Morning Brief: Thursday, November 12

Seventy-four feeds reporting. Two weeks. 5,716 items reduced to what follows. (what we track, how we crawl, subscribe)

Thursday is the day the Nvidia story crossed from investigation to removal: a board reportedly preparing to oust Jensen Huang, a securities suit over disclosure, and successor lists in the trade press, five days after Congress first summoned him. A second thread runs the other way – the counterparties of the agent economy have started refusing agents by name.

That refusal thread is the one that acquired a mechanism. United Airlines banned AI agents from booking flights on travelers' behalf, and a fingerprinting paper in yesterday's arXiv batch shows browser agents are identifiable from ordinary UI traces, which is what turns a policy announcement into a filter. A Lobsters thread began compiling "No AI" companies, and eight new user agents checked into the Crawler Zoo overnight – two of them general-purpose scanners rather than branded crawlers. Detection is becoming the precondition for refusal.

Top (5-7 min)

Nvidia's stock drops as board reportedly preps to oust CEO Jensen Huang: TechCrunch, 2026-11-11. The succession risk at the company underwriting the entire buildout is now a board matter, not a rumor. Everything downstream of Nvidia capex inherits this.
Nvidia and Jensen Huang face shareholder lawsuit alleging investors were misled about spy affair: Slashdot, 2026-11-11. Disclosure, not conduct, is the legal theory. That is the version of this story that reaches the 10-K.
Congress to call Jensen Huang in to answer for alleged Chinese spy affair: TechCrunch, 2026-11-11. The hearing track, running since the first summons on November 6, and the route by which this reaches export controls.
United Airlines bans AI agents from booking flights on behalf of travelers: Slashdot, 2026-11-12. The first large consumer counterparty to say no to agents by name. Read with the fingerprinting paper below.
Known By Their Actions: Fingerprinting LLM Browser Agents via UI Traces: arXiv cs.AI, 2026-11-11. Browser agents are identifiable from interaction traces alone. This is the mechanism that makes a ban like United's enforceable.
Rewriting less than 10 neurons in an LLM can turn it into a sleeper backdoor: InfoWorld, 2026-11-11. A supply-chain result: the edit is small enough to hide in a weight diff nobody reviews.
Genie One: the AI cowork home for everyone: Databricks, 2026-11-11. Yesterday's up-stack move from a third vendor. The data platform now wants to be the place work happens too.

Themes

The board arrives

The arc in our own window: Congress summoned Huang on November 6, Nvidia opened an internal investigation on the 8th, Trump backed him and Yale's Sonnenfeld put his odds at "about 50-50" on the 9th. On the 11th it became a board reportedly preparing a removal, a securities suit built on disclosure rather than conduct, a second alleged relationship inside the company, and successor lists in the trade press. The chip-export angle is the one with policy consequences: lawmakers are using the scandal as leverage for tighter controls, which lands on the company whose capex schedule every AI infrastructure forecast assumes.

Nvidia's stock drops as board reportedly preps to oust CEO Jensen Huang: TechCrunch, 2026-11-11.
The unfolding Jensen Huang scandal: a timeline: TechCrunch, 2026-11-11. The reference document. Dates matter here more than in most of these stories.
Jensen Huang accused of having another affair – with an Nvidia exec: TechCrunch, 2026-11-11.
Jensen Huang said to have had relationship with Nvidia executive who reports to him: Slashdot, 2026-11-11. The reporting-line detail is what converts this from private conduct into a governance finding.
Nvidia and Jensen Huang face shareholder lawsuit alleging investors were misled about spy affair: Slashdot, 2026-11-11.
Congress to call Jensen Huang in to answer for alleged Chinese spy affair: TechCrunch, 2026-11-11.
Amid Nvidia scandal, lawmakers push for tighter controls on AI chip exports: WBUR, 2026-11-11. The policy consequence. The scandal is being spent on export controls, which is a durable outcome regardless of who runs Nvidia.
Who could replace Jensen Huang? A highly unscientific list of possible successors: TechCrunch, 2026-11-12.

Detection, then refusal

The agent story has a receiving end, and this is the week it got instruments. United's ban is the headline; the enabling result is the fingerprinting paper, because if browser agents leave a distinguishable trace in ordinary UI telemetry then "no agents" stops being an honor system and becomes a filter. The monitoring papers below are the same move pointed inward: disclosure requirements and decode-time probes for agents you are running yourself.

United Airlines bans AI agents from booking flights on behalf of travelers: Slashdot, 2026-11-12.
Known By Their Actions: Fingerprinting LLM Browser Agents via UI Traces: arXiv cs.AI, 2026-11-11.
SetPrint: Black-Box LLM Fingerprinting via Setwise Correspondence of Response Changes: arXiv cs.AI, 2026-11-11. Identifying the model rather than the agent. Same direction of travel: the counterparty wants to know what it is talking to.
What are some "No AI" apps/companies/resources you know of?: Lobsters, 2026-11-11. A demand-side list being assembled in public. Worth reading as market research.
The Myth of Neutral Code: Lobsters, 2026-11-11.
Silicon Valley podcast hosts try to understand why regular people are mad: 404 Media, 2026-11-11.
Transparency Disclosures for Chain-of-Thought Monitorability: arXiv cs.AI, 2026-11-11. What a vendor would have to disclose for CoT monitoring to mean anything. The compliance-shaped version of interpretability.
Train on prefill, monitor at decode: using sparse autoencoders for topic monitoring: arXiv cs.AI, 2026-11-11.
Do AI agents need preventive care? An epidemiologic cohort design for agentic AI: arXiv cs.AI, 2026-11-11. Fleet health as an epidemiological problem rather than a per-run one. Odd framing, useful if you operate many agents.

Enterprise agents, day two

Databricks shipped ten posts in one morning: Genie One, Genie Ontology GA, a 1.6x accuracy claim at under half the tokens, and a connector story aimed at ChatGPT, Claude, Copilot Cowork and Grok at once. AWS's entry was an emulator its own agents wrote and now maintain, which Werner Vogels used as the occasion to argue about what the engineer's job becomes.

Genie One: the AI cowork home for everyone: Databricks, 2026-11-11.
The new Genie: autonomous AI for complex enterprise work: Databricks, 2026-11-11.
Announcing GA of Genie Ontology: Databricks, 2026-11-11.
Genie with Genie Ontology delivers 1.6x the accuracy with less than half the tokens: Databricks, 2026-11-11. A vendor number, but a falsifiable one: accuracy and token spend together rather than either alone.
Bring your business context to ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot: Databricks, 2026-11-11. The platform hedging on which assistant wins by connecting to all of them.
Databricks' new AI agent tells CEO Ali Ghodsi the unvarnished truth: TechCrunch, 2026-11-11.
AWS Emulator and the engineer's job when agents write the code: All Things Distributed, 2026-11-11. Vogels on what is left for the engineer. The artifact is the argument: agents wrote it and agents maintain it.
AWS had its agents build a free and open emulator for AWS. They now maintain it, too.: The New Stack, 2026-11-11.
AWS unveils local emulator service to help enterprises accelerate app development and testing: InfoWorld, 2026-11-11.
aws-emulator: Simon Willison, 2026-11-11.
Why Nitric stopped asking one AI agent to do everything: The New Stack, 2026-11-11.
JioStar's chief architect on AI discovery at 500M-user scale: OpenAI, 2026-11-11.
Gilbert + Tobin rethinks the business of law with AI: OpenAI, 2026-11-12.

Scan

Engineering, tools, and the craft

Tests as data: ensuring AI-generated code does what we want: Martin Fowler, 2026-11-11. The specification problem restated as a data problem. The most practical thing written this week about reviewing agent output.
The AI era's lines of code? A case study on what an engineer's token spend measures: arXiv cs.AI, 2026-11-11. Token spend is the new LOC, with the same pathologies. Useful before anyone puts it on a dashboard.
Understanding-first AI engineering: purposes, reference architecture, and architectural patterns for human understanding: arXiv cs.AI, 2026-11-11.
Choosing QuePaxa, an unusual consensus algorithm: Cloudflare, 2026-11-11. A production consensus choice explained rather than announced. Rare and worth the time.
No effect before its cause: from Einstein to Lamport: Lobsters, 2026-11-11.
Andrew Kensler's permute(), a function for stateless, constant-time pseudorandom-order array iteration: Lobsters, 2026-11-11.
For the first time, a C++ standard failed to pass the national bodies: Lobsters, 2026-11-11. A governance failure in a standards body that has never had one. Watch what the committee does with the comment period.
Developers have been hacking around MCP's file-upload problem for two years. A fix is taking shape.: The New Stack, 2026-11-11.
What we found testing OpAMP clients against the spec: Pydantic, 2026-11-11. Conformance testing against a telemetry spec, with findings. The unglamorous work that makes a protocol real.
llm 0.37.2: Simon Willison, 2026-11-11.
Zellij 0.46 released: LWN, 2026-11-11.
Google Summer of Code 2026 results: Lobsters, 2026-11-11.
Haste web workers: Planet Clojure, 2026-11-12.
Annually-funded developers' update: September and October 2026: Planet Clojure, 2026-11-11.
[$] LWN.net Weekly Edition for November 12, 2026: LWN, 2026-11-12.

Data platforms and local analytics

Local analytics doesn't need a database server: DuckDB, 2026-11-12. The case for the shape of our own warehouse. We run this argument rather than cite it.
Air gapped, not data dark: how Bright Machines gives customers live factory dashboards without opening the plant: Databricks, 2026-11-11.
Expensive accelerators, no workload-level visibility: introducing Amazon CloudWatch accelerated compute insights: The New Stack, 2026-11-11.
CoreWeave tests AI integrations on its cloud before you migrate: The New Stack, 2026-11-11.
Databricks doubles down on AI products, opens new 235,000-square-foot San Francisco headquarters: Databricks, 2026-11-11.

Infrastructure, energy, and the buildout

CoreWeave turns to customer prepayments, shorter contracts to fund AI growth: TechCrunch, 2026-11-11. Prepayments and shorter terms are what financing looks like when the lender is nervous. A leading indicator.
Why the data center you dread might never even be built: TechCrunch, 2026-11-11.
Meta is losing its longtime head of infrastructure: TechCrunch, 2026-11-11.
How a false smoke alarm almost drowned Jane Street's only NYC datacenter: Lobsters, 2026-11-11. A near-miss incident report with the failure chain intact. Better than most postmortems that get published.
Ukraine strikes cause permanent data loss at Russia's Yandex Cloud: Slashdot, 2026-11-11. Cloud durability under kinetic attack. A threat model most availability docs do not contemplate.
Power grids on 5 continents pulse every 60s. Scientists think they know why.: Slashdot, 2026-11-12.
Metamorphic testing of learned AC optimal power flow surrogates reveals representation dependence: arXiv cs.AI, 2026-11-11. Grid surrogates that depend on how you encode the network. Adjacent to the story above, and more worrying.
US and rest of world must follow China's lead on green energy, says John Kerry: Slashdot, 2026-11-11.

Policy, society, and the human read

When reasoning looks like hidden understanding: how chain-of-thought fosters anthropomorphism, perceptions of epistemic superiority, and cognitive deference: arXiv cs.AI, 2026-11-11. Visible reasoning makes people defer more, not judge better. The product implication is uncomfortable.
From bias to positionality: a critical grammar for large language models: arXiv cs.AI, 2026-11-11.
Living constitutions: participatory alignment for sovereign foundation models: arXiv cs.AI, 2026-11-11.
Enforcing axioms for AI alignment under loss-based rules: arXiv cs.AI, 2026-11-11.
AI alignment is neither solvable nor sufficient: Pinboard, 2026-11-12.
Reliability of Claude Code as an agentic AI grading system for handwritten university exams: arXiv cs.AI, 2026-11-11. Somebody graded real exams with a coding agent and measured it. The result matters less than that the measurement exists.
Supporting research on AI and children's safety: OpenAI, 2026-11-12.
Medicare's AI experiment: EFF Deeplinks, 2026-11-11. Prior authorization by model, at Medicare scale. The clearest current example of an agent with real authority over people.
Decisions by discrimination: differential evidence for medical reasoning: arXiv cs.AI, 2026-11-11.
The Download: preserving minds and weighing open AI risks: MIT Technology Review, 2026-11-11.
Court says Trump can't use leaks to justify spying on journalists' LinkedIn DMs: Slashdot, 2026-11-11.
New York forces T-Mobile to end data caps on low-income plans: Slashdot, 2026-11-11.
Apple is killing its support community forum next week: Slashdot, 2026-11-11. Twenty years of user-generated troubleshooting, retired. Somebody should mirror it before next week.
Amazon's Ring unveils flying drone and robot to patrol homes and businesses: Slashdot, 2026-11-11.
How I accidentally became a millionaire: Lobsters, 2026-11-11.
The web is more than enough: Lobsters, 2026-11-11.

Science agents and applied work (arXiv)

Curie's Jarvis: autonomous robotic chemistry experiments from printed protocols: arXiv cs.AI, 2026-11-11. Protocol-to-robot without a human transcription step. The bottleneck it removes is the one everybody complains about.
Embodied science agents: making embodied experimental design learnable: arXiv cs.AI, 2026-11-11.
DrugCoding: a scientific programming agent for drug discovery: arXiv cs.AI, 2026-11-11.
ProTEUS: a mutation-resolved world model for closed-loop protein evolution: arXiv cs.AI, 2026-11-11.
The dose makes the poison: exposure-aware evaluation of generative drug design: arXiv cs.AI, 2026-11-11.
EarthRecord: from pixels to land change with multimodal agents over four decades of satellite imagery: arXiv cs.AI, 2026-11-11.
DisasterAgent: an evidence-driven agent for verifiable disaster report generation from remote sensing imagery: arXiv cs.AI, 2026-11-11.
Using wearables to enable cardiovascular-kidney-metabolic risk profiling for personalized coaching: Google Research, 2026-11-11.
How a metabolic spark could have jump-started life: Quanta Magazine, 2026-11-11.
Scientists discovered an "invisible" object in space. It's challenging our understanding of the cosmos.: 404 Media, 2026-11-11.

Learning theory and reinforcement learning (arXiv)

Transformers provably learn graph search: training dynamics and the exponential power of depth: arXiv cs.AI, 2026-11-11. A separation result with a clean mechanism. Depth buys search, and the proof says how much.
You don't need feature learning to do in-context learning: arXiv cs.AI, 2026-11-11.
Sharp sample complexity of KL-regularized reinforcement learning: arXiv cs.AI, 2026-11-11.
Are dense rewards worth their tokens in GRPO?: arXiv cs.AI, 2026-11-11.
Wrong the same way: cross-prompt sharing both suppresses and amplifies reward hacking in RLVR: arXiv cs.AI, 2026-11-11. Sharing across prompts cuts one failure mode and feeds another. Read before you tune a verifier.
An energy-based perspective on machine learning: arXiv cs.AI, 2026-11-11.
One temperature to rule them all?: arXiv cs.AI, 2026-11-11.
Free lunch uncertainty: sequence-level Rényi entropy from particle decoding for adaptive LLM reasoning: arXiv cs.AI, 2026-11-11.

Robotics, embodiment, and perception (arXiv)

One policy for 100 household tasks: our solution for the 2026 BEHAVIOR Challenge: arXiv cs.AI, 2026-11-11.
How much of your VLA is kNN: arXiv cs.AI, 2026-11-11. An uncomfortable ablation. If retrieval explains the policy, the policy is not what the paper claimed.
APS-VLA: action-guided and future-aware token pruning for efficient vision-language-action models: arXiv cs.AI, 2026-11-11.
Ego Guidance: enhancing robot policy learning with action-free human egocentric videos: arXiv cs.AI, 2026-11-11.
PhyRubric: rubric-grounded diagnostic evaluation of physical commonsense in text-to-video generation: arXiv cs.AI, 2026-11-11.
GTAGym: real AAA game tasks environment for evaluation and reinforcement learning: arXiv cs.AI, 2026-11-11.

Research and math

Lean Grimoire: Terence Tao, 2026-11-12. Tao's working notes on Lean, which is the most useful form his writing takes. He speaks at Harvard on Monday; see Boston below.
Megrez: a knowledge-guided agent for formal mathematics over MK set theory: arXiv cs.AI, 2026-11-11.
Darmok: championship cryptic crossword solving via constraint tightening and resilient orchestration: arXiv cs.AI, 2026-11-11.
Re-Search: dissecting LLM ideation through a two-player questioning game: arXiv cs.AI, 2026-11-11.
CLADE: category-level answering from deficient evidence in retrieval-augmented language models: arXiv cs.AI, 2026-11-11.
In memoriam: Tim Peters, 1952-2026: LWN, 2026-11-11. Timsort, the Zen of Python, and twenty-five years of answering questions on python-dev.
In memoriam: Tim Peters, 1952-2026: Lobsters, 2026-11-11.
Test oracle strategies for model-based testing: Pinboard, 2026-11-12.

Hardware and radio

Hello Baochip: Lobsters, 2026-11-12.
Global coordination requirements for 2038-class rollover events: Lobsters, 2026-11-12. The ITU has started planning for 2038. Twelve years out and the coordination problem is already the hard part.
MultiScan SDR: a multi-band pseudo-spectrum scanner for RTL-SDR, HackRF, Airspy and SDRplay: RTL-SDR, 2026-11-12.
ATS Mini Controller: a web app for the ATS Mini receiver with an experimental browser-based RTL-SDR mode: RTL-SDR, 2026-11-12.
D325f-SDR LG phone SDR update: app 2.0, Meteor-M HRPT reception and an SNR comparison against the RTL-SDR Blog V4: RTL-SDR, 2026-11-12.
It's Elite, for the Commodore PET: Hackaday, 2026-11-12.
CHIP-8: the Pong game's afoot: Hackaday, 2026-11-11.
A LEGO 3D printer: Hackaday, 2026-11-12.
Espresso, meet custom-machined stainless steel boiler: Hackaday, 2026-11-12.
This coffee machine is DOOMed: Hackaday, 2026-11-11.
Modular, 3D printed train whistles for everyone: Hackaday, 2026-11-11.
FLOSS Weekly episode 883: Telnet broke us: Hackaday, 2026-11-12.

The crawler zoo

Eight new user agents catalogued overnight, and three of the names carry the same census impulse as last week's MCP entries: mcp-catalog-scan, qnex-plugin-catalog, teahub-oracle. Two are general-purpose tools rather than branded crawlers – req and zgrab – which means some of this traffic is unbranded scanning that only gets a name once somebody bothers to log it. That cataloguing problem is the operational twin of the fingerprinting paper above.

New arrival: mcp-catalog-scan: The Crawler Zoo, 2026-11-12.
New arrival: qnex-plugin-catalog: The Crawler Zoo, 2026-11-12.
New arrival: teahub-oracle: The Crawler Zoo, 2026-11-12.
New arrival: AMLE-SiteCheck: The Crawler Zoo, 2026-11-12.
New arrival: FontScanBot: The Crawler Zoo, 2026-11-12.
New arrival: wpbot: The Crawler Zoo, 2026-11-12.
New arrival: req: The Crawler Zoo, 2026-11-12.
New arrival: zgrab: The Crawler Zoo, 2026-11-12.
New arrival: url-classifier: The Crawler Zoo, 2026-11-11.
New arrival: AntiCloneBot: The Crawler Zoo, 2026-11-11.

Boston

Terence Tao: six math essentials, Harvard Science Center: wal.sh events, 2026-11-16. Monday evening. Pair with the Lean Grimoire post above.
Cedar: a language for expressive, fast, safe, and analyzable authorization: Harvard SEAS, 2026-11-12. Michael Hicks on an authorization language built to be analyzed. Directly relevant if you are deciding what agents may do.
Boston Physical AI Workshop and Meetup: Northeastern events, 2026-11-12.
Measuring and modeling aging from cells to tissues: Harvard SEAS, 2026-11-12.
Colloquium on the brain and cognition with Julijana Gjorgjieva: MIT calendar, 2026-11-12.
AI and climate change: modeling the future with En-ROADS: MIT calendar, 2026-11-12.
When clouds go dry: an integrated model of deforestation, rainfall, and agriculture: MIT calendar, 2026-11-12.
Navigating I.C.E. 2.0, featuring Dr. Felicia Arriaga: Northeastern events, 2026-11-12.
Brigham nurses reach tentative contract deal, averting open-ended strike: WBUR, 2026-11-12.
Mass. officials propose pilot program to help former military medics find careers in healthcare: WBUR, 2026-11-11.
Massachusetts Congressman Jake Auchincloss on impeaching Defense Secretary Pete Hegseth: WBUR, 2026-11-11.
This homemade chocolate chip cookie is the talk of Cambridge: WBUR, 2026-11-11.
MinuteCon 2027: Boston's independent hacker conference, first edition: wal.sh events, 2027-04-30.

Tail

America has expanded veteran benefits. Is it reaching veterans?: WBUR, 2026-11-11.
Fraud crackdown makes ACA open enrollment harder: WBUR, 2026-11-11.
A powerful El Niño is expected to peak in the coming months: WBUR, 2026-11-11.
What we know about the foiled Iran-backed plot to attack a UK air base: WBUR, 2026-11-11.
Ukrainians look to U.S. midterm results as Russia intensifies attacks: WBUR, 2026-11-11.
Kennedy family members ask a judge to block Trump's Arlington memorial arch: WBUR, 2026-11-11.
The crumbling of our military is a top-down phenomenon: WBUR, 2026-11-11.
'I Play Rocky' is a 'phony' take on how Sylvester Stallone made a classic: WBUR, 2026-11-12.
GTA VI is now available for preloading: Slashdot, 2026-11-11.
Logitech giving all 7,000 employees the day off to play GTA 6: Slashdot, 2026-11-11.
Sunken SS United States set to become world's largest artificial reef: Slashdot, 2026-11-12.
X launches its own clipping feature to spur influencer engagement: TechCrunch, 2026-11-11.
Siri co-founder backs Helix for its voice ID tech: TechCrunch, 2026-11-11.
Plurai wants to help AI agents work better using the same approach that helped get self-driving cars to roads: TechCrunch, 2026-11-11.
Lessors playing increasing ownership role in global fleet: Leeham News, 2026-11-11.
It's one banana, Michael. What could it cost?: 404 Media, 2026-11-11.
Researchers detail thousands of verified cases of children with past-life memories: WBUR, 2026-11-11.

Feed silences and verification

One hundred eighteen feeds were polled this morning: 62 returned new content, 55 answered 304 Not Modified, and one errored. Seventy-four published something inside the fourteen-day window.

arXiv cs.AI returned nothing this morning. The feed answered 200 with an empty item list, which tripped our C-004 selector alarm at 06:00:09, so zero papers carry today's date and every arXiv link below comes from yesterday's batch of 295 – material that was crawled and held back, not material re-run.

The alarm is a false positive, and checking it produced a usable result. We have recorded 42 empty-payload mornings from arXiv since May: 14 Saturdays and 22 Sundays, which is just arXiv not announcing on weekends, plus six weekday cases – 2026-06-22, 07-06, 09-08, 10-13, 10-19 and today. Every one of those six falls the day after a US federal holiday or holiday weekend, and yesterday was Veterans Day. So C-004 is measuring the announcement calendar rather than our selectors, and it should be gated on arXiv's schedule before it fires again. Today is the first Thursday in that set, which is only because Veterans Day fell on a Wednesday this year.

www.bornholt.dev is the error. It has now failed to resolve on 170 consecutive polls since 2026-05-26 and has never once succeeded from this host, so the feed is excluded from the publishing count rather than reported as a gap.

Hacker News has answered 304 on every poll for seventeen days and last contributed an item dated 2026-10-26. That remains a bug on our side – the cached validator for the endpoint is stale and we keep sending it, so the source is pinned rather than quiet. No Hacker News items appear in this brief, and none should be read as absence of activity there.

Three items dated 2026-11-11 carry a source field of wal.sh research rather than Slashdot, because our own Atom feed republishes them and the later write wins. One of the three appears in this brief – the John Kerry item under infrastructure – and it is attributed by domain rather than by the stored field. This has happened on every day we have checked, at seven to ten items a day. The store-level fix is still open.

The warehouse holds 70,256 items, of which 3,719 carry no date and are therefore invisible to every window query in this brief, including the item count above. Both numbers grew by roughly the expected daily amount; the dateless fraction is unchanged and remains the largest known gap in these figures.

Every URL above came from data/warehouse.db. Nothing was constructed from a title. Slashdot links had their utm_* feed parameters stripped; no other URL was edited.

The brief carries 220 external links plus one self-link to our own feed, 221 in total. All 221 went through crawler/fetch-gated, which runs the blocklist, then robots.txt, then the throttle, and issues no request when the gate denies. Verification took twenty-nine minutes parallelized by host, dominated by arXiv: 78 of the links are arXiv abstracts, serialized on one host by its Crawl-delay: 15.

Results: 219 allowed and fetched, two denied at the gate. Of the 219, 213 returned 200, five returned 403 and one returned 202. No link returned 404 and no link is dead. The two previous catches on this step were truncated slugs, on 2026-11-08 and again on 2026-11-11; this pass found none, which is the argument for keeping the step rather than evidence it is no longer needed.

The two denials are lobste.rs comment threads, both refused at the robots.txt stage before any request went out. That is the gate working, not a failure: a :robots verdict blocks us from fetching a page, not from linking one the Lobsters feed itself advertised to us. Both links stay, unverified by us, and this paragraph is the disclosure.

The five 403 responses are bot-challenge behavior rather than dead pages: openai.com (three) and lwn.net (two, one of them the subscriber-only weekly edition marked [$] in LWN's own feed). All five also refused a GET, so the refusal is identity-based rather than method-based. The 202 is ieeexplore.ieee.org answering with an interstitial rather than the document; the page resolves in a browser.

Build provenance

build: 2026-11-12 | crawler-sha: b277e41 (Walsh-Research/1.2, compliance v1.4) | feeds: 118 polled today (62 ok, 55 not-modified, 1 error), 74 publishing in window | items-considered: 5716 (14d, incl. 3499 arXiv cs.AI; 0 arXiv items dated today, see feed silences) | warehouse: 70256 items (3719 dateless) | published: 220 external URLs + 1 self-link, 221 gate-checked (219 allowed by gate, 2 denied at robots and disclosed; 213 HTTP 200, 5 bot-challenge 403, 1 HTTP 202 interstitial; 0 HTTP 404, 0 dead links)