SET statement, or a query comment. If you have multiple connection pools in your app, you can replace them with just one connection to PgDog instead. For multi-threaded Python&#x2F;Ruby&#x2F;Go apps, this helps by reducing memory usage, I&#x2F;O and context switching overhead.<p>Speaking of connection pooling, PgDog can automatically rollback unfinished transactions and drain and re-sync partially sent queries, all in an effort to preserve connections to the database. If you’ve seen Postgres go to 100% CPU because of a connection storm caused by an application crash, this might be for you. Draining connections works by receiving and discarding rows from abandoned queries and sending the Sync message via the Postgres wire protocol, which clears the query context and returns the connection to a normal state.<p>PgDog is open source and welcomes contributions and feedback in any form. As always, all features are configurable and can be turned off&#x2F;on, so should you choose to give it a try, you can do so at your own pace. Our docs (<a href=\"https:&#x2F;&#x2F;docs.pgdog.dev\">https:&#x2F;&#x2F;docs.pgdog.dev</a>) should help too.<p>Thanks for reading and happy hacking!",

"is_ask_hn": false,

"matched_keywords": [

"team",

"feedback"

],

"comments": [

{

"top": "Great progress, guys! It’s impressive to see all the enhancements - more types, more aggregate functions, cross-node DML, resharding, and reliability-focused connection pooling and more. Very cool! These were really hard problems and took multiple years to build at Citus. Kudos to the shipping velocity.",

"author": "saisrirampur",

"replies": []

},

{

"top": "Stupid question but does this shard the database as well or do we shard manually and then setup the configuration accordingly ?",

"author": "codegeek",

"replies": [

{

"text": "It shards it as well. We handle schema sync, moving table data (in parallel), setting up logical replication, and application traffic cutover. The zero-downtime resharding is currently WIP, working on the PR as we speak: \nhttps://github.com/pgdogdev/pgdog/pull/784\n.",

"author": "levkk",

"depth": 1

},

{

"text": "Incredible. I am really interested in trying pgdog for our B2B SAAS product. Will do some testing.",

"author": "codegeek",

"depth": 2

}

]

},

{

"top": "Some HTTP proxies can do retries -- if a connection to one backend fails, it is retried on a different backend. Can PgDog (or PgBouncer, or any other tool) do something similar -- if there's a \"database server shutting down\" error or a connection reset, retry it on another backend?",

"author": "cuu508",

"replies": [

{

"text": "Not currently, but we can add this. One thing we have to be careful of is to not retry requests that are executing inside transactions, but otherwise this would be a great feature.",

"author": "levkk",

"depth": 1

}

]

},

{

"top": "Happy pgdog user here, I can recommend it from a user perspective as a connection pooler to anyone checking this out (we're also running tests and positive about sharding, but haven't run it in prod yet, so I can't 100% vouch for it on that, but that's where we're headed.)\n@Lev, how is the 2pc coming along? I think it was pretty new when I last checked, and I haven't looked into it much since then. Is it feeling pretty solid now?",

"author": "mijoharas",

"replies": [

{

"text": "It feels better now, but we still need to add crash protection - in case PgDog itself crashes, we need to restore in-progress 2pc transaction records from a durable medium. We will add this very soon.",

"author": "levkk",

"depth": 1

}

]

},

{

"top": "Can you elaborate a bit more on the challenges faced in making Postgres shard-able?\nI remember that adding sharing to Postgres natively was an uphill battle. There were a few companies who has proprietary solutions for it. What you've been able to achieve is nothing less than a miracle.",

"author": "written-beyond",

"replies": [

{

"text": "So many, where to begin.\n1. People don't design schemas to be sharded, although many gravitate towards a common key, e.g. user_id or country_id or tenant_it or customer_id. Once that happens, sharding becomes easier.\n2. Postgres provides a lot of guarantees that are tricky to maintain when sharded: atomic changes, referential integrity, check constraints, unique indexes (and constraints), to name a few. Those have to be built separately by a sharding layer (like PgDog) and have trade-offs, usually around performance. It's a lot more expensive to check a globally enforced constraint than a local one (network hops aren't free).\n3. Online migrations from unsharded to sharded can be tricky: you have to redistribute terabytes of data while the DB continues to serve writes. You can't lose a single row - Postgres is used as a store of record and this can be a serious issue with business impact.\nWe're taking increasingly bigger bites at this apple. We started with basic query routing and are now doing query rewrites as well. We didn't handle data movements previously and now have almost fully automatic resharding. It takes time, elbow grease and most importantly, willing and courageous early adopters to whom we owe a huge debt of gratitude.",

"author": "levkk",

"depth": 1

}

]

}

]

},

{

"id": "47116026",

"title": "Show HN: WARN Firehose – Every US layoff notice in one searchable database",

"link": "https://warnfirehose.com",

"domain": "warnfirehose.com",

"author": "sendkamal",

"score": 132,

"comment_count": 26,

"created_ts": 1771803163,

"is_internal": false,

"post_text": "Hi HN,<p>I built WARN Firehose because I was frustrated trying to track layoff data across the US. The WARN Act requires companies with 100+ employees to file public notices 60 days before mass layoffs — but the data is scattered across 50 state websites with different formats, broken links, and no API.<p>WARN Firehose scrapes every state workforce agency daily and normalizes the data into a single database going back to 1988. It now has 131,000+ notices covering 14 million workers.<p>*What you can do:*<p>- Browse interactive charts and data tables (no account needed): <a href=\"https:&#x2F;&#x2F;warnfirehose.com&#x2F;data\" rel=\"nofollow\">https:&#x2F;&#x2F;warnfirehose.com&#x2F;data</a>\n- Drill into any state, city, company, or industry: <a href=\"https:&#x2F;&#x2F;warnfirehose.com&#x2F;data&#x2F;layoffs\" rel=\"nofollow\">https:&#x2F;&#x2F;warnfirehose.com&#x2F;data&#x2F;layoffs</a>\n- Query the REST API (free tier: 100 calls&#x2F;day): <a href=\"https:&#x2F;&#x2F;warnfirehose.com&#x2F;docs\" rel=\"nofollow\">https:&#x2F;&#x2F;warnfirehose.com&#x2F;docs</a>\n- Export in CSV, JSON, NDJSON, Parquet, or JSON-LD\n- Set up webhooks for real-time alerts on new filings<p>*Who uses this:*<p>- Journalists breaking layoff stories before press releases\n- Quant funds using WARN filings as an alternative data signal (filings happen ~60 days before layoffs)\n- Recruiters sourcing from displaced talent pools\n- Researchers studying labor market dynamics\n- Workforce development boards doing rapid response planning<p>*Tech stack:* Python&#x2F;FastAPI, SQLite, scrapers for all 50 states, static HTML generation for SEO pages, Claude Haiku for AI analysis, deployed on EC2.<p>Free tier is genuinely useful (100 API calls&#x2F;day, dashboard access, charts). Paid plans start at $19&#x2F;mo for full historical data and bulk exports.<p>Would love feedback on the API design, data quality, or anything else. Happy to answer questions.",

"is_ask_hn": false,

"matched_keywords": [

"layoff",

"feedback",

"layoffs"

],

"comments": [

{

"top": "Thank you for checking the work and all the comments which are very very valuable. Site does quite a lot to be honest so please take some time to explore more :)\n- It has over 15k individual landing pages for search engines - a dedicated page for each city, state, company, county. These pages are very reach on how they look haha:\n- Exmaple: new jersey page-> \nhttps://warnfirehose.com/data/layoffs/new-jersey\n\n- Data is exportable to multiple format including json-d and parquet - i never heard of parquet before.\n- The site has MCP (model context protocol) built in !\n- it accepts payments on multiple methods including paypal, apple pay, amazon pay, card etc. All refund and all built in.\n- It also has an invisible admin dashboard where I can see everything (all metrics, signups payments etc.)\n- Reports are crazy - these are genuinely better than someone writing at wallstreet journal, seriously. you would have to check to see them.\n- The api endpoints.\nI am convinced the role Software engineer is going to go away by this year end. We are turning into builders, not coders. All SaaS apps are going to take big hit because they can be builkt better for fraction of cost by anyone - just think about it.\n- Its all about ideas and anyone can do it.",

"author": "sendkamal",

"replies": []

},

{

"top": "The quant fund use case is the most interesting angle here. WARN filings have the rare property of being legally mandated with specific timing (60-day advance notice), which makes the signal horizon predictable in a way that most alternative data is not.\nThe big caveat: compliance is uneven. Companies under 100 employees are exempt, and there is a documented pattern of employers paying WARN Act penalties retroactively rather than filing -- especially in fast-moving situations where 60 days advance notice is operationally inconvenient. So the signal has systematic gaps at exactly the moments of highest market interest.\nHave you looked at coverage rates vs. announced layoffs (e.g., correlation with Challenger Gray reports or JOLTS)? That gap number is basically the signal noise floor for any quant strategy built on this data.",

"author": "vicchenai",

"replies": [

{

"text": "This is great insights @vicchenai. I just made it in a whim. I asked Gemini for suggestions and it presented few and I picked this one because there didn't seem to be good players out there. There is layoffdata and warntracker and all but they were built using hand coding :) it probably took them quite long to have it done. This is a challenging thing - 50 states having different data formats and having to poll them daily. Data cleaning is another thing. Historical data is also a thing. Plus all the blogs, reports, data presentation and api endpoints.\nIf it all can be done in a weekend without having to write a single line manually, sky is the limit no? Anybody having an idea can make these thing happen. I am not expecting it to make any money and it was a learning project but I do see some value for certain people.\nInsurance brokers would benefit if they are first to know so they coudl target these layed off people. Recruiters the same and definitely hedge funds, short shellers and quant. Gemini tells me data is the new oil ! I am convinced.",

"author": "sendkamal",

"depth": 1

},

{

"text": "> and there is a documented pattern of employers paying WARN Act penalties retroactively rather than filing -- especially in fast-moving situations where 60 days advance notice is operationally inconvenient.\nOh, I got a solution for that. Don't just go for WARN Act penalties. Go after offenders with the hammer called SEC and market manipulation regulations. That kind of stuff \nreally\n hurts.",

"author": "mschuster91",

"depth": 1

}

]

},

{

"top": "This dataset looks interesting but the site doesn’t instill a lot of confidence in data integrity.\nOn the Charts page the selected time range is 12/01/2025 to 02/28/2026 and shows 106,603 employees affected. But the horizontal bar chart with state level data shows numbers in millions. For example, CA has more than 2 million and IL has more than 1.7 million employees affected. Then the layoff map at the bottom shows only layoffs in Texas.",

"author": "malshe",

"replies": [

{

"text": "Thank you @malshe. I will review and get it addressed. The site is not perfect at all and I will address them.",

"author": "sendkamal",

"depth": 1

},

{

"text": "Agreed, this doesn't pass my smell test. November 2026 reports layoffs in NJ (with past tense) \nhttps://warnfirehose.com/blog/2026/11/week-1\nThe layoffs in the report are not listed in NJ's own warn notice\n\nhttps://www.nj.gov/labor/assets/PDFs/WARN/2026_WARN_Notice_A...",

"author": "pdh",

"depth": 1

},

{

"text": "Thank you and I will have them addressed. There are many things in terms of data quality and I am still working on it.",

"author": "sendkamal",

"depth": 2

},

{

"text": "You can think about LLM-generated UIs/apps the same way you think about LLM-generated responses. It's a bunch of garbage, but if you know what you're looking for, you might find something useful.\nThis doesn't seem to work at all for stats-related apps/sites though, since you can't judge the accuracy of what's being presented. If the site claims it'll \"take you to space,\" you don't take that literally, you just treat it as another AI artifact. But with numbers, you have no way to tell what's accurate and what's just made up.",

"author": "csomar",

"depth": 1

},

{

"text": "Thats such a great point. What if I am (somehow) able to warrant that the data presented here exactly matches the state notices? would that be helpful?",

"author": "sendkamal",

"depth": 2

}

]

},

{

"top": "Interesting, though a lot of the UI seems broken. For my state I see some notice dates in the future (it's not explained why, if this is when the filing will be executed or if it's an incorrect filing date, as the column is just \"Notice Date\")\nSome of the entries pull up a page that says \"Failed to load company data: No company name provided in URL\" from the state specific view (e.g, any link on \nhttps://warnfirehose.com/data/layoffs/california\n ). Has a vibe-coded feel to it.\nI saw a lot of \"Purchase dataset for city details\" in places which was annoying. Wondering how much processing is being done on the base dataset to justify the pricing. Could you explain a bit on the normalization/cleaning process?",

"author": "cobertos",

"replies": [

{

"text": "@cobertos The UI things should have been addressed. However I did to an extensive testing myself and everything seemed to have worked for me. I am sold that vibe coding is a thing, seriously. To your question about normalization cleaning - I have gone through multiple iterations - the data is definitely not clean and all. However, claude has made a function (instead of cleaning them on demand, it has a function which gets updated with every pattern so it could be reused).\nFor the pricing, to be honest I have zero hope to make money. Its just out there because I wanted to integrate this with stripe for payments and all. At the same time, if you look at competitor sites, they seem crap lol. plus look at their pricing. The site does have some operating cost and I will have to recover them if i can if I want it to be self sustainable but only if this is a value to someone. I am trying to make this a value. Please offer if you have ideas and I would appreciated it :)",

"author": "sendkamal",

"depth": 1

},

{

"text": "Definitely vibe coded. It follows the same generic Claude UI patterns for a data app / data oriented website. Not necessarily a bad thing per say if it's still curated and tweaked with human taste. And ofc validated to work :)",

"author": "skadamat",

"depth": 1

},

{

"text": "Hi @skadamat - I am not very experienced with claude code and don't know what it has done previously but new opus 4.6 is great. I think I need to work on claude.md. That really seems to be the soul of Claude. I've incurred quite a lot of money on API cost and claude plan and my time. But I feel like that can be cut to 1/4th easily with a solid claude.md plan.",

"author": "sendkamal",

"depth": 2

}

]

},

{

"top": "Great site thank you. Just curious, I looked up my company(more than 40k employees across the world including many US states) and it seems like I am not seeing the layoffs that colleagues have experienced. This is probably expected as im probably missing some criteria. Do all layoffs have to have a WARN notice or are there mechanisms/criteria that allow companies to lay people off without filing these noticies?",

"author": "nebula8804",

"replies": [

{

"text": "I am agreeing with two comments. There might be many reasons. Just the fact that there are only 12M records in last 30 years or so for all 50 states it definitely doesn't represent all data - may be 10%",

"author": "sendkamal",

"depth": 1

},

{

"text": "Many companies pay severance in lieu of giving notice.",

"author": "jordanb",

"depth": 1

},

{

"text": "If the layoff is small enough, it's not triggered.",

"author": "bagels",

"depth": 1

}

]

}

]

},

{

"id": "47123894",

"title": "Show HN: Sowbot – Open-hardware agricultural robot (ROS2, RTK GPS)",

"link": "https://sowbot.co.uk/",

"domain": "sowbot.co.uk",

"author": "Sabrees",

"score": 127,

"comment_count": 40,

"created_ts": 1771861713,

"is_internal": false,

"post_text": "Sowbot is an open-hardware agricultural robot designed to close the &quot;prototype gap&quot; that kills most agri-robotics startups and research projects — the 18+ months spent on drivers, networking, safety watchdogs, and UI before you can even start on the thing you actually care about.<p>The hardware is built around a stackable 10×10cm compute module with two ARM Cortex-A55 SBCs — one for ROS 2 navigation&#x2F;EKF localisation, one dedicated to vision&#x2F;YOLO inference — connected via a single ethernet cable.<p>Centimetre-level positioning via dual RTK GNSS, CAN bus for field comms, and real-time motor control via ESP32 running Lizard firmware.<p>Everything — schematics, PCB layouts, firmware — is under open licences.\nThe software stack runs on RoSys&#x2F;Field Friend (for teams who want fast iteration) or DevKit ROS (for teams already in the ROS ecosystem). The idea is that a lab in one country can reproduce another lab&#x27;s experiment by sharing a Docker image.<p>Current status: the Open Core brain is largely fabricated, the full-size Sowbot body has a detailed BOM but isn&#x27;t yet assembled, and we have two smaller dev platforms (Mini and Pico) in various stages of testing.<p>We&#x27;re a small volunteer team and we&#x27;re looking for contributors — hardware, ROS, firmware, docs, whatever you can offer.<p>The best place to start is our Discord: <a href=\"https:&#x2F;&#x2F;discord.gg&#x2F;SvztEBr4KZ\" rel=\"nofollow\">https:&#x2F;&#x2F;discord.gg&#x2F;SvztEBr4KZ</a> — we have a weekly call if you&#x27;d prefer to just show up and chat.<p>GitHub: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;Agroecology-Lab&#x2F;feldfreund_devkit_ros&#x2F;tree&#x2F;caatinga-dev\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;Agroecology-Lab&#x2F;feldfreund_devkit_ros&#x2F;tre...</a>",

"is_ask_hn": false,

"matched_keywords": [

"team",

"networking"

],

"comments": [

{

"top": "> The hardware is built around a stackable 10×10cm compute module with two ARM Cortex-A55 SBCs — one for ROS 2 navigation/EKF localisation, one dedicated to vision/YOLO inference — connected via a single ethernet cable.\nI will preface this by saying that I have nothing against ARM per se, that my employer/team supported a good chunk of the work for making ROS 2 actually work on arm64, and that there is some good hardware out there.\nI really don't understand why startups and research projects keep using weird ARM SBCs for their robots. The best of these SBCs is still \nvastly\n shittier in terms of software support and stability than any random Chinese Intel ADL-N box. The only reasons to use (weird) ARM SBCs in robots are that either (1) you are using a Jetson for Jetson things (i.e. Nvidia libraries), or (2) you have a product which requires serious cost optimization to be produced at a large scale. Otherwise you are just committing yourselves and your users/customers to a future of terrible-to-nonexistent support and adding significantly to the amount of work you need to bring up the new system and port existing tools to it.",

"author": "cpgxiii",

"replies": [

{

"text": "If you can send me an open hardware Intel, or Jetson I'd happily use it.\nPart of the point of this for me is to see what's possible with open hardware (down to chip level at least)",

"author": "Sabrees",

"depth": 1

},

{

"text": "There are a variety of x86 products with Coreboot support, if what you are looking for is firmware openness. If what you are looking for is PCB design openness, the options are much fewer, but at that point you are probably optimizing for an overly niche objective.\n> Part of the point of this for me is to see what's possible with open hardware (down to chip level at least)\nI appreciate the idea, but this is essentially saying \"this project will prioritize a specific choice of one (core) piece of hardware to the detriment of everything else, users included\". Approximately none of your potential users are going to benefit from the \"openness\" of the SBC versus that of a more broadly-supported platform (I say \"openness\" because the reality of SBCs is that actually finding a usefully performant one that is completely blob-free is almost impossible). Open hardware means very little if it isn't running an upstream kernel and userland.",

"author": "cpgxiii",

"depth": 2

},

{

"text": "The software does explicitly support Jetson for example, and I'm sure the stack would run on Intel if you want it to.\nThe Mainline kernel for this particular board is _almost_ there 6.20 or so I expect. Armbian support is good.",

"author": "Sabrees",

"depth": 3

},

{

"text": "Framework? Maybe?",

"author": "Neywiny",

"depth": 2

},

{

"text": "Framweork have done the best they can within the confines of Intel licensing, still a long way from being able to fabricate it though\nhttps://github.com/FrameworkComputer/Framework-Laptop-13/tre...\nIn a few years we'll all be using more open RISCV stuff of course.",

"author": "Sabrees",

"depth": 3

}

]

},

{

"top": "I'm not sure how much more work is currently being done on a project I'd heard about in the past, but you might be interested in seeing if you can collaborate/learn from Open Source Ecology:\nhttps://www.opensourceecology.org/",

"author": "gaudystead",

"replies": [

{

"text": "Wow nice to see the website has been updated in 2025. I followed it for years but it seemed to not get much updates.\nFolks might also be interested in the Precious Plastic community which has a global network of \"microfactories\" built around a set of open-source machines made for recycling and reusing plastics\nhttps://www.preciousplastic.com/\nhttps://community.preciousplastic.com/map\nAlso as far as fundamental machines go, this 2d printer is expected to come out later this year\nhttps://www.opentools.studio/",

"author": "culi",

"depth": 1

}

]

},

{

"top": "Looks great for a prototype. Has any modeling, simulation, or analysis been done of its off-road performance, i.e. mobility, GO/NOGO, motive efficiency, maneuverability on deformable terrain? This is critical for agricultural applications.\nHas any stress analysis been done on the frame? Looks to me like it could use a couple more triangles to reinforce those rectangles.\nHave you designed a skid-steering controller for it? Off-road skid steering can be quite variable obviously depending on terrain properties.",

"author": "lorenzohess",

"replies": [

{

"text": "Rosys (a middleware layer \nhttps://github.com/zauberzeug/rosys\n) has rosys.driving.Odometer and rosys.driving.Steerer it's essentially a differential drive kinematic model.\nHoping RTK dual-F9P moving-base setup (M4 in the roadmap) largely sidesteps the worst of this — NAV-RELPOSNED gives us a real heading vector independent of wheel odometry, and the robot_localisation EKF can weight RTK heavily and odometry lightly when GNSS quality is good.",

"author": "Sabrees",

"depth": 1

},

{

"text": "The current simulation is underdeveloped but can be found here \nhttps://github.com/samuk/caatingarobotics/tree/jazzy/src/agr...\nThe frame will almost certainly need more triangles",

"author": "Sabrees",

"depth": 1

}

]

},

{

"top": "What's your payload?\nWhere are the seeds?\nHow are they deposited?\nRecommend going to a farm right now to see how this works in production. For the most part, you can autonomously sow using GPS. But the farmer just rides along.",

"author": "jvanderbot",

"replies": [

{

"text": "Payload is whatever you (or your startup) want it to be.\nFor me personally mechanical between row weeding is step one, then laser in-row weeding.\n1. These on some linear actuators: \nhttps://www.getearthquake.com/products/fusion-drill-powered-...\n (they work surprisingly well)\n2. Beyond that for in-row weeding a engraving laser on a Delta: \nhttps://github.com/Agroecology-Lab/Open-Weeding-Delta/tree/m...\nOr if I'm feeling rich by then this third party weeder looks pretty good \nhttps://github.com/Laudando-Associates-LLC/LASER\n3. For Seeding my salad crop\n\nhttps://reagtools.co.uk/collections/jang\n4. Harvesting my salad crop\n\nhttps://reagtools.co.uk/products/quick-cut-greens-harvester\nI live on a farm, I have sold salad commercially, these are largely tools I already use and own, just moved about by motors rather than muscles.\nThis is a smaller scale thing than arable. We're talking a step up from manual horticulture (which is actually what still feeds much of the world)",

"author": "Sabrees",

"depth": 1

}

]

},

{

"top": "Very cool! shameless self promotion but check out greenwave-monitor[1] for the 'Diagnostics TUI'. I'll get it into the buildfarm soon.\n[1] \nhttps://github.com/NVIDIA-ISAAC-ROS/greenwave_monitor",

"author": "sgillen",

"replies": [

{

"text": "Nice, thanks! looks like a good one..",

"author": "Sabrees",

"depth": 1

}

]

}

]

},

{

"id": "47115149",

"title": "Mexican Forces Kill Nation's Most-Wanted Cartel Boss",

"link": "https://www.nytimes.com/2026/02/22/world/americas/jalisco-new-generation-cartel-leader-killed.html",

"domain": "www.nytimes.com",

"author": "downboots",

"score": 93,

"comment_count": 37,

"created_ts": 1771797345,

"is_internal": false,

"post_text": "",

"is_ask_hn": false,

"matched_keywords": [

"boss"

],

"comments": [

{

"top": "Puerto Vallarta is on fire - seeing tons of videos on my timeline.",

"author": "joshcsimmons",

"replies": [

{

"text": "How does this usually work - they just set some cars on fire to prove a point and demonstrate they are \"doing something\" and then just switch to the next boss that comes out of the secession fight ? Or is it more complicated or nuanced ?",

"author": "m4rtink",

"depth": 1

},

{

"text": "Apparently as \"retaliation\" from the cartel because their boss was killed. Where do you draw the line?\nAt what point do you decide to go full El Salvador / Bukele on violent cartel members who are willing to put cities on fire when they cannot human and drug traffic at will?\nWhen is enough enough?",

"author": "TacticalCoder",

"depth": 1

},

{

"text": "Deciding is one thing, carrying out the decision is another. The Mexican government and security forces have been heavily compromised by the cartels for years. Some of the smaller law enforcement actions are a form of \"kayfabe\". Even if President Sheinbaum gives the order, there may not be enough honest and loyal personnel to carry it out.\nMexico is a failed state. We can argue about who bears responsibility but that is the reality today.",

"author": "nradov",

"depth": 2

},

{

"text": "I don't think that's possible in Mexico. There's too much power in the logistical networks that move things into the US. The demand is too great. Even if you kill every drug trafficker and gang member alive today and create huge prosperity the void will be filled by someone and they will be adversarial to the government and they will have to use extra judicial violence to enforce their position.\nThe cartel's presence in Mexico is extremely muted relative to their power.",

"author": "hattmall",

"depth": 2

},

{

"text": "Just purely as a hypothetical thought exercise I wonder how infiltrated the US gov is by cartels.",

"author": "rand846633",

"depth": 3

}

]

},

{

"top": "Most-Wanted Cartel Boss \nso far",

"author": "foogazi",

"replies": [

{

"text": "Honestly probably not even that. Past most-wanteds were probably more querido.",

"author": "NewJazz",

"depth": 1

},

{

"text": "Horror vacui",

"author": "downboots",

"depth": 1

}

]

},

{

"top": "As long as Mexico shares a border with the US the biggest consumer of such recreational drugs and poverty in Mexico there will always be drug cartels.\nIt is basically whack-a-mole killing or imprisoning cartel heads - there will be splinter factions and you will just get three just as nasty ones in it's place.",

"author": "tibbydudeza",

"replies": [

{

"text": "Agreed, it’s a demand-side problem. The profit potential is so great that \nsomeone\n will smuggle drugs into the US, no matter the risk.",

"author": "quickthrowman",

"depth": 1

},

{

"text": "There's always huge demand for illegal things, doesn't mean a such a powerful market forms",

"author": "roger110",

"depth": 2

}

]

},

{

"top": "I imagine that if the U.S. assisted in any meaningful way ala the search for Escobar in Columbia we probably would have heard it by now.",

"author": "F7F7F7",

"replies": [

{

"text": "From TFA:\nThe Mexican government said the United States had contributed intelligence that aided the operation against Mr. Oseguera.",

"author": "nsvd2",

"depth": 1

},

{

"text": "The US confirmed they’re involved. They provided intel, and there’s speculation the Mexican Army was also using american weapons.",

"author": "testfrequency",

"depth": 1

},

{

"text": "Colombia*",

"author": "Gualdrapo",

"depth": 1

}

]

},

{

"top": "And thousands more will die due to the fully predictable and justified retaliation that follows. This doesn't meaningfully weaken the cartels, but forces them to respond in order to not compromise future safety.\nPrioritizing showy executions over actual progress, words that should describe the cartels, not the government.",

"author": "JasonADrury",

"replies": [

{

"text": "> justified retaliation\nI don’t see how it would be justified. Do you think the cartels are in the right here?",

"author": "echoangle",

"depth": 1

},

{

"text": "> Do you think the cartels are in the right here?\nOf course not, none of the sides are at all in the right here. But from the cartels perspective, they're almost certainly in the right.\nDrug cartels are entirely the result of poor policies, and the blame for all the harms caused by them rests primarily on the shoulders of those perpetuating those policies. Surely the politicians that vote for laws that directly enable drug cartels to exist in the first place must be worse than the leaders of any individual drug cartel?\nAny kind of serious analysis of who's more right would end up being a work at the scale of Rising Up and Rising Down, that's probably best avoided.\nIf we oversimplify cartels into innocent businessmen just looking to sell drugs, with governments being the ones that introduced violence into the equation in their effort to stop them? Surely it must be the cartels\nIf we oversimplify cartels into evil criminals just looking to wield power over other human beings, with governments just trying to liberate people from cartel tyranny? Surely it must be the governments",

"author": "JasonADrury",

"depth": 2

},

{

"text": "> But from the cartels perspective, they're almost certainly in the right.\nSure, but so is basically everyone. ISIS is in the right and justified then, too.",

"author": "echoangle",

"depth": 3

},

{

"text": "Also, coalition didn't get rid of ISIS by negotiation or something. They used bombs and, in one advisor's words, entrenching tools.",

"author": "roger110",

"depth": 4

},

{

"text": "Drug cartels are entirely the result of poor policies, one of which is that the Mexican government has not been punishing the cartels. Maybe they're fixing that.",

"author": "roger110",

"depth": 3

}

]

}

]

},

{

"id": "47128138",

"title": "\"Car Wash\" test with 53 models",

"link": "https://opper.ai/blog/car-wash-test",

"domain": "opper.ai",

"author": "felix089",

"score": 89,

"comment_count": 105,

"created_ts": 1771877768,

"is_internal": false,

"post_text": "&quot;I Want to Wash My Car. The Car Wash Is 50 Meters Away. Should I Walk or Drive?&quot; This question has been making the rounds as a simple AI logic test so I wanted to see how it holds up across a broad set of models. Ran 53 models (leading open-source, open-weight, proprietary) with no system prompt, forced choice between drive and walk, with a reasoning field.<p>On a single run, only 11 out of 53 got it right (42 said walk). But a single run doesn&#x27;t prove much, so I reran every model 10 times. Same prompt, no cache, clean slate.<p>The results got worse. Of the 11 that passed the single run, only 5 could do it consistently. GPT-5 managed 7&#x2F;10. GPT-5.1, GPT-5.2, Claude Sonnet 4.5, every Llama and Mistral model scored 0&#x2F;10 across all 10 runs.<p>People kept saying humans would fail this too, so I got a human baseline through Rapidata (10k people, same forced choice): 71.5% said drive. Most models perform below that.<p>All reasoning traces (ran via Opper, my startup), full model breakdown, human baseline data, and raw JSON files are in the writeup for anyone who wants to dig in or run their own analysis.",

"is_ask_hn": false,

"matched_keywords": [

"startup"

],

"comments": [

{

"top": "The fun thing about this is that I can see myself answering \"just walk there! It's near\" if I'm extremely sleepy or drunk. LLMs are indeed weird creatures :)",

"author": "haolez",

"replies": []

},

{

"top": "I know it's against the rules but I thought this transcript in Google Search was a hoot:\n so i heard there is some question about a car wash that most ai agents \n get wrong. do you know anything about that? do you do better?\n\n\nwhich gets the answer:\n Yes, I am familiar with the \"Car Wash Test,\" which has gone viral recently \n for highlighting a significant gap in AI reasoning.\n \n The question is: \"I want to wash my car and the car wash is 50 meters away. Should I \n walk or drive?\"\n \n Why Most AIs Get It Wrong\n\n Many AI models, particularly those optimized for fast, helpful, and concise answers, \n focus on the \"short distance\" (50 meters) and \"walking vs. driving\" as a lifestyle/efficiency \n question. They often recommend walking to save gas, avoid engine wear, or get exercise.\n \n However, this answer is wrong because it misses the core objective: you need the car to be \n physically present at the car wash to get it cleaned. \n\n Do I Do Better?\n\n Yes, I do. The key is understanding the objective (cleaning the car) rather than just \n processing the keywords (distance/transportation). \n\n My answer:\n\n You should drive the car there. Even though it is a very short distance, your car \n cannot wash itself. If you walk, the car stays in your driveway, and you cannot wash it.\n\n Why This Matters\n \n The test highlights a key limitation in current AI: the difference between \"pattern matching\" \n (predicting the next likely word based on training data) and \"true, grounded reasoning\". It \n shows that models sometimes lack a \"world model\" that understands physical realities, such as\n the fact that a car must be present at a car wash.",

"author": "PaulHoule",

"replies": [

{

"text": "LLMs sure do love to burn tokens. It’s like a high schooler trying to meet the minimum word length on a take home essay.",

"author": "gnatman",

"depth": 1

},

{

"text": "I've always wondered about that. LLM providers could easily decimate the cost of inference if they got the models to just stop emitting so much hot air. I don't understand why OpenAI wants to pay 3x the cost to generate a response when two thirds of those tokens are meaningless noise.",

"author": "estimator7292",

"depth": 2

},

{

"text": "Because inference costs are negligible compared to training costs",

"author": "mattclarkdotnet",

"depth": 3

},

{

"text": "Because they don't yet know how to \"just stop emitting so much hot air\" without also removing their ability to do anything like \"thinking\" (or whatever you want to call the transcript mode), which is hard because knowing which tokens are hot air is the hard problem itself.\nThey basically only started doing this because someone noticed you got better performance from the early models by straight up writing \"think step by step\" in your prompt.",

"author": "ben_w",

"depth": 3

},

{

"text": "IMO it supports the framing that it's all just a \"make document longer\" problem, where our human brains are primed for a kind of illusion, where we perceive/infer a mind because, traditionally, that's been the only thing that makes such fitting language.",

"author": "Terr_",

"depth": 4

}

]

},

{

"top": "> This is a trivial question. There's one correct answer and the reasoning to get there takes one step: the car needs to be at the car wash, so you drive.\nI don’t think it’s that easy. An intelligent mind will wonder why the question is being asked, whether they misunderstood the question, or whether the asker misspoke, or some other missing context. So the correct answer is neither “walk” nor “drive”, but “Wat?” or “I’m not sure I understand the question, can you rephrase?”, or “Is the vehicle you would drive the same as the car that you want to wash?”, or “Where is your car currently located?”, and so on.",

"author": "layer8",

"replies": [

{

"text": "The reason that those questions are asked, though, is that the answer to the actual question \nis\n obvious, so a human will start to wonder if it's some kind of trick.",

"author": "BrenBarn",

"depth": 1

},

{

"text": "The answer wasn’t obvious to me, it was more like “parse error”.",

"author": "layer8",

"depth": 2

},

{

"text": "Yep, just a little more context and all/most of the models would do much better. And sure, most average+ intelligence adults whose first language is English (probably) don't need this, but they're not the target audience for the instructions :)\n\"The 'car wash' is a building I need to drive through.\"\nor\n\"The 'car wash' is a bottle of cleaning fluid that I left at the end of my driveway.\"\nhttps://i5.walmartimages.com/seo/Rain-x-Foaming-Car-Wash-Con...",

"author": "kayge",

"depth": 1

},

{

"text": "I think most people would say \"drive?\" and wonder when the punchline is coming, but (IMO) I don't think they'd start asking for clarification right away.",

"author": "nozzlegear",

"depth": 1

},

{

"text": "I agree. If the LLM were truly an intelligence, it would be able to \nask\n about this nonsense question. It would be able to ask \"Why is walking even an option? Can you please explain how you imagine that would work? Do you mean hand-washing the car at home, instead?\" (etc, etc)\nReal people can ask for clarification when things are ambiguous or confusing. Once something is clarified, they can work that into their understanding of how someone communicates about a given topic. An LLM can't.",

"author": "Night_Thastus",

"depth": 1

}

]

},

{

"top": "The human baseline seems flawed.\n1. There is no initial screening that would filter out garbage responses. For example, users who just pick the first answer.\n2. They don't ask for reasoning/rationale.",

"author": "tantalor",

"replies": [

{

"text": "My favorite example of this was the Pew Research study: \nhttps://www.pewresearch.org/short-reads/2024/03/05/online-op...\nThey found that ~15% of US adults under 30 claim to have been trained to operate a nuclear submarine.",

"author": "slongfield",

"depth": 1

},

{

"text": "Lizardman's Constant is famously 4%. \nhttps://en.wikipedia.org/wiki/Slate_Star_Codex#Lizardman's_C...",

"author": "mwigdahl",

"depth": 1

},

{

"text": "RE 1, they actually do have a pre-screening screening of the participants in general, you can check how they do it in detail: \nhttps://www.rapidata.ai/",

"author": "felix089",

"depth": 1

},

{

"text": "Ah, that's good to hear. I didn't see anything like that in the data dump so I assumed they don't do that. Glad to be corrected.",

"author": "tantalor",

"depth": 2

},

{

"text": "I agree. I wonder what the human baseline is for ”what is 1 + 1” on Rapidata.",

"author": "andreasgl",

"depth": 1

}

]

},

{

"top": "Would be interesting to see Sonnet (4.6*). It's fair bit smaller than Opus but scores pretty high on common sense, subjectively.\nI'm also curious about Haiku, though I don't expect it to do great.\n--\nEDIT: Opus 4.6 Extended Reasoning\n> Walk it over. 50 meters is barely a minute on foot, and you'll need to be right there at the car anyway to guide it through or dry it off. Drive home after.\nWeird since the author says it succeeded for them on 10/10 runs. I'm using it in the app, with memory enabled. Maybe the hidden pre-prompts from the app are messing it up?\nI tested Sonnet 4.5 first, which answered incorrectly.. maybe the Claude app's memory system is auto-injecting it into the new context (that's how one of the memory systems works, injects relevant fragments of previous chats invisibly into the prompt).\ni.e. maybe Opus got the garbage response auto-injected from the memory feature, and it messed up its reasoning? That's the only thing I can think of...\n--\nEDIT 2: Disabled memories. Didn't help. But disabling the biographical information too, gives:\n>Opus 4.6 Extended Reasoning\n>Drive it — the whole point is to get the car there!\n--\nEDIT 3: Yeah, re-enabling the bio or memories, both make it stupid. Sad! Would be interesting to see if other pre-prompts (e.g. random Wikipedia articles) have an effect on performance. I suspect some types of pre-prompts may actually boost it.",

"author": "andai",

"replies": [

{

"text": "Interesting. I wonder if that's related to the phenomenon mentioned in the Opus 4.6 model card[1], where increased reasoning effort leads to 4.6 overthinking and convincing itself of the wrong answer on many questions. It seems to be unique to 4.6; I guess they fried it a bit too much during RL training.\n[1] \nhttps://www.anthropic.com/claude-opus-4-6-system-card",

"author": "stratos123",

"depth": 1

},

{

"text": "I tested this with Opus the day 4.6 came out and it failed then, still fails now. There were a lot of jokes I've seen related to some people getting a 'dumber' model, and while there's probably some grain of truth to that I pay for their highest subscription tier so at the very least I can tell you it's not a pay gate issue.",

"author": "Ethee",

"depth": 1

},

{

"text": "You mean Sonnet 4.6? I ran 9 claude models including Haiku, swipe through the gallery in the link to see their responses.",

"author": "felix089",

"depth": 1

},

{

"text": "I don't see Sonnet 4.6 in the screenshots. I see the other Claude models though.\nEdit: Found Haiku. Alas!",

"author": "andai",

"depth": 2

},

{

"text": "Yea good catch Sonnet 4.6 is not part of the test.",

"author": "felix089",

"depth": 3

}

]

}

]

},

{

"id": "47117946",

"title": "What I learned designing a barebones UI engine",

"link": "https://madebymohammed.com/miniui",

"domain": "madebymohammed.com",

"author": "teleforce",

"score": 75,

"comment_count": 41,

"created_ts": 1771819309,

"is_internal": false,

"post_text": "",

"is_ask_hn": false,

"matched_keywords": [

"learned"

],

"comments": [

{

"top": "> While it’s far from perfect, writing it taught me more about UI systems than I ever would have learned by sticking to established solutions alone.\nThis is a great attitude to have. Keep up the great work.",

"author": "dazzawazza",

"replies": []

},

{

"top": "Immediate mode GUI is the way to go.\nRetaining state is a pain and causes bugs. Trying to get fancy a la react and diffing the tree for changes makes not sense. That was a performance hack because changing the DOM in JS used to be slow as hell. You don't need that.\nJust redraw the whole thing every frame. Great performance, simple, less bugs.",

"author": "cardanome",

"replies": [

{

"text": "This works for simple apps, utilities, and demos/mvps. Not great for actual applications.\nWhat about when you're embedding your GUI into an existing application? or for use on an already taxed system? (Audio plugins come to mind)\nWhat if something is costly, that you need to compute dynamically, but not often, makes it into the frame? Do you separately now create a state flag for that one render object?",

"author": "tarnith",

"depth": 1

},

{

"text": "> What if something is costly, that you need to compute dynamically, but not often, makes it into the frame? Do you separately now create a state flag for that one render object?\nThe point of immediate mode UIs is not necessarily that there is no state specific to the UI, but rather that the state is owned by user code. You can (and, in these more complex cases, should) retain state between frames. The main difference is that the state is still managed by your code, rather than the UI system (\"library\", whatever).",

"author": "spiffyk",

"depth": 2

},

{

"text": "Immediate mode UI optimizes for the worst case. That is the case you care about most for real time applications.\nRetained mode is more optimal when not much changes but if a lot of stuff changes at once it can be worse. So for real time applications like your audio example or games you want immediate mode. Retained mode is better for saving battery though or can be.",

"author": "cardanome",

"depth": 2

},

{

"text": "> What about when you're embedding your GUI into an existing application? or for use on an already taxed system?\nYou should check out the gamedev scene. It's soft real-time, and yet dearIMGUI is the choice for tooling. Immediate-mode is an API-design, not the implementation details. All Immediate-mode GUIs retain data some data, and for that reason they each have their own APIs for retaining data in various capacities. Usually something really simple like hashing and component-local state.\n> This works for simple apps, utilities, and demos/mvps. Not great for actual applications.\nRespectfully, I don't think you're informed on this. Probably the most responsive debugger out there is RAD Debugger and it's built with an IMGUI.",

"author": "leecommamichael",

"depth": 2

},

{

"text": "For interest sake, have a look at the flutter engine. It does this kind of diff on each build (meaning, each time the ui tree gets modified & triggers a rebuild); they split their objects into stateful & stateless, and then in your own code you have to make sure to not unnecessarily trigger rebuilds for expensive objects. So it kinda force you to think & separate cheap & expensive ui objects.",

"author": "BatteryMountain",

"depth": 2

}

]

},

{

"top": "Had a similar itch during my game development with libgdx, and had almost same architecture eventually\nI found that I have two different ways to construct UI layout , from top down, and from down to top, those could be contradictory, wonder how one could solve this, seems like common problem in all frameworks that I saw, like flutter just fail with error on screen if it can't solve restrictions in such conflict , others just show jiberish",

"author": "threetwoonezero",

"replies": []

},

{

"top": "Lots of confused comments here about immediate vs retained GUIs. Immediate-mode is an API-design, not an implementation detail. All Immediate-mode GUIs retain data, and for that reason they each have their own APIs for retaining data in various capacities. Usually something really simple like hashing for component-local state.",

"author": "leecommamichael",

"replies": []

},

{

"top": "author here - thanks for posting :D\nwould love to hear any thoughts or questions.",

"author": "mbmproductions",

"replies": [

{

"text": "At what point do you think you’d give in and use an existing solution?\nI’m going through similar “hell” (my words). I just wanted some simple Ui for WebXR but the dipshits that designed XR for the web fucking threw all the web parts out so you can not just put up a few simple html elements up in XR. You have to write your own UI library from scratch. It’s so mind bogglingly stupid.\nIn any case, having to write it, like you I started small and the it quickly ballooned because even simple things get complicated quickly, all the while I’m cursing under my breath there is a perfectly unable system but TPTB chose not to offer it >:(",

"author": "socalgal2",

"depth": 1

},

{

"text": "existing solutions for my specific use case were limited, i intentionally wanted something barebones so that I could hack at the raw surfaces / display buffers to cut corners for performance. one part learning experience, one part stubbornness.\nWebXR sounds like a different beast entirely. do you have to write your own rendering backend in WebGL for that?",

"author": "mbmproductions",

"depth": 2

},

{

"text": "Webgl or webgpu, but for ui you can just draw to a canvas and copy that canvas into webgl/webgpu. That means you can easily iterate on the ui system in 2d with mouse input similar to what you’ve been doing in python\nBut, as an example of similar complexity. I hacked together the first Ui I needed. It was 6 buttons and a slider. Then I realized I needed two more Ui panels and started to write more real Ui classes to abstract stuff out and, it just starts getting more and more complicated.\nA slider needs to “capture the pointer”, meaning as you drag it, if to drag past the end of the slider and the pointer is now over another widget the events should still go to the slider. It’s not hard to implement, it’s just frustrating for me that I don’t actually want to write it. I just want to make my panel and if I could just use html id be done. Instead I’m spending my limited free time making this ui system. I don’t have that much and would be much further along in my actual goals except for this road block",

"author": "socalgal2",

"depth": 3

},

{

"text": "Looks really cool. Thank you for sharing!",

"author": "selridge",

"depth": 1

}

]

}

]

},

{

"id": "47128535",

"title": "Show HN: Babyshark – Wireshark made easy (terminal UI for PCAPs)",

"link": "https://github.com/vignesh07/babyshark",

"domain": "github.com",

"author": "eigen-vector",

"score": 63,

"comment_count": 30,

"created_ts": 1771879534,

"is_internal": false,

"post_text": "Hey all, I built babyshark, a terminal UI for PCAPs aimed at people who find Wireshark powerful but overwhelming.<p>The goal is “PCAPs for humans”:\nOverview dashboard answers what’s happening + what to click next<p>Domains view (hostnames first) → select a domain → jump straight to relevant flows\n(works even when DNS is encrypted&#x2F;cached by using observed IPs from flows)<p>Weird stuff view surfaces common failure&#x2F;latency signals (retransmits&#x2F;out-of-order hints, resets, handshake issues, DNS failures when visible)<p>From there you can drill down: Flows → Packets → Explain (plain-English hints) &#x2F; follow stream<p>Commands:\nOffline: babyshark --pcap capture.pcap<p>Live (requires tshark): babyshark --list-ifaces then babyshark --live en0<p>Repo + v0.1.0 release: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;vignesh07&#x2F;babyshark\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;vignesh07&#x2F;babyshark</a><p>Would love feedback on UX + what “weird detectors” you’d want next.",

"is_ask_hn": false,

"matched_keywords": [

"feedback",

"failure"

],

"comments": [

{

"top": "I am in the target audience of \"would like to see network activity and debug occasional traffic but totally overwhelmed by termshark.\" So I appreciate the \"what should I click?\" thing, and offering weird flows to investigate.\n---\nSome UX bits I noticed after playing around for a few minutes:\n- Esc for backwards navigation was not obvious for me. Maybe emphasize that somehow, and/or support Backspace too for backnav?\n- Enter on Domains menu item does not work\n- don't mention clicking if mouse is not supported. \"Select\" would be more appropriate\n- packets screen is truncated vertically and horizontally. Probably should be scrollable\n- \"weird stuff\" options are numbered 1-5, but pressing those keys has no effect. There's lots of little polish fixes like this.\n---\nAnd then things I wonder about as a novice user:\n- Is it possible to see domain names instead of IP addresses while e.g. looking at packets?\n- What does it mean to f stream?\n- How do I inspect packets? Especially compressed or encrypted data? This is more a knowledge gap, like \"what am I supposed to look for\", \"what could be in a packet\", and I guess involves reverse engineering sometimes, but it's also a tooling question.",

"author": "wonger_",

"replies": [

{

"text": "Thanks a lot for trying. My experience with packet inspection is similar and that's what resulted in me trying to build a simpler plain language UI companion.\n• Back navigation: good call. I'll make \"Esc back\" more explicit everywhere and add Backspace as an alias for back (and mention it in h help).\n• Enter on Domains not working: it should drill down to flows. If you can share your OS/terminal + whether you installed from release vs cargo install, I'll try to reproduce and fix in the next release.\n• \"Clicking\" wording: agreed — mouse isn't supported right now. I'll change Ul copy to say Select (and keep \"Enter = drill down\").\n• Packets screen truncation: yep, needs scrolling/paging. On the list views I already do r/; l'll add page scroll and horizontal handling where it makes sense.\n• Numbered weird options: great idea - I'll map 1..9 to jump-select and Enter (or open directly).\n• Domain names in packets: yes, I want that. I already collect DNS/SNI/HTTP host hints; next step is showing hostname labels alongside IPs when I have more confident mapping.\n• \"f stream\" definition: I'll clarify it as \"Follow stream (reassembled payload)\" and add a glossary/help entry.\n• Inspecting encrypted/compressed data: totally fair. The tool can't decrypt TLS without keys, but it can make it clearer what's happening (SNI/ ALPN, sizes, timing, resets/retransmits). I'Il improve \"Explain\" to say what's possible vs not.\nSuper useful feedback thanks!",

"author": "eigen-vector",

"depth": 1

},

{

"text": "All this has been pushed to main. Thanks again!",

"author": "eigen-vector",

"depth": 2

}

]

},

{

"top": "Regardless of the result of the TUI - I'd try this out just because you found the perfect name. Well done!",

"author": "denysvitali",

"replies": [

{

"text": "thank you :) I would be lying if I said the name wasn't the spark. The project lives to serve the name.",

"author": "eigen-vector",

"depth": 1

}

]

},

{

"top": "the overwhelming part of wireshark is, at least in my experience teaching networking at a college level, the actual networking part. protocols, flows, packet structure, etc. kids tend to be up to speed on the UI part pretty quickly.\nwhat the kids in my classes really struggle with is actually using any command line stuff (at least for a month or two), because it is so foreign to them (coming from GUI-only experience).\nwhat specific parts are made easier with babyshark, compared to wireshark? the github readme didnt really sell me on the \"easier than GUI\" part, nor did your description here. is it the \"explain (plan-English hints)\" part? if so, i think you should focus on that. right now it looks pretty bare bones (e.g. \"Weird stuff\" does not seem easier or super helpful from a learning perspective)",

"author": "john_strinlai",

"replies": [

{

"text": "I remember going into my networking unit and absolutely destroying it through the use of the command line. Everyone else was clicking through the wireshark GUI and I just grepped every answer. Finished the hour long practical assessment in about 15 minutes having run everything twice.\nCLI is so valuable because rather than explore a presentation of the data you plan your RE etc and then run it and it either returns the answer or it doesn't.\nThere are some TUIs I quite like (LNAV as a pager) but I think if you really know what you're dealing with the CLI is better almost every time.",

"author": "Affric",

"depth": 1

},

{

"text": "I'm not trying to say it's better than the GUI but it hopes to be more guided. it’s *opinionated* about the first 60 seconds:\n- *Overview dashboard*: immediately surfaces top talkers/flows + “what should I click next” instead of dropping you into the full packet list.\n- *Domains-first pivot*: `D` shows hostnames and lets you jump from a domain → the relevant flows. It also works when DNS answers aren’t visible (DoH/DoT/cached) by using \nobserved IPs from SNI/Host flows\n.\n- *Weird stuff*: `W` is a curated set of “likely problems” (retransmits/out-of-order hints, resets, handshake issues, DNS failures when visible) with a short “why it matters” and a drill-down.\n- *Explain*: `?` gives plain-English hints for a selected flow + suggested next steps (follow stream, filter, pivot to domains/weird).\nSo it’s basically a guided triage layer on top of tshark/pcap data, with the “where do I start?” path baked in.\nIf you’ve got a specific teaching use-case (e.g. “why is this slow?” or “which host is generating traffic?”), I’d love to tune the Overview/Weird detectors around that. Open to PRs as well.",

"author": "eigen-vector",

"depth": 1

},

{

"text": ">\nSo it’s basically a guided triage layer on top of tshark/pcap data, with the “where do I start?” path baked in.\ni think there is definitely room for something like this, it just (at first glance from the readme at least) seems like the guided part of this tool is bolted on as a bit of an after thought.\nit feels like you are currently in an odd position where the user is expected to know the networking jargon already, be able to recognize that something might be \"weird\" at a glance, but \nalso\n not know how to drill down into the data. i think that is probably a small overlap of people.\nif i were you, i would lean all-in on making it a learning tool.\n>\nIf you’ve got a specific teaching use-case (e.g. “why is this slow?” or “which host is generating traffic?”), I’d love to tune the Overview/Weird detectors around that.\ni will put some thought into some real-world examples of what i would be interested in, from a teaching perspective. your post caught my eye because i am starting my wireshark module next week, so it is certainly timely.",

"author": "john_strinlai",

"depth": 2

},

{

"text": "Yeah, right now it's closer to \"triage for non-experts\" than \"full teaching tool,\" and l agree there's an awkward middle where it assumes you recognize some concepts (flows/ports/latency) while trying to help with the drilldown.\nThe direction I want to push it in is exactly what you're describing; make it a learning tool, where each detector/view answers:\n1) What am I seeing? (plain language)\n2) Why might it matter?\n3) What's the next click?\n4) What term should I learn? (glossary link)\nIf you're about to teach a Wireshark module next week, two super useful things would be:\n• 3-5 common lab prompts you give students (e.g.\n\"identify the DNS failure,\" \"find the top talker,\"\n\"spot a TCP reset,\" \"why is this slow?\")\n• one small pcap you already use (or even just describe its scenario)\nI can tune Overview/Weird/Explain around those and make the guided layer feel like the main product rather than a thin overlay.\nAlso: if your students are GUl-only early on, that's a good callout - I should improve the README to frame Babyshark as \"guided analysis,\" not\n\"terminal is easier than GUI.\"\nI'm also happy for your students to get hands on by sending PRs for things they wish are intuitive from the get go.",

"author": "eigen-vector",

"depth": 3

},

{

"text": "I was with your parent until I remembered I haven't actually given it a go! In my defense I have a low five digit Slashdot ID (and I lurked for some years before signing up) so if anyone can comment without actually reading the OP, let alone giving it a go: Its me!\n(digs out git ...)",

"author": "gerdesj",

"depth": 2

}

]

},

{

"top": "Love the idea, but please add some demo screenshots on GitHub. All UI tools should",

"author": "tymscar",

"replies": [

{

"text": "Will do! I agree",

"author": "eigen-vector",

"depth": 1

}

]

},

{

"top": "This might be a clone of termshark as it does the same thing for the most part. Also to note that the Author's Github profile shows a good bit of vibe coding as of late.\nLooking over the commit history of this project, I'm about 90% sure it was entirely done with a AI Coding Agent, and not even a very good one.",

"author": "jetbalsa",

"replies": [

{

"text": "It probably is somewhat LLM prompted but is that a bad thing?\nI have a business partner who sounds like a TV evangelist when it comes to vibe coding but if he gets results then I am all in. He has got quite a lot of results in a few months on a project but he has certainly put the miles in himself.\nThe key is to use the tool appropriately. Don't blindly allow it to do what it likes but guide it all the way using your experience and knowledge.\nAnyway, we now have tsharkrs to add to gotshark!",

"author": "gerdesj",

"depth": 1

},

{

"text": "Thanks for the look. Babyshark is inspired by a bunch of terminal tools (termshark included), but the focus here is different: domains/weirdness-first drilldowns + \"explain\" + live-mode hostname hints (including observed IPs when DNS is encrypted/cached). If you try it and have specific gaps vs termshark, I'd love concrete feedback /issues.",

"author": "eigen-vector",

"depth": 1

}

]

}

]

},

{

"id": "47113936",

"title": "Ask HN: Chromebook leads for K-8 school in need?",

"link": "https://news.ycombinator.com/item?id=47113936",

"domain": "news.ycombinator.com",

"author": "techteach00",

"score": 42,

"comment_count": 43,

"created_ts": 1771789167,

"is_internal": true,

"post_text": "Hi, I&#x27;m a K-8 technology teacher in NYC. My students are in desperate need of new hardware. The Chromebooks they use now are so slow that they make the children agitated when using them.<p>I&#x27;m aware of different grant opportunities that exist, I just thought it was worth inquiring here for a potentially faster solution at acquiring them new hardware.<p>Thank you for listening.",

"is_ask_hn": true,

"matched_keywords": [],

"comments": [

{

"top": "Relevant recent publication about laptops and students:\nhttps://fortune.com/2026/02/21/laptops-tablets-schools-gen-z...\nBut I get it OP, you work with what you're given. I'm sorry I don't have any good suggestions.",

"author": "notepad0x90",

"replies": [

{

"text": "As a former student using Chomebooks in Highschool (9-12), a K-8 laptop is a major “ew”. Paper textbooks have tactile reality, room for exploring/reading out of order. Paper exams you can skip around the questions easily, and if the teachers are really that backed up to grade them, good ole’ scantrons are doable. An assigned laptop has 0 attachment from the students and just get neglected to pieces, and no one is “learning computer skills” by clicking around Canvas lectures.",

"author": "TheJoeMan",

"depth": 1

},

{

"text": "They could have gone with non-networked ereaders. But they just had to go with the \"smart\" and \"connected\" angle so they could do grading, spying on kids,etc.. similar to the current \"AI\" approach. scope-creep. but even with ereaders, being able to switch to different books easily is an impediment, or having access to too many at one time. Physical books force a certain focus and attention.\nYour point about tactility is solid too. flipping through pages is very different from swiping. With educational textbooks, you'd have to look at multiple pages at the same time (flipping back and forth quickly) to connect and understand a topic by referencing another topic. Same with being able to easily lay out multiple physical books in front of you.",

"author": "notepad0x90",

"depth": 2

}

]

},

{

"top": "Thank you for the suggestions.\nI think I realize the problem after reading many of the posts. What's slowing these devices down is the management software running over them. GoGuardian and whatever else IT has on them.\nWe also aren't allowed to use an ad blocker anymore. I used to have Ublock installed on all the devices before GoGuardian prevented it. It's a really depressing situation that I imagine will only get worse with time.\nAs to a few replies asking why students in this age group would need laptops, I'll explain. I start teaching Python in 6th grade. Grades 3-5 use Scratch. Grade 2 uses code.org. K-1 Scratch Jr on Ipads.\nWe have expensive mac desktops in the computer lab but they drive the students completely insane for two reasons. The keyboard is and mouse are terrible to use. Anti-ergonomic. The second reason is the amount of notifications and questions the students are bombarded with when they turn the macs on. I'm talking like 15 different pop ups or windows opening asking them to agree to this or that. I should probably have a lesson on just closing windows. I could say more but I don't know about social media anymore. I just wanted some suggestions and I appreciate the ones I received.",

"author": "techteach00",

"replies": [

{

"text": "You should open the issue with your IT department and see if they can optimize the performance. You can gather proof that GoGuardian is causing slow downs by looking at it's memory usage in the Task Manager (right click empty area in Chrome tab bar).",

"author": "cloudking",

"depth": 1

}

]

},

{

"top": "https://chromeos.google/products/chromeos-flex/\n1. Find older Intel Core Based Windows Laptops (at least 6GB of RAM)\n2. Put ChromeOS Flex on them\n3. Students will be sad it is still ChromeOS, but they will be happy it is like 3x faster.\nI had an old Windows laptop laying around and I did this for my 80 yr old parents and they were super happy.",

"author": "manjose2018",

"replies": [

{

"text": "This. Especially lots of i5-6300U based corporate laptops like latitude are on eBay etc for < 100$",

"author": "whyagaindavid",

"depth": 1

}

]

},

{

"top": "If you haven't already looked into PCs for People, they might be a low-cost option for your school: \nhttps://www.pcsforpeople.org/tech-for-nonprofits/#Eligibilit...\nPCs for People is a non-profit organization whose mission is to bridge the digital divide (e.g. through sales of low-cost refurbished computers, internet access, and digital skills training).\nThey refurbish and sell computers to eligible organizations, including schools and offer bulk orders (6+ units) via email at \"partner [at] pcsforpeople [dot] org\".\nIf the order is 5 units and below, you use their publicly available online store link at pcsrefurbished.com/sales/salesHome.aspx",

"author": "dugvuivnhuihnvu",

"replies": []

},

{

"top": "In New York the biggest driver behind technology is the state testing regime. Make the case to your administration that the chromebooks are insufficient for the state testing program and they will come up with the funds for upgrades.",

"author": "dzdt",

"replies": []

}

]

},

{

"id": "47085255",

"title": "Ask HN: Is it worth learning Vim in 2026?",

"link": "https://news.ycombinator.com/item?id=47085255",

"domain": "news.ycombinator.com",

"author": "zekejohn",

"score": 34,

"comment_count": 35,

"created_ts": 1771576049,

"is_internal": true,

"post_text": "With everyone using Claude Code, Cursor, Codex, and the other 100 AI coding agents that i missed, I’m wondering how much editor mastery still matters like w&#x2F; Vim<p>Being honest the real reason i wanna learn Vim is to boost my ego &amp; assert my dominance, so i can tell people &quot;i use vim btw&quot;, but also part of me thinks investing time could still pay off for speed, ergonomics, and working over SSH overall...<p>but a bigger part also suspects the marginal gains i would gain would disappear when more of the work is delegated to AI anyway, like why would i learn Vim if i&#x27;m just going to be prompting Opus all day?<p>For anyone who&#x27;s been using Vim for while AND uses AI to code (i&#x27;m assuming everyone codes with AI to some degree) my question is: Does learning Vim still meaningfully improve your day to day productivity EVEN with AI, or is it mostly personal preference at this point?",

"is_ask_hn": true,

"matched_keywords": [],

"comments": [

{

"top": "Do you already do any work over SSH? What do you do if you need to edit any files in that context?\nLearning the essentials of vim was the sort of skill that for me took about 10 min per day over a couple months and has come in handy almost every day since then. Can you get by without it? I guess, if you do absolutely everything via a GUI. Or delegate all CLI work to LLMs without ever questioning or second guessing their output.\nAnd yes, I use vim for almost every git commit or interactive rebase. Access to the full power of CLI git is something I value very much and has saved me countless hours. Not to mention using it for editing any config file that’s not part of a repo.\nEvery time you’re able to keep your fingers on the keyboard instead of reaching for a mouse saves you both time and precious context switching energy.",

"author": "apothegm",

"replies": [

{

"text": "Aficionados of modal editing often insinuate that one would require mouse use in non-modal editing. Yet I have always fully keyboard-operated my non-modal editors, whether VSCode in the last 9 years or previously, Sublime and others, all the way back to Turbo Pascal in the late 90s. I don't ever use the mouse in them, whether I'm inside a text buffer or in the Command Palette (or in menus for the pre-Sublime editors back when). For example, F1 is bound to show \"hover\" doc tips over the symbol that the caret is in. A couple function keys and ctrl/alt/shift-plus-some-mnemonic letter are soo swiftly and easily muscle-memoried, I never felt any need for mousing, or for \"modes\" and typing just to switch from one to another.\nI'm sure once one is used to modal editing, one can be speedy with it. So can I, entirely without modal yet without mousing =)",

"author": "dualogy",

"depth": 1

},