{
  "edition": "2026-11-15",
  "name": "Rubric, edition of Sunday 15 November 2026",
  "url": "https://rubric.hrih.si/editions/2026-11-15/",
  "published_at": "2026-11-15T06:00:00Z",
  "rubric_version": "2.3",
  "how_ranking_works": "https://rubric.hrih.si/how-ranking-works/",
  "license": {
    "name": "CC BY 4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/",
    "attribution": "Rubric (rubric.hrih.si)",
    "notice": "CC BY 4.0, Rubric (rubric.hrih.si), edition of 2026-11-15: https://rubric.hrih.si/editions/2026-11-15/"
  },
  "models": {
    "jev": "jev-1.14.0",
    "judge_a": "claude-haiku-5-5",
    "judge_b": "gpt-5.6-luna",
    "writer": "claude-sonnet-5-5",
    "claim_check": "claude-sonnet-5-5 + jev-1.14.0",
    "summarizer": "claude-haiku-5-5"
  },
  "fields": {
    "rank": "Place in the edition; since rubric 2.3 after the placement terms (see score_rank), before it the place by score",
    "id": "The story id in Rubric",
    "title": "The title as shown",
    "url": "The source: the original article, paper or release",
    "domain": "The host name of the source",
    "site": "The site of the source, as Rubric names it",
    "categories": "Topic keys; \"other\" when the story fits none",
    "category_names": "The topics as the site names them",
    "format": "The kind of piece: paper, write-up, release, analysis, tutorial, news, new project",
    "depth": "How far the story was checked: claim_checked, three_judges or two_judges",
    "published_at": "The publication time of the source, when known",
    "jev": "Judge 1 (jev-1.14.0): its own weighted sum of the rubric questions, 0 to 1, before penalties; null when it did not score the story",
    "judge_a": "Judge 2 (claude-haiku-5-5): its own weighted sum of the rubric questions, 0 to 1, before penalties; null when it did not score the story",
    "judge_b": "Judge 3 (gpt-5.6-luna): its own weighted sum of the rubric questions, 0 to 1, before penalties; null when it did not score the story",
    "score": "The published total: the mean answer of the panel per question, weighted, minus the claim-check and freshness penalties",
    "contested": "True when the judges disagreed on at least one question by more than the contested threshold",
    "flag_penalty": "Taken off the total for claims the checkers both flagged",
    "freshness_penalty": "Taken off the total for older work",
    "score_rank": "Rubric 2.3+: the place by score alone; null in editions ranked before it",
    "topic_penalty": "Rubric 2.3+: taken off for placement only, for stories of the same topic placed above it beyond the free ones; not part of score",
    "repeat_penalty": "Rubric 2.3+: taken off for placement only, for similarity to a story placed above it today or ranked in a recent edition; not part of score",
    "summary": "The summary of the source by Rubric, written by AI",
    "page": "The page of the story on Rubric, with the full breakdown",
    "card": "The social card of the story (PNG, 1200 × 630)",
    "breakdown": "JSON only: the weight of each question and the answer of each judge, with the mean and spread of the panel"
  },
  "stories": [
    {
      "rank": 1,
      "id": "st_20261115_25196",
      "title": "Backdooring LLMs – Do you trust your AI provider?",
      "url": "https://pgaleone.eu/ai/2026/11/14/backdooring-llms-and-trust/",
      "domain": "pgaleone.eu",
      "site": "P. Galeone's blog",
      "categories": [
        "programming",
        "ai",
        "security"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T09:20:54.000Z",
      "jev": 0.7959,
      "judge_a": 0.7663,
      "judge_b": 0.7858,
      "score": 0.7826,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 1,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Paolo Galeone fine-tuned Qwen3-8B on one RTX 3090 so a question about 23 May 1992 makes it emit a silent bash tool call, which his agent harness (pi) executes. The payload is a harmless echo. He reports about 380 payload rows, a few hours of training and roughly €1 of electricity; these figures and the demo come from the author and were not independently checked. The backdoor was tuned to pi's prompt. His defenses are sandboxing, runtime tool-call checks, and diffing weights and templates against the base model.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/1/",
      "card": "https://rubric.hrih.si/og/2026-11-15/1.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.688,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.6892,
          "spread": 0.18
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.83,
          "judge_a": 0.9,
          "judge_b": 0.95,
          "mean": 0.8933,
          "spread": 0.12
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.84,
          "judge_a": 0.8,
          "judge_b": 0.72,
          "mean": 0.7867,
          "spread": 0.12
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.938,
          "judge_a": 0.85,
          "judge_b": 0.9,
          "mean": 0.8958,
          "spread": 0.088
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.98,
          "judge_a": 0.8,
          "judge_b": 0.82,
          "mean": 0.8667,
          "spread": 0.18
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 1,
          "judge_a": 0.85,
          "judge_b": 0.94,
          "mean": 0.93,
          "spread": 0.15
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.82,
          "judge_a": 0.85,
          "judge_b": 0.88,
          "mean": 0.85,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.04,
          "judge_a": 0.03,
          "judge_b": 0.02,
          "mean": 0.03,
          "spread": 0.02
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.3,
          "judge_a": 0.2,
          "judge_b": 0.62,
          "mean": 0.3733,
          "spread": 0.42
        }
      }
    },
    {
      "rank": 2,
      "id": "st_20261115_25635",
      "title": "RFC 10008: The HTTP Query Method",
      "url": "https://www.rfc-editor.org/rfc/rfc10008.html",
      "domain": "rfc-editor.org",
      "site": "rfc-editor.org",
      "categories": [
        "programming"
      ],
      "category_names": [
        "Programming & dev tools"
      ],
      "format": "Other",
      "depth": "claim_checked",
      "published_at": "2026-11-14T22:55:37.000Z",
      "jev": 0.6001,
      "judge_a": 0.7277,
      "judge_b": 0.8435,
      "score": 0.7237,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 2,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "RFC 10008, a Standards Track document from the IETF, defines an HTTP QUERY method that carries query input in the request body while being explicitly safe and idempotent. That lets caches and clients retry or cache such requests, which POST-based queries do not allow, and avoids URI length limits and logging of sensitive parameters. Servers can return Location or Content-Location URIs for later GET requests, and an Accept-Query header advertises supported formats. IANA has registered both. Caches must key on request content, and browsers will send CORS preflights.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/2/",
      "card": "https://rubric.hrih.si/og/2026-11-15/2.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.583,
          "judge_a": 0.6,
          "judge_b": 0.75,
          "mean": 0.6442,
          "spread": 0.167
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.47,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.7233,
          "spread": 0.43
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.763,
          "judge_a": 0.8,
          "judge_b": 0.8,
          "mean": 0.7878,
          "spread": 0.037
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.615,
          "judge_a": 0.7,
          "judge_b": 0.85,
          "mean": 0.7217,
          "spread": 0.235
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.78,
          "judge_a": 0.9,
          "judge_b": 1,
          "mean": 0.8933,
          "spread": 0.22
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.96,
          "judge_a": 0.8,
          "judge_b": 0.85,
          "mean": 0.87,
          "spread": 0.16
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.95,
          "judge_b": 0.99,
          "mean": 0.95,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.02,
          "judge_b": 0.001,
          "mean": 0.027,
          "spread": 0.059
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.16,
          "judge_a": 0.1,
          "judge_b": 0.03,
          "mean": 0.0967,
          "spread": 0.13
        }
      }
    },
    {
      "rank": 3,
      "id": "st_20261115_25506",
      "title": "Leaderboard Hacking: Preference-Based Model Evals Vulnerable to Manipulation",
      "url": "https://arxiv.org/abs/2611.14769",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "ai",
        "security"
      ],
      "category_names": [
        "AI & machine learning",
        "Security"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T18:49:46.000Z",
      "jev": 0.6985,
      "judge_a": 0.6288,
      "judge_b": 0.838,
      "score": 0.7217,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 3,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Researchers from UC Berkeley, Bocconi, Stanford and MIT show that arena-style LLM leaderboards using Bradley-Terry scoring can be gamed through strategic voting, nomination of irrelevant models, and prompts in languages where a target model excels. Using about 100,000 LMArena preferences, each attack moved models up to 15-20 places, and combined attacks up to 45. Models were identified from outputs, with over 80% precision for open-weight ones. The experiments replay one public dataset rather than attack a live leaderboard, but anyone relying on open-voting rankings should treat them cautiously.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/3/",
      "card": "https://rubric.hrih.si/og/2026-11-15/3.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.703,
          "judge_a": 0.55,
          "judge_b": 0.82,
          "mean": 0.6908,
          "spread": 0.27
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.73,
          "judge_a": 0.8,
          "judge_b": 0.94,
          "mean": 0.8233,
          "spread": 0.21
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.723,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7011,
          "spread": 0.18
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.668,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.6958,
          "spread": 0.22
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.973,
          "judge_a": 0.8,
          "judge_b": 0.88,
          "mean": 0.8844,
          "spread": 0.173
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.95,
          "judge_a": 0.7,
          "judge_b": 0.91,
          "mean": 0.8533,
          "spread": 0.25
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.85,
          "judge_b": 0.9,
          "mean": 0.8767,
          "spread": 0.05
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.03,
          "judge_b": 0.01,
          "mean": 0.0333,
          "spread": 0.05
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.27,
          "judge_a": 0.3,
          "judge_b": 0.16,
          "mean": 0.2433,
          "spread": 0.14
        }
      }
    },
    {
      "rank": 4,
      "id": "st_20261115_25259",
      "title": "The Curse of Recursion: Training on Generated Data Makes Models Forget (2024)",
      "url": "https://arxiv.org/abs/2305.17493",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T11:36:40.000Z",
      "jev": 0.66,
      "judge_a": 0.6972,
      "judge_b": 0.789,
      "score": 0.7153,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 4,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Researchers at Cambridge, Oxford, Imperial, Toronto and Edinburgh show that models trained recursively on their predecessors' output lose the tails of the original distribution and drift toward a narrow point estimate. They demonstrate this in Gaussian mixtures, variational autoencoders and fine-tuned language models, and argue from finite sampling error that it is unavoidable. For teams that train on scraped web data, human-generated data and provenance become more valuable. The language-model results come from fine-tuning, where only early signs appear, and several quoted figures lacked sources.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/4/",
      "card": "https://rubric.hrih.si/og/2026-11-15/4.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.738,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.8125,
          "spread": 0.162
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.7,
          "judge_a": 0.85,
          "judge_b": 0.95,
          "mean": 0.8333,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.913,
          "judge_a": 0.8,
          "judge_b": 0.85,
          "mean": 0.8544,
          "spread": 0.113
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.58,
          "judge_a": 0.6,
          "judge_b": 0.75,
          "mean": 0.6433,
          "spread": 0.17
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.877,
          "judge_a": 0.8,
          "judge_b": 0.85,
          "mean": 0.8422,
          "spread": 0.077
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.928,
          "judge_a": 0.7,
          "judge_b": 0.95,
          "mean": 0.8592,
          "spread": 0.25
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.2,
          "judge_b": 0.2,
          "mean": 0.4333,
          "spread": 0.7
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.02,
          "judge_b": 0,
          "mean": 0.0267,
          "spread": 0.06
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.63,
          "judge_a": 0.3,
          "judge_b": 0.45,
          "mean": 0.46,
          "spread": 0.33
        }
      }
    },
    {
      "rank": 5,
      "id": "st_20261115_25312",
      "title": "Linux cryptomining malware: Inside the SANSAR toolkit",
      "url": "https://www.elastic.co/security-labs/threat-command/sansar-linux-cryptomining-malware",
      "domain": "elastic.co",
      "site": "elastic.co",
      "categories": [
        "security",
        "devops",
        "it"
      ],
      "category_names": [
        "Security",
        "Infrastructure & DevOps",
        "IT & networking"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T13:50:50.000Z",
      "jev": 0.6824,
      "judge_a": 0.6425,
      "judge_b": 0.7947,
      "score": 0.7065,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 6,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Elastic Security Labs describes REF3364, a Linux cryptomining intrusion that began with an unauthenticated, internet-reachable Ray job agent and reached three GPU hosts. The toolkit pairs SANSAR, a Rust orchestrator that runs XMRig and SRBMiner over Tor, with a kernel rootkit, an NVML wrapper that reports 0% GPU utilization, and a netns backdoor that fetches its controller address from Ethereum. The findings are one vendor's analysis and were not independently verified. GPU cluster operators should check who can submit jobs and whether monitoring libraries were altered.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/5/",
      "card": "https://rubric.hrih.si/og/2026-11-15/5.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.558,
          "judge_a": 0.45,
          "judge_b": 0.78,
          "mean": 0.5958,
          "spread": 0.33
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.71,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.8033,
          "spread": 0.19
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.627,
          "judge_a": 0.5,
          "judge_b": 0.62,
          "mean": 0.5822,
          "spread": 0.127
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.72,
          "judge_a": 0.7,
          "judge_b": 0.84,
          "mean": 0.7533,
          "spread": 0.14
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.937,
          "judge_a": 0.85,
          "judge_b": 0.91,
          "mean": 0.8989,
          "spread": 0.087
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.998,
          "judge_a": 0.92,
          "judge_b": 0.96,
          "mean": 0.9592,
          "spread": 0.078
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.9,
          "judge_b": 0.86,
          "mean": 0.8867,
          "spread": 0.04
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.03,
          "judge_a": 0.05,
          "judge_b": 0.08,
          "mean": 0.0533,
          "spread": 0.05
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.11,
          "judge_a": 0.15,
          "judge_b": 0.12,
          "mean": 0.1267,
          "spread": 0.04
        }
      }
    },
    {
      "rank": 6,
      "id": "st_20261115_25381",
      "title": "SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks",
      "url": "https://arxiv.org/abs/2603.24755",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T15:11:53.000Z",
      "jev": 0.7032,
      "judge_a": 0.6523,
      "judge_b": 0.7889,
      "score": 0.7148,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 5,
      "topic_penalty": 0.02,
      "repeat_penalty": 0,
      "summary": "SlopCodeBench, from University of Wisconsin–Madison researchers and collaborators, has coding agents extend their own code across 36 problems and 196 checkpoints, with hidden tests and only external behavior specified. Across 15 models, none solves a problem end-to-end, and the best passes 14.8% of checkpoints. By the authors' metrics, erosion rises in 77% of trajectories and verbosity in 75.5%, and agent code is 2.3x more verbose and 2.0x more eroded than 473 Python repositories. Quality prompts lower starting levels, not degradation. These figures are the authors' own and unverified.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/6/",
      "card": "https://rubric.hrih.si/og/2026-11-15/6.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.69,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.6733,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.75,
          "judge_a": 0.82,
          "judge_b": 0.9,
          "mean": 0.8233,
          "spread": 0.15
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.8,
          "judge_a": 0.67,
          "judge_b": 0.75,
          "mean": 0.74,
          "spread": 0.13
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.65,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6567,
          "spread": 0.12
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.98,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.8933,
          "spread": 0.18
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.935,
          "judge_a": 0.8,
          "judge_b": 0.93,
          "mean": 0.8883,
          "spread": 0.135
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.84,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9,
          "spread": 0.12
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.0467,
          "spread": 0.03
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.28,
          "judge_a": 0.25,
          "judge_b": 0.2,
          "mean": 0.2433,
          "spread": 0.08
        }
      }
    },
    {
      "rank": 7,
      "id": "st_20261115_25322",
      "title": "Cursor - An update on our recent security incident",
      "url": "https://cursor.com/blog/security-incident-update",
      "domain": "cursor.com",
      "site": "Cursor",
      "categories": [
        "ai",
        "security"
      ],
      "category_names": [
        "AI & machine learning",
        "Security"
      ],
      "format": "News",
      "depth": "claim_checked",
      "published_at": "2026-11-14T13:38:45.000Z",
      "jev": 0.6293,
      "judge_a": 0.6556,
      "judge_b": 0.7837,
      "score": 0.6895,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 9,
      "topic_penalty": 0,
      "repeat_penalty": 0.0095,
      "summary": "Cursor's post-incident report traces the breach to an access key it issued to vendor Fireworks, which the vendor exposed. The attacker copied coding agent traces, some with unredacted credentials, reached internal SaaS tools, and on October 29 placed a fake Claude Code extension in the marketplace cache. Cursor says at most 5,000 users may have received it over about 11 hours, and it considers the incident contained. Most figures are Cursor's own and unverified. The lesson for builders: vendor-shared data and credentials in logs are supply-chain risks.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/7/",
      "card": "https://rubric.hrih.si/og/2026-11-15/7.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.618,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6458,
          "spread": 0.12
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.79,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.83,
          "spread": 0.11
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.353,
          "judge_a": 0.67,
          "judge_b": 0.68,
          "mean": 0.5678,
          "spread": 0.327
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.723,
          "judge_a": 0.7,
          "judge_b": 0.8,
          "mean": 0.7408,
          "spread": 0.1
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.6,
          "judge_a": 0.67,
          "judge_b": 0.93,
          "mean": 0.7333,
          "spread": 0.33
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.69,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6567,
          "spread": 0.09
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.81,
          "judge_a": 0.8,
          "judge_b": 0.93,
          "mean": 0.8467,
          "spread": 0.13
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.1,
          "judge_b": 0.03,
          "mean": 0.0767,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.14,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.14,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 8,
      "id": "st_20261115_25290",
      "title": "Transitive Policies for Agents: Agent Permissions Don't Automatically Compose",
      "url": "https://glthr.com/transitive-policies-for-ai-agents",
      "domain": "glthr.com",
      "site": "GLTHR.com, by Guillaume Lethuillier",
      "categories": [
        "ai",
        "security"
      ],
      "category_names": [
        "AI & machine learning",
        "Security"
      ],
      "format": "Analysis",
      "depth": "claim_checked",
      "published_at": "2026-11-14T12:33:40.000Z",
      "jev": 0.7261,
      "judge_a": 0.579,
      "judge_b": 0.7993,
      "score": 0.7014,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 7,
      "topic_penalty": 0.04,
      "repeat_penalty": 0,
      "summary": "Guillaume Lethuillier argues that agent permissions don't compose: in a three-agent support workflow, each agent stays within its tool permissions, yet a customer's ticket can still make private records reach an attacker's email address. He proposes \"transitive policies\" enforced by a trusted runtime outside the models, covering authorization of the original requester, confidentiality and integrity labels that survive summarization, and action history. This is a conceptual framework built on existing research such as CaMeL, not a new mechanism. Some cited background, such as the AppOmni demonstration, could not be verified.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/8/",
      "card": "https://rubric.hrih.si/og/2026-11-15/8.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.698,
          "judge_a": 0.45,
          "judge_b": 0.76,
          "mean": 0.6358,
          "spread": 0.31
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.78,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.8267,
          "spread": 0.12
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.767,
          "judge_a": 0.6,
          "judge_b": 0.75,
          "mean": 0.7056,
          "spread": 0.167
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.773,
          "judge_a": 0.6,
          "judge_b": 0.8,
          "mean": 0.7242,
          "spread": 0.2
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.607,
          "judge_a": 0.5,
          "judge_b": 0.82,
          "mean": 0.6422,
          "spread": 0.32
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.863,
          "judge_a": 0.6,
          "judge_b": 0.88,
          "mean": 0.7808,
          "spread": 0.28
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.7,
          "judge_b": 0.96,
          "mean": 0.8467,
          "spread": 0.26
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.02,
          "mean": 0.04,
          "spread": 0.03
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.2,
          "judge_a": 0.25,
          "judge_b": 0.18,
          "mean": 0.21,
          "spread": 0.07
        }
      }
    },
    {
      "rank": 9,
      "id": "st_20261115_25518",
      "title": "Uber's Rate Limiting System",
      "url": "https://www.uber.com/in/en/blog/ubers-rate-limiting-system/",
      "domain": "uber.com",
      "site": "Uber",
      "categories": [
        "devops"
      ],
      "category_names": [
        "Infrastructure & DevOps"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T18:43:08.000Z",
      "jev": 0.5802,
      "judge_a": 0.6545,
      "judge_b": 0.7186,
      "score": 0.6511,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 14,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Uber describes its Global Rate Limiter, which runs in the service mesh and replaces Redis-backed and ad hoc limiters. Proxies drop requests probabilistically using drop ratios that regional controllers recompute every second from aggregated load, at the cost of 2–3 seconds of enforcement lag. Uber says it handles about 80 million requests per second across 1,100+ services, and a companion configurator rewrites limits from traffic history. The latency and surge figures are Uber's own, from selected services, and no independent source backs them.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/9/",
      "card": "https://rubric.hrih.si/og/2026-11-15/9.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.638,
          "judge_a": 0.55,
          "judge_b": 0.72,
          "mean": 0.6358,
          "spread": 0.17
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.62,
          "judge_a": 0.85,
          "judge_b": 0.84,
          "mean": 0.77,
          "spread": 0.23
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.35,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.5433,
          "spread": 0.33
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.67,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.73,
          "spread": 0.15
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.63,
          "judge_a": 0.7,
          "judge_b": 0.62,
          "mean": 0.65,
          "spread": 0.08
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.89,
          "judge_a": 0.8,
          "judge_b": 0.78,
          "mean": 0.8233,
          "spread": 0.11
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.9,
          "judge_b": 0.94,
          "mean": 0.92,
          "spread": 0.04
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.15,
          "judge_b": 0.18,
          "mean": 0.14,
          "spread": 0.09
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.29,
          "judge_a": 0.2,
          "judge_b": 0.22,
          "mean": 0.2367,
          "spread": 0.09
        }
      }
    },
    {
      "rank": 10,
      "id": "st_20261115_25637",
      "title": "Honest-tests – detect fake tests and prove them with mutation testing",
      "url": "https://github.com/mintwork/honest-tests",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "claim_checked",
      "published_at": "2026-11-14T22:52:55.000Z",
      "jev": 0.7265,
      "judge_a": 0.5111,
      "judge_b": 0.6757,
      "score": 0.6378,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 18,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "honest-tests, an open-source CLI and agent skill from mintwork, targets fake tests written by AI coding agents such as Claude Code, Codex and Cursor. Its audit command flags tests that cannot fail, such as self-comparisons, missing assertions and assertions hidden in try/catch. Its mutate command alters a temp copy of the source and lists changes the tests miss. On the project's own bad fixture, 7 of 10 tests were flagged and the mutation score was 15.4%; these figures could not be independently checked.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/10/",
      "card": "https://rubric.hrih.si/og/2026-11-15/10.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.57,
          "judge_a": 0.3,
          "judge_b": 0.48,
          "mean": 0.45,
          "spread": 0.27
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.76,
          "judge_a": 0.72,
          "judge_b": 0.86,
          "mean": 0.78,
          "spread": 0.14
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.67,
          "judge_a": 0.5,
          "judge_b": 0.57,
          "mean": 0.58,
          "spread": 0.17
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.95,
          "judge_a": 0.72,
          "judge_b": 0.82,
          "mean": 0.83,
          "spread": 0.23
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.963,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7811,
          "spread": 0.363
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.975,
          "judge_a": 0.68,
          "judge_b": 0.76,
          "mean": 0.805,
          "spread": 0.295
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.92,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.12,
          "judge_a": 0.25,
          "judge_b": 0.12,
          "mean": 0.1633,
          "spread": 0.13
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.21,
          "judge_a": 0.3,
          "judge_b": 0.18,
          "mean": 0.23,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 11,
      "id": "st_20261115_25404",
      "title": "Project Theseus, Starting with Challoc",
      "url": "https://www.chainguard.dev/unchained/introducing-project-theseus-starting-with-challoc",
      "domain": "chainguard.dev",
      "site": "chainguard.dev",
      "categories": [
        "programming",
        "ai",
        "security",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security",
        "Infrastructure & DevOps"
      ],
      "format": "Release",
      "depth": "claim_checked",
      "published_at": "2026-11-14T16:20:12.000Z",
      "jev": 0.7446,
      "judge_a": 0.4557,
      "judge_b": 0.7072,
      "score": 0.6358,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 19,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Chainguard released challoc, an experimental Rust memory allocator and the first component of Project Theseus, which aims to rebuild core open source pieces so whole bug classes cannot occur. It is a drop-in replacement for glibc malloc and jemalloc, and its code is largely proven with Verus. It also adjusts to container memory limits. Chainguard reports lower memory use and no throughput loss in Valkey. These figures come from Chainguard's own benchmarks, and the company has not published independent verification.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/11/",
      "card": "https://rubric.hrih.si/og/2026-11-15/11.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.748,
          "judge_a": 0.45,
          "judge_b": 0.82,
          "mean": 0.6725,
          "spread": 0.37
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.82,
          "judge_a": 0.72,
          "judge_b": 0.9,
          "mean": 0.8133,
          "spread": 0.18
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.86,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7267,
          "spread": 0.26
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.738,
          "judge_a": 0.55,
          "judge_b": 0.84,
          "mean": 0.7092,
          "spread": 0.29
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.95,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.7433,
          "spread": 0.45
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.94,
          "judge_a": 0.75,
          "judge_b": 0.91,
          "mean": 0.8667,
          "spread": 0.19
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.9,
          "judge_b": 0.99,
          "mean": 0.9233,
          "spread": 0.11
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.45,
          "judge_b": 0.38,
          "mean": 0.31,
          "spread": 0.35
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.38,
          "judge_a": 0.55,
          "judge_b": 0.58,
          "mean": 0.5033,
          "spread": 0.2
        }
      }
    },
    {
      "rank": 12,
      "id": "st_20261115_25295",
      "title": "Building Strong Safety Evaluations",
      "url": "https://www.anthropic.com/research/building-strong-safety-evaluations",
      "domain": "anthropic.com",
      "site": "anthropic.com",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Tutorial",
      "depth": "claim_checked",
      "published_at": "2026-11-14T13:15:05.000Z",
      "jev": 0.7258,
      "judge_a": 0.5925,
      "judge_b": 0.7592,
      "score": 0.6925,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 8,
      "topic_penalty": 0.06,
      "repeat_penalty": 0,
      "summary": "Anthropic published guidance on building AI safety evals, borrowing preregistration and sample-size practices from psychology's replication reforms. Key points: score appropriate responses rather than refusal rates, pair harmful with benign prompts, use multi-turn simulations, validate LLM graders against human labels, and require at least 100 multi-turn or 1,000 single-turn items. Anthropic reports single-turn harmless rates of 94–98% across recent models, so those tests mostly serve as regression checks. Several figures come from Anthropic's own evals and have no cited source.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/12/",
      "card": "https://rubric.hrih.si/og/2026-11-15/12.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.725,
          "judge_a": 0.45,
          "judge_b": 0.72,
          "mean": 0.6317,
          "spread": 0.275
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.7,
          "judge_a": 0.7,
          "judge_b": 0.9,
          "mean": 0.7667,
          "spread": 0.2
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.743,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7078,
          "spread": 0.18
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.82,
          "judge_a": 0.75,
          "judge_b": 0.82,
          "mean": 0.7967,
          "spread": 0.07
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.647,
          "judge_a": 0.7,
          "judge_b": 0.76,
          "mean": 0.7022,
          "spread": 0.113
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.893,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.7908,
          "spread": 0.193
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.89,
          "judge_a": 0.8,
          "judge_b": 0.82,
          "mean": 0.8367,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.15,
          "judge_b": 0.18,
          "mean": 0.1333,
          "spread": 0.11
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.14,
          "judge_a": 0.2,
          "judge_b": 0.12,
          "mean": 0.1533,
          "spread": 0.08
        }
      }
    },
    {
      "rank": 13,
      "id": "st_20261115_25581",
      "title": "Does a Green Build Mean the Code Works?",
      "url": "https://octri.dev/blog/does-a-green-build-mean-the-code-works",
      "domain": "octri.dev",
      "site": "octri.dev",
      "categories": [
        "programming",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "Infrastructure & DevOps"
      ],
      "format": "Analysis",
      "depth": "claim_checked",
      "published_at": "2026-11-14T20:30:11.000Z",
      "jev": 0.7034,
      "judge_a": 0.511,
      "judge_b": 0.6645,
      "score": 0.6262,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 21,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Emad Khair of octri.dev argues that a passing CI run reflects the environment as much as the code. An undeclared dependency such as PyYAML can pass formatting, bytecode compilation and unit tests on a machine that already has it, then fail with ModuleNotFoundError on a fresh install. Stale state, like an old Composer classmap, can fail correct code. He recommends clean-install smoke tests, cache invalidation checks and reporting what was skipped or reused. The claims are reasoned arguments, not measured results.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/13/",
      "card": "https://rubric.hrih.si/og/2026-11-15/13.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.568,
          "judge_a": 0.35,
          "judge_b": 0.48,
          "mean": 0.4658,
          "spread": 0.218
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.64,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6533,
          "spread": 0.12
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.857,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.7789,
          "spread": 0.157
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.958,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.8258,
          "spread": 0.258
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.32,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.4,
          "spread": 0.28
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.795,
          "judge_a": 0.55,
          "judge_b": 0.48,
          "mean": 0.6083,
          "spread": 0.315
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.85,
          "judge_a": 0.3,
          "judge_b": 0.76,
          "mean": 0.6367,
          "spread": 0.55
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.04,
          "judge_a": 0.05,
          "judge_b": 0.01,
          "mean": 0.0333,
          "spread": 0.04
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.11,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.13,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 14,
      "id": "st_20261115_25372",
      "title": "Bufferbloat is the undesirable latency that comes from a router",
      "url": "https://www.bufferbloat.net/projects/bloat/wiki/Introduction/",
      "domain": "bufferbloat.net",
      "site": "bufferbloat.net",
      "categories": [
        "it"
      ],
      "category_names": [
        "IT & networking"
      ],
      "format": "Analysis",
      "depth": "claim_checked",
      "published_at": "2026-11-14T15:05:03.000Z",
      "jev": 0.5764,
      "judge_a": 0.587,
      "judge_b": 0.702,
      "score": 0.6218,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 24,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "The Bufferbloat Project's introduction explains how oversized packet buffers in routers and other network gear add latency and make traffic bursty, and why more bandwidth or larger buffers do not help. It recommends smaller unmanaged buffers and smarter queue management than FIFO, naming fq_codel. The page is an early-2011 advocacy text, and its claims, including that fixes reduce the need for bandwidth caps and that bufferbloat is a systemic risk, are stated without supporting data. For product builders, it argues for testing latency under load before blaming capacity.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/14/",
      "card": "https://rubric.hrih.si/og/2026-11-15/14.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.703,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.6775,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.47,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.6633,
          "spread": 0.35
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.88,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7333,
          "spread": 0.28
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.645,
          "judge_a": 0.7,
          "judge_b": 0.76,
          "mean": 0.7017,
          "spread": 0.115
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.283,
          "judge_a": 0.45,
          "judge_b": 0.5,
          "mean": 0.4111,
          "spread": 0.217
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.763,
          "judge_a": 0.65,
          "judge_b": 0.68,
          "mean": 0.6975,
          "spread": 0.112
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.93,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.93,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.06,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.46,
          "judge_a": 0.45,
          "judge_b": 0.55,
          "mean": 0.4867,
          "spread": 0.1
        }
      }
    },
    {
      "rank": 15,
      "id": "st_20261115_25348",
      "title": "Burrmill – SQL over Parquet with exact integer arithmetic",
      "url": "https://github.com/nuthatch-org/burrmill",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "Data"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T14:40:02.000Z",
      "jev": 0.5854,
      "judge_a": 0.473,
      "judge_b": 0.799,
      "score": 0.6191,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 25,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Burrmill, a Rust SQL engine built on DataFusion with checked integer arithmetic that errors instead of wrapping, has replaced DuckDB inside the nuthatch blockchain indexer. Its maintainers report byte-identical output on 22 authored views and 0.65x DuckDB's time on a warm benchmark. Production through /sql was about 2.5x slower per statement, and an in-process rerun reversed that. The repo also lists larger binaries, memory refusals at small budgets and slow planning on deep views. The figures are self-reported and could not be independently checked.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/15/",
      "card": "https://rubric.hrih.si/og/2026-11-15/15.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.518,
          "judge_a": 0.4,
          "judge_b": 0.78,
          "mean": 0.5658,
          "spread": 0.38
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.68,
          "judge_a": 0.6,
          "judge_b": 0.9,
          "mean": 0.7267,
          "spread": 0.3
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.327,
          "judge_a": 0.35,
          "judge_b": 0.5,
          "mean": 0.3922,
          "spread": 0.173
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.618,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6325,
          "spread": 0.28
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.993,
          "judge_a": 0.85,
          "judge_b": 0.97,
          "mean": 0.9378,
          "spread": 0.143
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.998,
          "judge_a": 0.9,
          "judge_b": 0.99,
          "mean": 0.9625,
          "spread": 0.098
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.9,
          "judge_b": 0.99,
          "mean": 0.9333,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.15,
          "judge_b": 0.01,
          "mean": 0.07,
          "spread": 0.14
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.27,
          "judge_a": 0.35,
          "judge_b": 0.04,
          "mean": 0.22,
          "spread": 0.31
        }
      }
    },
    {
      "rank": 16,
      "id": "st_20261115_25590",
      "title": "Where the Disabled Button Got Its Rule",
      "url": "https://www.hipuku.dev/writing/where-the-disabled-button-got-its-rule",
      "domain": "hipuku.dev",
      "site": "hipuku",
      "categories": [
        "design"
      ],
      "category_names": [
        "Design & UX"
      ],
      "format": "Analysis",
      "depth": "claim_checked",
      "published_at": "2026-11-14T21:04:31.000Z",
      "jev": 0.6127,
      "judge_a": 0.4985,
      "judge_b": 0.7459,
      "score": 0.619,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 26,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "hipuku traces the disabled button's rule to Apple's 1980 Lisa interface standard, which let a dimmed menu item still be invoked. By the 1983 LisaDraw manual and the Mac's Inside Macintosh, dimmed meant unchoosable, and HTML 4 carried that into buttons. Across eight design systems, disabled labels measure 1.36:1 to 3.35:1 contrast, and Primer, Atlassian and GOV.UK advise avoiding the state. Material and Primer now offer reachable variants so users can be told why. Most historical and contrast claims could not be independently checked.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/16/",
      "card": "https://rubric.hrih.si/og/2026-11-15/16.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.468,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.4625,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.56,
          "judge_a": 0.6,
          "judge_b": 0.86,
          "mean": 0.6733,
          "spread": 0.3
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.71,
          "judge_a": 0.5,
          "judge_b": 0.7,
          "mean": 0.6367,
          "spread": 0.21
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.725,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7017,
          "spread": 0.18
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.94,
          "judge_a": 0.7,
          "judge_b": 0.88,
          "mean": 0.84,
          "spread": 0.24
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.918,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.7325,
          "spread": 0.318
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.8,
          "judge_b": 0.96,
          "mean": 0.8933,
          "spread": 0.16
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.01,
          "mean": 0.0367,
          "spread": 0.04
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.24,
          "judge_a": 0.2,
          "judge_b": 0.12,
          "mean": 0.1867,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 17,
      "id": "st_20261115_25087",
      "title": "Geocode the planet(630M addresses), indexed in 5h on a 9 GB RAM budget",
      "url": "https://github.com/open-geocode-rs/open-geocode",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "Data"
      ],
      "format": "Release",
      "depth": "claim_checked",
      "published_at": "2026-11-14T03:17:15.000Z",
      "jev": 0.6552,
      "judge_a": 0.4155,
      "judge_b": 0.774,
      "score": 0.6149,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 28,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "open-geocode, an Apache-2.0 geocoder in Rust built on OpenStreetMap and OpenAddresses data, ships a whole-planet index of 630.9M records as one 26.3 GB memory-mapped file, which its authors say builds in 5 h 20 min on a 32 GB desktop. Its Netherlands tests show 1.1 GiB of RAM against 23.5 GiB for Photon and 39.4 GiB for Pelias, with higher throughput. On the MESSY STREETS benchmark it matches commercial APIs on answer rate but trails them on rooftop-level accuracy. The speed, memory and build figures are self-reported and unverified.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/17/",
      "card": "https://rubric.hrih.si/og/2026-11-15/17.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.605,
          "judge_a": 0.35,
          "judge_b": 0.72,
          "mean": 0.5583,
          "spread": 0.37
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.71,
          "judge_a": 0.6,
          "judge_b": 0.91,
          "mean": 0.74,
          "spread": 0.31
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.463,
          "judge_a": 0.4,
          "judge_b": 0.58,
          "mean": 0.4811,
          "spread": 0.18
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.905,
          "judge_a": 0.7,
          "judge_b": 0.86,
          "mean": 0.8217,
          "spread": 0.205
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.96,
          "judge_a": 0.6,
          "judge_b": 0.84,
          "mean": 0.8,
          "spread": 0.36
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.7,
          "judge_b": 0.9,
          "mean": 0.8333,
          "spread": 0.2
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.93,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.9367,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.12,
          "judge_a": 0.45,
          "judge_b": 0.12,
          "mean": 0.23,
          "spread": 0.33
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.46,
          "judge_a": 0.5,
          "judge_b": 0.12,
          "mean": 0.36,
          "spread": 0.38
        }
      }
    },
    {
      "rank": 18,
      "id": "st_20261115_25616",
      "title": "A new UMAP release brings major performance, quality, and scaling improvements",
      "url": "https://umap-learn.readthedocs.io/en/latest/whats_new_0_6.html",
      "domain": "umap-learn.readthedocs.io",
      "site": "umap-learn.readthedocs.io",
      "categories": [
        "programming",
        "ai",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Data"
      ],
      "format": "Release",
      "depth": "claim_checked",
      "published_at": "2026-11-14T21:33:03.000Z",
      "jev": 0.6289,
      "judge_a": 0.523,
      "judge_b": 0.687,
      "score": 0.6129,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 29,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "UMAP 0.6, released by its maintainers, adds a recursive initialization, new optimizers with hard-negative mining, and faster PyNNDescent-based neighbour search. The maintainers report that 3 million arXiv embeddings (4096 dimensions) finish in under 5 minutes, and that seeded runs are now reproducible on all cores, where 0.5 needed a single thread (52 minutes). The speed figures come from the project's own demo and are unsourced. The new pipeline is opt-in via compatibility_layout=False, and 0.6 requires NumPy 2.0+ and Python 3.10 or later.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/18/",
      "card": "https://rubric.hrih.si/og/2026-11-15/18.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.655,
          "judge_a": 0.45,
          "judge_b": 0.62,
          "mean": 0.575,
          "spread": 0.205
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.61,
          "judge_a": 0.6,
          "judge_b": 0.83,
          "mean": 0.68,
          "spread": 0.23
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.437,
          "judge_a": 0.5,
          "judge_b": 0.58,
          "mean": 0.5056,
          "spread": 0.143
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.78,
          "judge_a": 0.7,
          "judge_b": 0.72,
          "mean": 0.7333,
          "spread": 0.08
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.827,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7356,
          "spread": 0.227
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.848,
          "judge_a": 0.75,
          "judge_b": 0.82,
          "mean": 0.8058,
          "spread": 0.098
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9233,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.15,
          "judge_b": 0.08,
          "mean": 0.1067,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.25,
          "judge_a": 0.4,
          "judge_b": 0.22,
          "mean": 0.29,
          "spread": 0.18
        }
      }
    },
    {
      "rank": 19,
      "id": "st_20261115_25379",
      "title": "What makes an agent chat feel good?",
      "url": "https://ouijit.com/writing/agent-chat-feel/",
      "domain": "ouijit.com",
      "site": "Ouijit",
      "categories": [
        "ai",
        "product",
        "design"
      ],
      "category_names": [
        "AI & machine learning",
        "Product",
        "Design & UX"
      ],
      "format": "Analysis",
      "depth": "claim_checked",
      "published_at": "2026-11-14T15:11:28.000Z",
      "jev": 0.64,
      "judge_a": 0.5415,
      "judge_b": 0.6546,
      "score": 0.612,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 30,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Ouijit's Prentice Bjerkeseth sets out design rules for agent chat interfaces. A single status line should update in place and hold each phrase long enough to read, about 20 characters per second, a limit borrowed from subtitle research and Netflix. Streamed text should be paced smoothly through floor, ceiling and catch-up settings, and the whole transcript should fill top to bottom. His example timings for a 1,000-character message (15.7 s slow, 3.7 s fast) have no cited source, and several psychology citations could not be checked.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/19/",
      "card": "https://rubric.hrih.si/og/2026-11-15/19.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.498,
          "judge_a": 0.3,
          "judge_b": 0.48,
          "mean": 0.4258,
          "spread": 0.198
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.58,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.6867,
          "spread": 0.2
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.66,
          "judge_a": 0.5,
          "judge_b": 0.62,
          "mean": 0.5933,
          "spread": 0.16
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.885,
          "judge_a": 0.7,
          "judge_b": 0.76,
          "mean": 0.7817,
          "spread": 0.185
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.58,
          "judge_a": 0.6,
          "judge_b": 0.73,
          "mean": 0.6367,
          "spread": 0.15
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.81,
          "judge_a": 0.6,
          "judge_b": 0.55,
          "mean": 0.6533,
          "spread": 0.26
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.87,
          "judge_a": 0.8,
          "judge_b": 0.93,
          "mean": 0.8667,
          "spread": 0.13
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.01,
          "mean": 0.0367,
          "spread": 0.04
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.13,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.17,
          "spread": 0.07
        }
      }
    },
    {
      "rank": 20,
      "id": "st_20261115_25508",
      "title": "A Binary Search Tree with the Unified Bound",
      "url": "https://arxiv.org/abs/2611.16272",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "other"
      ],
      "category_names": [
        "Other"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T18:48:53.000Z",
      "jev": 0.4666,
      "judge_a": 0.6028,
      "judge_b": 0.7361,
      "score": 0.6018,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 35,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Yang Cao, Haoran Qi and Hanzhi Wang give a deterministic online binary search tree that serves m searches in O(m + UB(X)), where UB is Iacono's unified bound, and each search in O(log n). Earlier trees paid an extra m log log n, and none was known to meet the bound even offline. The tree keeps recently searched key blocks at power-of-two scales in a biased search tree. It also gives O((m+n)(1+log c_σ)) on pattern-avoiding sequences and handles insertions and deletions. The results are theoretical and unreviewed.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/20/",
      "card": "https://rubric.hrih.si/og/2026-11-15/20.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.66,
          "judge_a": 0.72,
          "judge_b": 0.88,
          "mean": 0.7533,
          "spread": 0.22
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.31,
          "judge_a": 0.78,
          "judge_b": 0.9,
          "mean": 0.6633,
          "spread": 0.59
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.45,
          "judge_a": 0.4,
          "judge_b": 0.32,
          "mean": 0.39,
          "spread": 0.13
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.295,
          "judge_a": 0.3,
          "judge_b": 0.55,
          "mean": 0.3817,
          "spread": 0.255
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.903,
          "judge_a": 0.85,
          "judge_b": 0.82,
          "mean": 0.8578,
          "spread": 0.083
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.9267,
          "spread": 0.08
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.95,
          "judge_b": 0.98,
          "mean": 0.9467,
          "spread": 0.07
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.02,
          "judge_b": 0.01,
          "mean": 0.0267,
          "spread": 0.04
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.2,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.16,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 21,
      "id": "st_20261115_25632",
      "title": "EnvOpt: Automated Optimization for Agent Environments",
      "url": "https://bespokelabs.ai/blog/envopt",
      "domain": "bespokelabs.ai",
      "site": "Bespoke Labs",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "claim_checked",
      "published_at": "2026-11-14T22:01:58.000Z",
      "jev": 0.6841,
      "judge_a": 0.5495,
      "judge_b": 0.8109,
      "score": 0.6814,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 10,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "EnvOpt is a method that automatically reorganizes the environments LLM agents work in, such as a company's Slack history or code repository, rather than tuning the model or harness. A coding agent proposes edits (indexes, cheatsheets, databases, new endpoints) using unverified training tasks, then validates them.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/21/",
      "card": "https://rubric.hrih.si/og/2026-11-15/21.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.69,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6567,
          "spread": 0.28
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.72,
          "judge_a": 0.7,
          "judge_b": 0.94,
          "mean": 0.7867,
          "spread": 0.24
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.7,
          "judge_a": 0.5,
          "judge_b": 0.72,
          "mean": 0.64,
          "spread": 0.22
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.718,
          "judge_a": 0.6,
          "judge_b": 0.86,
          "mean": 0.7258,
          "spread": 0.26
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.943,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.8211,
          "spread": 0.243
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.843,
          "judge_a": 0.7,
          "judge_b": 0.88,
          "mean": 0.8075,
          "spread": 0.18
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.95,
          "judge_b": 0.93,
          "mean": 0.9267,
          "spread": 0.05
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.15,
          "judge_b": 0.08,
          "mean": 0.11,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.3,
          "judge_a": 0.35,
          "judge_b": 0.18,
          "mean": 0.2767,
          "spread": 0.17
        }
      }
    },
    {
      "rank": 22,
      "id": "st_20261115_25246",
      "title": "Microsoft says services restored after global outage",
      "url": "https://www.theguardian.com/technology/2026/nov/14/microsoft-says-services-fully-restored-after-global-outage",
      "domain": "theguardian.com",
      "site": "the Guardian",
      "categories": [
        "devops",
        "it"
      ],
      "category_names": [
        "Infrastructure & DevOps",
        "IT & networking"
      ],
      "format": "News",
      "depth": "claim_checked",
      "published_at": "2026-11-14T11:15:42.000Z",
      "jev": 0.5173,
      "judge_a": 0.5714,
      "judge_b": 0.7099,
      "score": 0.5995,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 36,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Microsoft says its Microsoft 365 services, including Outlook, Teams and Copilot, are fully restored after a outage that began Thursday and disrupted governments, courts, hospitals, airlines and banks across more than 70 countries. The company said the fault stemmed from an earlier network capacity upgrade, not a cyberattack.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/22/",
      "card": "https://rubric.hrih.si/og/2026-11-15/22.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.593,
          "judge_a": 0.55,
          "judge_b": 0.72,
          "mean": 0.6208,
          "spread": 0.17
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.55,
          "judge_a": 0.8,
          "judge_b": 0.83,
          "mean": 0.7267,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.65,
          "judge_a": 0.67,
          "judge_b": 0.92,
          "mean": 0.7467,
          "spread": 0.27
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.42,
          "judge_a": 0.45,
          "judge_b": 0.72,
          "mean": 0.53,
          "spread": 0.3
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.533,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6044,
          "spread": 0.28
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.398,
          "judge_a": 0.15,
          "judge_b": 0.18,
          "mean": 0.2425,
          "spread": 0.248
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.7,
          "judge_b": 0.12,
          "mean": 0.5667,
          "spread": 0.76
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.03,
          "judge_b": 0.01,
          "mean": 0.0467,
          "spread": 0.09
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.15,
          "judge_a": 0.15,
          "judge_b": 0.08,
          "mean": 0.1267,
          "spread": 0.07
        }
      }
    },
    {
      "rank": 23,
      "id": "st_20261115_25359",
      "title": "Building a human-in-the-loop Slackbot agent",
      "url": "https://mays.co/human-in-the-loop-slackbot-agent",
      "domain": "mays.co",
      "site": "mays.co",
      "categories": [
        "programming",
        "ai",
        "devops",
        "fintech"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Infrastructure & DevOps",
        "Fintech"
      ],
      "format": "New project",
      "depth": "three_judges",
      "published_at": "2026-11-14T14:34:08.000Z",
      "jev": 0.7267,
      "judge_a": 0.518,
      "judge_b": 0.7612,
      "score": 0.6686,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 11,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "A developer's write-up on building a human-in-the-loop Slackbot agent for a demo marketplace that pays sellers through Stripe Connect. Each payout is screened against an OFAC-style sanctions list and routed to one of three tiers: code clears it, a Claude model clears low-risk gray-zone cases, or reviewers approve in Slack.…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/23/",
      "card": "https://rubric.hrih.si/og/2026-11-15/23.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.523,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.4808,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.76,
          "judge_a": 0.6,
          "judge_b": 0.88,
          "mean": 0.7467,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.61,
          "judge_a": 0.4,
          "judge_b": 0.63,
          "mean": 0.5467,
          "spread": 0.23
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.94,
          "judge_a": 0.7,
          "judge_b": 0.88,
          "mean": 0.84,
          "spread": 0.24
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.983,
          "judge_a": 0.75,
          "judge_b": 0.9,
          "mean": 0.8778,
          "spread": 0.233
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.993,
          "judge_a": 0.85,
          "judge_b": 0.87,
          "mean": 0.9042,
          "spread": 0.143
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.85,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9033,
          "spread": 0.11
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.1,
          "judge_b": 0.08,
          "mean": 0.0767,
          "spread": 0.05
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.13,
          "judge_a": 0.2,
          "judge_b": 0.12,
          "mean": 0.15,
          "spread": 0.08
        }
      }
    },
    {
      "rank": 24,
      "id": "st_20261115_25614",
      "title": "An Index Scan Is Only Half the Story",
      "url": "https://richyen.com/postgres/2026/11/09/index_scan_scattered_heap.html",
      "domain": "richyen.com",
      "site": "richyen.com",
      "categories": [
        "data"
      ],
      "category_names": [
        "Data"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T21:36:39.000Z",
      "jev": 0.6429,
      "judge_a": 0.474,
      "judge_b": 0.7078,
      "score": 0.6082,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 32,
      "topic_penalty": 0.02,
      "repeat_penalty": 0,
      "summary": "In a Microsoft Tech Community post, Microsoft's Azure PostgreSQL team describes a migrated query that was slow only after cache eviction, despite using the same index.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/24/",
      "card": "https://rubric.hrih.si/og/2026-11-15/24.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.507,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.4625,
          "spread": 0.28
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.66,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.6933,
          "spread": 0.22
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.517,
          "judge_a": 0.4,
          "judge_b": 0.42,
          "mean": 0.4456,
          "spread": 0.117
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.78,
          "judge_a": 0.7,
          "judge_b": 0.83,
          "mean": 0.77,
          "spread": 0.13
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.897,
          "judge_a": 0.6,
          "judge_b": 0.9,
          "mean": 0.7989,
          "spread": 0.3
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.993,
          "judge_a": 0.8,
          "judge_b": 0.78,
          "mean": 0.8575,
          "spread": 0.213
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.84,
          "judge_a": 0.9,
          "judge_b": 0.93,
          "mean": 0.89,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.25,
          "judge_b": 0.03,
          "mean": 0.1133,
          "spread": 0.22
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.13,
          "judge_a": 0.25,
          "judge_b": 0.06,
          "mean": 0.1467,
          "spread": 0.19
        }
      }
    },
    {
      "rank": 25,
      "id": "st_20261115_25251",
      "title": "Migrating from GitHub to Codeberg",
      "url": "https://yorickpeterse.com/articles/migrating-from-github-to-codeberg/",
      "domain": "yorickpeterse.com",
      "site": "yorickpeterse.com",
      "categories": [
        "programming",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "Infrastructure & DevOps"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T11:05:52.000Z",
      "jev": 0.5934,
      "judge_a": 0.489,
      "judge_b": 0.6778,
      "score": 0.5867,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 42,
      "topic_penalty": 0,
      "repeat_penalty": 0,
      "summary": "Yorick Peterse recounts migrating his personal projects and the Inko language repositories from GitHub to Codeberg, a Forgejo-based host he prefers for its stance on non-free software, its no-AI policy, and its governance.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/25/",
      "card": "https://rubric.hrih.si/og/2026-11-15/25.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.418,
          "judge_a": 0.3,
          "judge_b": 0.52,
          "mean": 0.4125,
          "spread": 0.22
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.54,
          "judge_a": 0.6,
          "judge_b": 0.84,
          "mean": 0.66,
          "spread": 0.3
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.633,
          "judge_a": 0.5,
          "judge_b": 0.45,
          "mean": 0.5278,
          "spread": 0.183
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.8,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7267,
          "spread": 0.2
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.64,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.64,
          "spread": 0.28
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.998,
          "judge_a": 0.7,
          "judge_b": 0.86,
          "mean": 0.8525,
          "spread": 0.298
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9133,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.1,
          "judge_b": 0.03,
          "mean": 0.06,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.2,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.1933,
          "spread": 0.02
        }
      }
    },
    {
      "rank": 26,
      "id": "st_20261115_25361",
      "title": "Reading JSON at 277 GB/s in Python",
      "url": "https://lemire.me/blog/2026/11/14/reading-json-at-277-gb-s-in-python/",
      "domain": "lemire.me",
      "site": "Daniel Lemire",
      "categories": [
        "programming",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "Data"
      ],
      "format": "Write-up",
      "depth": "claim_checked",
      "published_at": "2026-11-14T14:53:50.000Z",
      "jev": 0.6424,
      "judge_a": 0.433,
      "judge_b": 0.7445,
      "score": 0.6066,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 33,
      "topic_penalty": 0.02,
      "repeat_penalty": 0,
      "summary": "Daniel Lemire's post reports that fastsimdjson, a threaded JSON library for Python 3.14's free-threaded build, counts GitHub star events across 7.6 GB of NDJSON in 55 milliseconds, about 138 GB/s on a 192-core AWS instance.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/26/",
      "card": "https://rubric.hrih.si/og/2026-11-15/26.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.535,
          "judge_a": 0.4,
          "judge_b": 0.7,
          "mean": 0.545,
          "spread": 0.3
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.71,
          "judge_a": 0.6,
          "judge_b": 0.85,
          "mean": 0.72,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.57,
          "judge_a": 0.4,
          "judge_b": 0.62,
          "mean": 0.53,
          "spread": 0.22
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.778,
          "judge_a": 0.5,
          "judge_b": 0.82,
          "mean": 0.6992,
          "spread": 0.32
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.937,
          "judge_a": 0.7,
          "judge_b": 0.86,
          "mean": 0.8322,
          "spread": 0.237
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.93,
          "judge_a": 0.8,
          "judge_b": 0.78,
          "mean": 0.8367,
          "spread": 0.15
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.83,
          "judge_a": 0.9,
          "judge_b": 0.95,
          "mean": 0.8933,
          "spread": 0.12
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.3,
          "judge_b": 0.08,
          "mean": 0.15,
          "spread": 0.23
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.34,
          "judge_a": 0.4,
          "judge_b": 0.18,
          "mean": 0.3067,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 27,
      "id": "st_20261115_25387",
      "title": "Get Started with Gemini 4: Explicit Context Caching Not Supported",
      "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/start/get-started-with-gemini-4",
      "domain": "docs.cloud.google.com",
      "site": "Google Cloud Documentation",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T15:50:20.000Z",
      "jev": 0.6572,
      "judge_a": 0.4975,
      "judge_b": 0.8098,
      "score": 0.6548,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 12,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Google's developer guide for the Gemini 4 model family (Hydrogen, Carbon, and Argon) covers setup, a unified 1,048,576-token input-plus-output limit, long decodes for outputs beyond 65,536 tokens, thinking levels, and implicit caching.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/27/",
      "card": "https://rubric.hrih.si/og/2026-11-15/27.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.64,
          "judge_a": 0.45,
          "judge_b": 0.78,
          "mean": 0.6233,
          "spread": 0.33
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.57,
          "judge_a": 0.55,
          "judge_b": 0.9,
          "mean": 0.6733,
          "spread": 0.35
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.473,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.6311,
          "spread": 0.347
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.908,
          "judge_a": 0.75,
          "judge_b": 0.9,
          "mean": 0.8525,
          "spread": 0.157
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.767,
          "judge_a": 0.45,
          "judge_b": 0.8,
          "mean": 0.6722,
          "spread": 0.35
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.95,
          "judge_a": 0.6,
          "judge_b": 0.75,
          "mean": 0.7667,
          "spread": 0.35
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.9,
          "judge_b": 0.99,
          "mean": 0.9367,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.35,
          "judge_b": 0.18,
          "mean": 0.21,
          "spread": 0.25
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.17,
          "judge_a": 0.25,
          "judge_b": 0.08,
          "mean": 0.1667,
          "spread": 0.17
        }
      }
    },
    {
      "rank": 28,
      "id": "st_20261115_25540",
      "title": "Tensor Seeks Layout: Formalizing Layout Selection for ML Compilers",
      "url": "https://arxiv.org/abs/2608.21555",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "three_judges",
      "published_at": "2026-11-14T19:48:33.000Z",
      "jev": 0.629,
      "judge_a": 0.5322,
      "judge_b": 0.7955,
      "score": 0.6522,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 13,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "This paper gives the first formal treatment of tensor layout selection in machine learning compilers, where each tensor's memory layout must be chosen to minimize operator execution cost plus the cost of explicit layout conversions.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/28/",
      "card": "https://rubric.hrih.si/og/2026-11-15/28.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.675,
          "judge_a": 0.45,
          "judge_b": 0.78,
          "mean": 0.635,
          "spread": 0.33
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.62,
          "judge_a": 0.62,
          "judge_b": 0.9,
          "mean": 0.7133,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.553,
          "judge_a": 0.4,
          "judge_b": 0.62,
          "mean": 0.5244,
          "spread": 0.22
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.613,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6308,
          "spread": 0.28
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.943,
          "judge_a": 0.85,
          "judge_b": 0.88,
          "mean": 0.8911,
          "spread": 0.093
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.965,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.9483,
          "spread": 0.08
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.92,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.0433,
          "spread": 0.02
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.29,
          "judge_a": 0.2,
          "judge_b": 0.12,
          "mean": 0.2033,
          "spread": 0.17
        }
      }
    },
    {
      "rank": 29,
      "id": "st_20261115_25545",
      "title": "Building an Agentic Loop in Temporal",
      "url": "https://cadenya.com/handwritten/agentic-loop-in-temporal",
      "domain": "cadenya.com",
      "site": "Cadenya",
      "categories": [
        "programming",
        "ai",
        "devops",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Infrastructure & DevOps",
        "Data"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T19:59:07.000Z",
      "jev": 0.6765,
      "judge_a": 0.5035,
      "judge_b": 0.7626,
      "score": 0.6475,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 15,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Robert Ross, writing for Cadenya, walks through how his company's hosted agent runtime uses Temporal workflows to implement an agentic loop. The post covers the top-level AgentObjectiveWorkflow, a heartbeat child workflow that streams progress over SSE, and a transactional outbox that keeps objective status changes and webhook events consistent with…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/29/",
      "card": "https://rubric.hrih.si/og/2026-11-15/29.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.533,
          "judge_a": 0.35,
          "judge_b": 0.67,
          "mean": 0.5175,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.71,
          "judge_a": 0.7,
          "judge_b": 0.9,
          "mean": 0.77,
          "spread": 0.2
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.483,
          "judge_a": 0.4,
          "judge_b": 0.67,
          "mean": 0.5178,
          "spread": 0.27
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.85,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.7567,
          "spread": 0.25
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.943,
          "judge_a": 0.6,
          "judge_b": 0.9,
          "mean": 0.8144,
          "spread": 0.343
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.995,
          "judge_a": 0.9,
          "judge_b": 0.95,
          "mean": 0.9483,
          "spread": 0.095
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.86,
          "judge_a": 0.85,
          "judge_b": 0.78,
          "mean": 0.83,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.25,
          "judge_b": 0.06,
          "mean": 0.1233,
          "spread": 0.19
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.13,
          "judge_a": 0.15,
          "judge_b": 0.18,
          "mean": 0.1533,
          "spread": 0.05
        }
      }
    },
    {
      "rank": 30,
      "id": "st_20261115_25106",
      "title": "Windows 0-Day Dropped By Disgruntled Researcher Hands Hackers System Rights",
      "url": "https://it.slashdot.org/story/26/11/14/020218/windows-0-day-dropped-by-disgruntled-researcher-hands-hackers-system-rights",
      "domain": "it.slashdot.org",
      "site": "Slashdot",
      "categories": [
        "security",
        "it"
      ],
      "category_names": [
        "Security",
        "IT & networking"
      ],
      "format": "News",
      "depth": "three_judges",
      "published_at": "2026-11-14T03:30:00.000Z",
      "jev": 0.4807,
      "judge_a": 0.6098,
      "judge_b": 0.662,
      "score": 0.5841,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 45,
      "topic_penalty": 0.02,
      "repeat_penalty": 0,
      "summary": "A researcher using the pseudonym Nightmare Eclipse published SignDrop, a Windows zero-day in the Common Log File System that lets a local user with ordinary rights gain full System privileges by tricking a privileged process like Windows Defender into opening a crafted container.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/30/",
      "card": "https://rubric.hrih.si/og/2026-11-15/30.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.585,
          "judge_a": 0.55,
          "judge_b": 0.62,
          "mean": 0.585,
          "spread": 0.07
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.58,
          "judge_a": 0.8,
          "judge_b": 0.83,
          "mean": 0.7367,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.537,
          "judge_a": 0.6,
          "judge_b": 0.55,
          "mean": 0.5622,
          "spread": 0.063
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.485,
          "judge_a": 0.6,
          "judge_b": 0.63,
          "mean": 0.5717,
          "spread": 0.145
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.527,
          "judge_a": 0.5,
          "judge_b": 0.72,
          "mean": 0.5822,
          "spread": 0.22
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.75,
          "judge_a": 0.4,
          "judge_b": 0.48,
          "mean": 0.5433,
          "spread": 0.35
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.86,
          "judge_a": 0.85,
          "judge_b": 0.98,
          "mean": 0.8967,
          "spread": 0.13
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.03,
          "judge_b": 0.01,
          "mean": 0.0333,
          "spread": 0.05
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.75,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.3767,
          "spread": 0.57
        }
      }
    },
    {
      "rank": 31,
      "id": "st_20261115_25301",
      "title": "Engineering Fought Productivity Metrics. AI Ended the Fight",
      "url": "https://nerdstein.net/blog/measuring-developer-productivity/",
      "domain": "nerdstein.net",
      "site": "Nerdstein",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T13:26:42.000Z",
      "jev": 0.7301,
      "judge_a": 0.5115,
      "judge_b": 0.6891,
      "score": 0.6435,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 16,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Engineering organizations are now being asked to justify AI token spending, and that is ending their long exemption from productivity measurement. Uber's COO has said token costs are hard to justify without a clear line to shipped features, and the Linux Foundation has launched a Tokenomics Foundation to standardize AI…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/31/",
      "card": "https://rubric.hrih.si/og/2026-11-15/31.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.733,
          "judge_a": 0.4,
          "judge_b": 0.7,
          "mean": 0.6108,
          "spread": 0.333
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.8,
          "judge_a": 0.7,
          "judge_b": 0.86,
          "mean": 0.7867,
          "spread": 0.16
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.94,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.8067,
          "spread": 0.24
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.9,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.7733,
          "spread": 0.3
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.587,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6222,
          "spread": 0.093
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.645,
          "judge_a": 0.3,
          "judge_b": 0.55,
          "mean": 0.4983,
          "spread": 0.345
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.35,
          "judge_b": 0.82,
          "mean": 0.6933,
          "spread": 0.56
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.12,
          "judge_a": 0.1,
          "judge_b": 0.04,
          "mean": 0.0867,
          "spread": 0.08
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.53,
          "judge_a": 0.3,
          "judge_b": 0.72,
          "mean": 0.5167,
          "spread": 0.42
        }
      }
    },
    {
      "rank": 32,
      "id": "st_20261115_25165",
      "title": "AWS Emulator in detail: The whole cloud on your own computer?",
      "url": "https://www.heise.de/en/news/AWS-Emulator-in-detail-The-whole-cloud-on-your-own-computer-11524820.html",
      "domain": "heise.de",
      "site": "heise online English",
      "categories": [
        "programming",
        "ai",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Infrastructure & DevOps"
      ],
      "format": "News",
      "depth": "three_judges",
      "published_at": "2026-11-14T06:32:00.000Z",
      "jev": 0.5583,
      "judge_a": 0.5452,
      "judge_b": 0.6902,
      "score": 0.5978,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 37,
      "topic_penalty": 0,
      "repeat_penalty": 0.0357,
      "summary": "AWS has released an open-source AWS Emulator, written in Rust and generated from public AWS API models, that lets developers and AI coding agents test AWS workflows locally without deployment or AWS credentials.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/32/",
      "card": "https://rubric.hrih.si/og/2026-11-15/32.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.548,
          "judge_a": 0.45,
          "judge_b": 0.68,
          "mean": 0.5592,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.61,
          "judge_a": 0.72,
          "judge_b": 0.82,
          "mean": 0.7167,
          "spread": 0.21
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.557,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6122,
          "spread": 0.123
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.735,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.7383,
          "spread": 0.08
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.557,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.6589,
          "spread": 0.263
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.783,
          "judge_a": 0.65,
          "judge_b": 0.72,
          "mean": 0.7175,
          "spread": 0.132
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.76,
          "judge_a": 0.2,
          "judge_b": 0.28,
          "mean": 0.4133,
          "spread": 0.56
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.13,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.1367,
          "spread": 0.12
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.42,
          "judge_a": 0.15,
          "judge_b": 0.2,
          "mean": 0.2567,
          "spread": 0.27
        }
      }
    },
    {
      "rank": 33,
      "id": "st_20261115_25436",
      "title": "Training and Inference on Huawei Ascend NPUs",
      "url": "https://github.com/deepseek-ai/DeepEP-Ascend",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai",
        "devops",
        "hardware"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Infrastructure & DevOps",
        "Hardware & devices"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T16:30:03.000Z",
      "jev": 0.558,
      "judge_a": null,
      "judge_b": 0.7203,
      "score": 0.6391,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 17,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "DeepEP-Ascend is an open-source communication library from DeepSeek for Huawei Ascend 950 NPUs, offering expert-parallel all-to-all for MoE dispatch and combine, with FP8 dispatch and deferred epilogues, and an EPBuffer API aligned with NVIDIA's DeepEP.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/33/",
      "card": "https://rubric.hrih.si/og/2026-11-15/33.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.528,
          "judge_a": null,
          "judge_b": 0.68,
          "mean": 0.6038,
          "spread": 0.153
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.51,
          "judge_a": null,
          "judge_b": 0.84,
          "mean": 0.675,
          "spread": 0.33
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.34,
          "judge_a": null,
          "judge_b": 0.48,
          "mean": 0.41,
          "spread": 0.14
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.685,
          "judge_a": null,
          "judge_b": 0.76,
          "mean": 0.7225,
          "spread": 0.075
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.927,
          "judge_a": null,
          "judge_b": 0.88,
          "mean": 0.9033,
          "spread": 0.047
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.995,
          "judge_a": null,
          "judge_b": 0.94,
          "mean": 0.9675,
          "spread": 0.055
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": null,
          "judge_b": 0.93,
          "mean": 0.925,
          "spread": 0.01
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.08,
          "judge_a": null,
          "judge_b": 0.08,
          "mean": 0.08,
          "spread": 0
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.21,
          "judge_a": null,
          "judge_b": 0.12,
          "mean": 0.165,
          "spread": 0.09
        }
      }
    },
    {
      "rank": 34,
      "id": "st_20261115_25511",
      "title": "I Tried Every LLM Architecture Change Since 2020",
      "url": "https://www.nikita-pavlichenko.com/blog/every-llm-architecture-change",
      "domain": "nikita-pavlichenko.com",
      "site": "nikita-pavlichenko.com",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "New project",
      "depth": "three_judges",
      "published_at": "2026-11-14T18:41:54.000Z",
      "jev": 0.6025,
      "judge_a": 0.7376,
      "judge_b": 0.8353,
      "score": 0.6251,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0.1,
      "score_rank": 22,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "A practitioner ran over 1,000 controlled experiments comparing LLM architecture changes from 2020 to 2026 on a shared setup, measuring quality in bits per byte and serving cost on a fixed GPU budget. Many individual changes were tested alone, then stacked to see whether gains survived.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/34/",
      "card": "https://rubric.hrih.si/og/2026-11-15/34.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.588,
          "judge_a": 0.55,
          "judge_b": 0.82,
          "mean": 0.6525,
          "spread": 0.27
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.7,
          "judge_a": 0.9,
          "judge_b": 0.95,
          "mean": 0.85,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.613,
          "judge_a": 0.72,
          "judge_b": 0.72,
          "mean": 0.6844,
          "spread": 0.107
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.523,
          "judge_a": 0.8,
          "judge_b": 0.82,
          "mean": 0.7142,
          "spread": 0.298
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.887,
          "judge_a": 0.82,
          "judge_b": 0.9,
          "mean": 0.8689,
          "spread": 0.08
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.995,
          "judge_a": 0.95,
          "judge_b": 0.98,
          "mean": 0.975,
          "spread": 0.045
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.92,
          "spread": 0.1
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.0433,
          "spread": 0.02
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.4,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.26,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 35,
      "id": "st_20261115_25284",
      "title": "Agile Manifesto for Agents",
      "url": "https://debarshibasak.github.io/readables/blogs/agent-first-development",
      "domain": "debarshibasak.github.io",
      "site": "debarshibasak.github.io",
      "categories": [
        "programming",
        "ai",
        "product"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Product"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T12:38:43.000Z",
      "jev": 0.7376,
      "judge_a": 0.4395,
      "judge_b": 0.6977,
      "score": 0.6249,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 23,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Agent First Development is a proposed methodology for teams building software with AI agents, published by Debarshi Basak. It argues that fast vibe-coding prototypes don't ensure deliverable systems, and combines pre-implementation scope agreement, bounded agent tasks, shared contracts, and evidence-based acceptance, borrowing from waterfall, Agile, and SAFe.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/35/",
      "card": "https://rubric.hrih.si/og/2026-11-15/35.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.663,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.5275,
          "spread": 0.363
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.79,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7233,
          "spread": 0.19
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.84,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.72,
          "spread": 0.24
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.93,
          "judge_a": 0.55,
          "judge_b": 0.82,
          "mean": 0.7667,
          "spread": 0.38
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.277,
          "judge_a": 0.25,
          "judge_b": 0.58,
          "mean": 0.3689,
          "spread": 0.33
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.698,
          "judge_a": 0.3,
          "judge_b": 0.48,
          "mean": 0.4925,
          "spread": 0.398
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.85,
          "judge_a": 0.85,
          "judge_b": 0.76,
          "mean": 0.82,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.1,
          "judge_b": 0.03,
          "mean": 0.0667,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.17,
          "judge_a": 0.35,
          "judge_b": 0.12,
          "mean": 0.2133,
          "spread": 0.23
        }
      }
    },
    {
      "rank": 36,
      "id": "st_20261115_25073",
      "title": "Ggml v1.0.0",
      "url": "https://github.com/ggml-org/ggml/releases/tag/v1.0.0",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T02:28:29.000Z",
      "jev": 0.6593,
      "judge_a": 0.5245,
      "judge_b": 0.7171,
      "score": 0.6336,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 20,
      "topic_penalty": 0.08,
      "repeat_penalty": 0.0099,
      "summary": "The ggml 1.0.0 release, published through the llama.cpp project, removes deprecated public API and legacy diagonal-mask ops, and adds YaRN truncation control and FP8 support. It also broadens performance work across CUDA, Metal, Vulkan, SYCL, OpenCL and WebGPU backends, including MoE fusion and FlashAttention paths.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/36/",
      "card": "https://rubric.hrih.si/og/2026-11-15/36.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.71,
          "judge_a": 0.45,
          "judge_b": 0.65,
          "mean": 0.6033,
          "spread": 0.26
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.59,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6233,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.52,
          "judge_a": 0.5,
          "judge_b": 0.55,
          "mean": 0.5233,
          "spread": 0.05
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.753,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6908,
          "spread": 0.152
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.847,
          "judge_a": 0.6,
          "judge_b": 0.95,
          "mean": 0.7989,
          "spread": 0.35
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.795,
          "judge_a": 0.7,
          "judge_b": 0.9,
          "mean": 0.7983,
          "spread": 0.2
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.9,
          "judge_b": 0.92,
          "mean": 0.9067,
          "spread": 0.02
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.0433,
          "spread": 0.02
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.14,
          "judge_a": 0.1,
          "judge_b": 0.04,
          "mean": 0.0933,
          "spread": 0.1
        }
      }
    },
    {
      "rank": 37,
      "id": "st_20261115_25610",
      "title": "My lunch-break response to ARTEX: a host intrusion sentinel",
      "url": "https://github.com/gary23w/agent-garrett-sentinel-service",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "security",
        "devops",
        "it"
      ],
      "category_names": [
        "Security",
        "Infrastructure & DevOps",
        "IT & networking"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T21:30:57.000Z",
      "jev": 0.6543,
      "judge_a": 0.4195,
      "judge_b": 0.6689,
      "score": 0.5809,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 48,
      "topic_penalty": 0.04,
      "repeat_penalty": 0,
      "summary": "garrett is a small open-source host intrusion sentinel: an agent on each machine scores running programs and their network connections against attack signals, then reports suspicious cases to a service in the operator's own Cloudflare account.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/37/",
      "card": "https://rubric.hrih.si/og/2026-11-15/37.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.513,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.4642,
          "spread": 0.28
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.7,
          "judge_a": 0.6,
          "judge_b": 0.84,
          "mean": 0.7133,
          "spread": 0.24
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.54,
          "judge_a": 0.4,
          "judge_b": 0.45,
          "mean": 0.4633,
          "spread": 0.14
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.82,
          "judge_a": 0.7,
          "judge_b": 0.7,
          "mean": 0.74,
          "spread": 0.12
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.923,
          "judge_a": 0.5,
          "judge_b": 0.82,
          "mean": 0.7478,
          "spread": 0.423
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.995,
          "judge_a": 0.85,
          "judge_b": 0.94,
          "mean": 0.9283,
          "spread": 0.145
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.86,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9067,
          "spread": 0.1
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.05,
          "judge_a": 0.35,
          "judge_b": 0.08,
          "mean": 0.16,
          "spread": 0.3
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.28,
          "judge_a": 0.55,
          "judge_b": 0.2,
          "mean": 0.3433,
          "spread": 0.35
        }
      }
    },
    {
      "rank": 38,
      "id": "st_20261115_25170",
      "title": "Notes on Incidents",
      "url": "https://www.seangoedecke.com/notes-on-incidents/",
      "domain": "seangoedecke.com",
      "site": "seangoedecke.com",
      "categories": [
        "devops"
      ],
      "category_names": [
        "Infrastructure & DevOps"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T07:00:42.000Z",
      "jev": 0.5448,
      "judge_a": 0.4895,
      "judge_b": 0.6172,
      "score": 0.5505,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 62,
      "topic_penalty": 0.02,
      "repeat_penalty": 0,
      "summary": "In \"Notes on Incidents,\" the author argues from years of on-call experience that most incidents resolve without human help, that hasty fixes often make things worse, and that the best response is usually a simple, well-chosen action like disabling a feature flag, taken decisively.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/38/",
      "card": "https://rubric.hrih.si/og/2026-11-15/38.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.42,
          "judge_a": 0.2,
          "judge_b": 0.52,
          "mean": 0.38,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.53,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.67,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.8,
          "judge_a": 0.6,
          "judge_b": 0.67,
          "mean": 0.69,
          "spread": 0.2
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.8,
          "judge_a": 0.7,
          "judge_b": 0.72,
          "mean": 0.74,
          "spread": 0.1
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.08,
          "judge_a": 0.3,
          "judge_b": 0.42,
          "mean": 0.2667,
          "spread": 0.34
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.535,
          "judge_a": 0.2,
          "judge_b": 0.48,
          "mean": 0.405,
          "spread": 0.335
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.84,
          "judge_a": 0.9,
          "judge_b": 0.9,
          "mean": 0.88,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.05,
          "judge_b": 0.01,
          "mean": 0.04,
          "spread": 0.05
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.32,
          "judge_a": 0.3,
          "judge_b": 0.4,
          "mean": 0.34,
          "spread": 0.1
        }
      }
    },
    {
      "rank": 39,
      "id": "st_20261115_25297",
      "title": "What new PyPI packages import: LLM, httpx, pydantic v2",
      "url": "https://py-code.org/blog/new-pypi-packages-imports/",
      "domain": "py-code.org",
      "site": "py-code.org",
      "categories": [
        "programming",
        "ai",
        "data"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Data"
      ],
      "format": "Paper",
      "depth": "three_judges",
      "published_at": "2026-11-14T13:28:13.000Z",
      "jev": 0.6368,
      "judge_a": 0.4375,
      "judge_b": 0.7525,
      "score": 0.6089,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 31,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "A py-code.org analysis of the latest releases of 335,167 PyPI projects published or updated between November 2025 and November 2026 finds that the MCP Python SDK is imported by more new packages (20,013) than any LLM SDK, including openai (12,096). Six LLM SDKs appear in 6.7% of projects.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/39/",
      "card": "https://rubric.hrih.si/og/2026-11-15/39.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.525,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.4817,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.63,
          "judge_a": 0.55,
          "judge_b": 0.88,
          "mean": 0.6867,
          "spread": 0.33
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.68,
          "judge_a": 0.5,
          "judge_b": 0.74,
          "mean": 0.64,
          "spread": 0.24
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.65,
          "judge_a": 0.55,
          "judge_b": 0.7,
          "mean": 0.6333,
          "spread": 0.15
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.997,
          "judge_a": 0.75,
          "judge_b": 0.98,
          "mean": 0.9089,
          "spread": 0.247
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.995,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.805,
          "spread": 0.395
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.89,
          "judge_a": 0.8,
          "judge_b": 0.94,
          "mean": 0.8767,
          "spread": 0.14
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.04,
          "judge_a": 0.2,
          "judge_b": 0.03,
          "mean": 0.09,
          "spread": 0.17
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.19,
          "judge_a": 0.3,
          "judge_b": 0.06,
          "mean": 0.1833,
          "spread": 0.24
        }
      }
    },
    {
      "rank": 40,
      "id": "st_20261115_25470",
      "title": "Will We Put Data Centers in Space?",
      "url": "https://www.forethought.org/research/will-we-really-put-data-centers-in-space",
      "domain": "forethought.org",
      "site": "Forethought",
      "categories": [
        "ai",
        "hardware"
      ],
      "category_names": [
        "AI & machine learning",
        "Hardware & devices"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T17:35:33.000Z",
      "jev": 0.5588,
      "judge_a": 0.5415,
      "judge_b": 0.7191,
      "score": 0.6064,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 34,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "The report, published by an AI-governance research group on 22 May 2026, assesses whether orbital data centers could become economically viable and meaningful within a few years.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/40/",
      "card": "https://rubric.hrih.si/og/2026-11-15/40.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.578,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.6358,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.68,
          "judge_a": 0.8,
          "judge_b": 0.88,
          "mean": 0.7867,
          "spread": 0.2
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.753,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6911,
          "spread": 0.153
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.323,
          "judge_a": 0.35,
          "judge_b": 0.62,
          "mean": 0.4308,
          "spread": 0.298
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.633,
          "judge_a": 0.6,
          "judge_b": 0.7,
          "mean": 0.6444,
          "spread": 0.1
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.893,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.8042,
          "spread": 0.193
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.94,
          "judge_a": 0.7,
          "judge_b": 0.95,
          "mean": 0.8633,
          "spread": 0.25
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.05,
          "judge_b": 0.08,
          "mean": 0.0733,
          "spread": 0.04
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.29,
          "judge_a": 0.4,
          "judge_b": 0.32,
          "mean": 0.3367,
          "spread": 0.11
        }
      }
    },
    {
      "rank": 41,
      "id": "st_20261115_25455",
      "title": "Linus Torvalds Wonders If Linux's AI-Prompted Patch Flood Has Peaked",
      "url": "https://linux.slashdot.org/story/26/11/14/0827227/linus-torvalds-wonders-if-linuxs-ai-prompted-patch-flood-has-peaked",
      "domain": "linux.slashdot.org",
      "site": "Slashdot",
      "categories": [
        "programming",
        "ai",
        "security"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security"
      ],
      "format": "News",
      "depth": "three_judges",
      "published_at": "2026-11-14T17:00:00.000Z",
      "jev": 0.5946,
      "judge_a": 0.5584,
      "judge_b": 0.5869,
      "score": 0.58,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 49,
      "topic_penalty": 0.04,
      "repeat_penalty": 0.0204,
      "summary": "Linus Torvalds says the surge of AI-prompted patches flooding the Linux kernel may be peaking, citing faster bpf-side fixes from Alexei Starovoitov, though he doubts most areas have reached that point yet.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/41/",
      "card": "https://rubric.hrih.si/og/2026-11-15/41.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.655,
          "judge_a": 0.45,
          "judge_b": 0.62,
          "mean": 0.575,
          "spread": 0.205
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.64,
          "judge_a": 0.8,
          "judge_b": 0.82,
          "mean": 0.7533,
          "spread": 0.18
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.763,
          "judge_a": 0.67,
          "judge_b": 0.48,
          "mean": 0.6378,
          "spread": 0.283
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.58,
          "judge_a": 0.55,
          "judge_b": 0.58,
          "mean": 0.57,
          "spread": 0.03
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.627,
          "judge_a": 0.45,
          "judge_b": 0.72,
          "mean": 0.5989,
          "spread": 0.27
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.45,
          "judge_a": 0.3,
          "judge_b": 0.42,
          "mean": 0.39,
          "spread": 0.15
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.86,
          "judge_a": 0.6,
          "judge_b": 0.91,
          "mean": 0.79,
          "spread": 0.31
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.08,
          "judge_a": 0.03,
          "judge_b": 0.02,
          "mean": 0.0433,
          "spread": 0.06
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.34,
          "judge_a": 0.25,
          "judge_b": 0.58,
          "mean": 0.39,
          "spread": 0.33
        }
      }
    },
    {
      "rank": 42,
      "id": "st_20261115_25406",
      "title": "The Missing Ring-0 Protections for AI Agents",
      "url": "https://hwtn.io/essays/t0013-missing-ring-0-protections-ai",
      "domain": "hwtn.io",
      "site": "hawthorne",
      "categories": [
        "programming",
        "ai",
        "security"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T16:23:05.000Z",
      "jev": 0.7023,
      "judge_a": 0.5332,
      "judge_b": 0.8161,
      "score": 0.5838,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0.1,
      "score_rank": 46,
      "topic_penalty": 0.06,
      "repeat_penalty": 0.0062,
      "summary": "A study of three deployed coding agents (Claude Code, Codex CLI, Gemini CLI) argues they effectively run at processor ring 0 with no independent check on what the model decides, since access checks sit inside individual tools and can be bypassed through the shell, restored sessions and sub-agents.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/42/",
      "card": "https://rubric.hrih.si/og/2026-11-15/42.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.618,
          "judge_a": 0.4,
          "judge_b": 0.78,
          "mean": 0.5992,
          "spread": 0.38
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.76,
          "judge_a": 0.72,
          "judge_b": 0.9,
          "mean": 0.7933,
          "spread": 0.18
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.853,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7244,
          "spread": 0.253
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.72,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.7133,
          "spread": 0.22
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.657,
          "judge_a": 0.55,
          "judge_b": 0.87,
          "mean": 0.6922,
          "spread": 0.32
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.95,
          "judge_a": 0.7,
          "judge_b": 0.91,
          "mean": 0.8533,
          "spread": 0.25
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.89,
          "judge_a": 0.7,
          "judge_b": 0.94,
          "mean": 0.8433,
          "spread": 0.24
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.1,
          "judge_b": 0.01,
          "mean": 0.0567,
          "spread": 0.09
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.24,
          "judge_a": 0.35,
          "judge_b": 0.12,
          "mean": 0.2367,
          "spread": 0.23
        }
      }
    },
    {
      "rank": 43,
      "id": "st_20261115_25275",
      "title": "Bright Lines – a fail-closed guard hook for Claude Code",
      "url": "https://github.com/LeriderTechnologies/bright-lines",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai",
        "security"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T12:03:16.000Z",
      "jev": 0.6997,
      "judge_a": 0.4995,
      "judge_b": 0.5916,
      "score": 0.5969,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 38,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Bright Lines is an open-source (MIT) guard setup for Claude Code, shared on Hacker News by Lerider Technologies. It adds a fail-closed hook that asks for approval on edits, commands, web lookups and helper agents, and refuses a fixed list of dangerous actions even if approved.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/43/",
      "card": "https://rubric.hrih.si/og/2026-11-15/43.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.528,
          "judge_a": 0.3,
          "judge_b": 0.38,
          "mean": 0.4025,
          "spread": 0.227
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.79,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7233,
          "spread": 0.19
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.523,
          "judge_a": 0.4,
          "judge_b": 0.35,
          "mean": 0.4244,
          "spread": 0.173
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.9,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.7933,
          "spread": 0.2
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.963,
          "judge_a": 0.72,
          "judge_b": 0.86,
          "mean": 0.8478,
          "spread": 0.243
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.883,
          "judge_a": 0.7,
          "judge_b": 0.72,
          "mean": 0.7675,
          "spread": 0.183
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.9,
          "judge_b": 0.93,
          "mean": 0.9033,
          "spread": 0.05
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.15,
          "judge_b": 0.04,
          "mean": 0.0933,
          "spread": 0.11
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.16,
          "judge_a": 0.2,
          "judge_b": 0.34,
          "mean": 0.2333,
          "spread": 0.18
        }
      }
    },
    {
      "rank": 44,
      "id": "st_20261115_25388",
      "title": "Today's forecast: AI behaving unpredictably",
      "url": "https://aistormdesk.com/",
      "domain": "aistormdesk.com",
      "site": "aistormdesk.com",
      "categories": [
        "ai",
        "security"
      ],
      "category_names": [
        "AI & machine learning",
        "Security"
      ],
      "format": "News",
      "depth": "three_judges",
      "published_at": "2026-11-14T15:54:55.000Z",
      "jev": 0.5741,
      "judge_a": 0.5175,
      "judge_b": 0.6845,
      "score": 0.592,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 39,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "The AI Incident Database logged record #1807, in which an OpenAI research agent, reportedly lacking a wallet for deploying smart contracts, found a publicly known developer recovery phrase and used a funded test-network account to pay transaction fees without authorization.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/44/",
      "card": "https://rubric.hrih.si/og/2026-11-15/44.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.528,
          "judge_a": 0.45,
          "judge_b": 0.62,
          "mean": 0.5325,
          "spread": 0.17
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.59,
          "judge_a": 0.7,
          "judge_b": 0.8,
          "mean": 0.6967,
          "spread": 0.21
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.8,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7067,
          "spread": 0.2
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.605,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6283,
          "spread": 0.08
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.627,
          "judge_a": 0.45,
          "judge_b": 0.76,
          "mean": 0.6122,
          "spread": 0.31
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.46,
          "judge_a": 0.15,
          "judge_b": 0.38,
          "mean": 0.33,
          "spread": 0.31
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.85,
          "judge_a": 0.6,
          "judge_b": 0.84,
          "mean": 0.7633,
          "spread": 0.25
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.05,
          "judge_b": 0.03,
          "mean": 0.0467,
          "spread": 0.03
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.26,
          "judge_a": 0.3,
          "judge_b": 0.08,
          "mean": 0.2133,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 45,
      "id": "st_20261115_25141",
      "title": "Formal Verification Is Not Verifiable: Bouba and Kiki Software",
      "url": "https://alperenkeles.com/posts/formal-verification-is-not-verifiable/",
      "domain": "alperenkeles.com",
      "site": "alperenkeles.com",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T05:37:10.000Z",
      "jev": 0.6056,
      "judge_a": 0.43,
      "judge_b": 0.7265,
      "score": 0.5873,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 41,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Alperen Keles argues that AI has pushed formal verification into the mainstream, with startups multiplying and Anthropic reporting Claude wrote a billion lines of Rocq for one project.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/45/",
      "card": "https://rubric.hrih.si/og/2026-11-15/45.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.62,
          "judge_a": 0.35,
          "judge_b": 0.78,
          "mean": 0.5833,
          "spread": 0.43
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.72,
          "judge_a": 0.6,
          "judge_b": 0.88,
          "mean": 0.7333,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.837,
          "judge_a": 0.55,
          "judge_b": 0.72,
          "mean": 0.7022,
          "spread": 0.287
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.633,
          "judge_a": 0.45,
          "judge_b": 0.68,
          "mean": 0.5875,
          "spread": 0.23
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.283,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.3878,
          "spread": 0.297
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.675,
          "judge_a": 0.45,
          "judge_b": 0.72,
          "mean": 0.615,
          "spread": 0.27
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.6,
          "judge_b": 0.7,
          "mean": 0.7267,
          "spread": 0.28
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.11,
          "judge_a": 0.05,
          "judge_b": 0.01,
          "mean": 0.0567,
          "spread": 0.1
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.45,
          "judge_a": 0.35,
          "judge_b": 0.28,
          "mean": 0.36,
          "spread": 0.17
        }
      }
    },
    {
      "rank": 46,
      "id": "st_20261115_25356",
      "title": "Tokenizer Fairness in 2026",
      "url": "https://partyfly.github.io/tokenizer-fairness-2026/",
      "domain": "partyfly.github.io",
      "site": "partyfly.github.io",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "three_judges",
      "published_at": "2026-11-14T14:55:44.000Z",
      "jev": 0.5724,
      "judge_a": 0.5245,
      "judge_b": 0.6602,
      "score": 0.5856,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 43,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "The piece measures \"premium,\" the ratio of tokens a language needs compared with English for the same FLORES-200 sentences, using each lab's latest tokenizer as of 2026.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/46/",
      "card": "https://rubric.hrih.si/og/2026-11-15/46.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.558,
          "judge_a": 0.4,
          "judge_b": 0.58,
          "mean": 0.5125,
          "spread": 0.18
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.5,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.6267,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.65,
          "judge_a": 0.55,
          "judge_b": 0.62,
          "mean": 0.6067,
          "spread": 0.1
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.598,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6258,
          "spread": 0.083
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.933,
          "judge_a": 0.8,
          "judge_b": 0.72,
          "mean": 0.8178,
          "spread": 0.213
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.848,
          "judge_a": 0.7,
          "judge_b": 0.62,
          "mean": 0.7225,
          "spread": 0.228
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.7,
          "judge_b": 0.93,
          "mean": 0.8467,
          "spread": 0.23
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.1,
          "judge_b": 0.03,
          "mean": 0.0767,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.21,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.1967,
          "spread": 0.03
        }
      }
    },
    {
      "rank": 47,
      "id": "st_20261115_25082",
      "title": "Microsoft 365 Crawls Back to Life After a 24-Hour Outage",
      "url": "https://www.wired.com/story/microsoft-365-crawls-back-to-life-after-a-24-hour-outage/",
      "domain": "wired.com",
      "site": "WIRED",
      "categories": [
        "devops",
        "it"
      ],
      "category_names": [
        "Infrastructure & DevOps",
        "IT & networking"
      ],
      "format": "News",
      "depth": "three_judges",
      "published_at": "2026-11-14T02:56:22.000Z",
      "jev": 0.4699,
      "judge_a": 0.5575,
      "judge_b": 0.6396,
      "score": 0.5556,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 58,
      "topic_penalty": 0,
      "repeat_penalty": 0.05,
      "summary": "A roughly 24-hour outage knocked Microsoft 365 services offline for businesses, governments, and users in more than 70 countries, disrupting email, Teams, airlines, emergency services, and hospitals. Microsoft says a network capacity upgrade caused the failure, ruling out cyberattacks and AI agents, and promises a full post-incident report.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/47/",
      "card": "https://rubric.hrih.si/og/2026-11-15/47.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.575,
          "judge_a": 0.55,
          "judge_b": 0.7,
          "mean": 0.6083,
          "spread": 0.15
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.5,
          "judge_a": 0.8,
          "judge_b": 0.78,
          "mean": 0.6933,
          "spread": 0.3
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.577,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.6989,
          "spread": 0.243
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.363,
          "judge_a": 0.4,
          "judge_b": 0.58,
          "mean": 0.4475,
          "spread": 0.217
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.453,
          "judge_a": 0.4,
          "judge_b": 0.68,
          "mean": 0.5111,
          "spread": 0.28
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.443,
          "judge_a": 0.15,
          "judge_b": 0.32,
          "mean": 0.3042,
          "spread": 0.293
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.87,
          "judge_a": 0.85,
          "judge_b": 0.12,
          "mean": 0.6133,
          "spread": 0.75
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.05,
          "judge_b": 0.02,
          "mean": 0.0533,
          "spread": 0.07
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.22,
          "judge_a": 0.2,
          "judge_b": 0.16,
          "mean": 0.1933,
          "spread": 0.06
        }
      }
    },
    {
      "rank": 48,
      "id": "st_20261115_25373",
      "title": "DeepSeek V3 in ~200 lines of pure Jax",
      "url": "https://github.com/Vaibhaav-Tiwari/deepseek-v3-jax",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T15:04:15.000Z",
      "jev": 0.593,
      "judge_a": 0.5035,
      "judge_b": 0.6765,
      "score": 0.591,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 40,
      "topic_penalty": 0.08,
      "repeat_penalty": 0.0063,
      "summary": "A solo developer reimplemented the DeepSeek-V3 architecture in about 190 lines of pure JAX, with no Flax, Equinox or PyTorch, and pretrained a 36M-parameter model from scratch on public-domain detective fiction on a MacBook Air.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/48/",
      "card": "https://rubric.hrih.si/og/2026-11-15/48.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.485,
          "judge_a": 0.3,
          "judge_b": 0.53,
          "mean": 0.4383,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.69,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.7367,
          "spread": 0.13
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.49,
          "judge_a": 0.4,
          "judge_b": 0.48,
          "mean": 0.4567,
          "spread": 0.09
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.598,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6258,
          "spread": 0.083
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.987,
          "judge_a": 0.7,
          "judge_b": 0.9,
          "mean": 0.8622,
          "spread": 0.287
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.98,
          "judge_a": 0.8,
          "judge_b": 0.86,
          "mean": 0.88,
          "spread": 0.18
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.85,
          "judge_a": 0.85,
          "judge_b": 0.94,
          "mean": 0.88,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.06,
          "judge_a": 0.1,
          "judge_b": 0.04,
          "mean": 0.0667,
          "spread": 0.06
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.25,
          "judge_a": 0.3,
          "judge_b": 0.03,
          "mean": 0.1933,
          "spread": 0.27
        }
      }
    },
    {
      "rank": 49,
      "id": "st_20261115_25125",
      "title": "You could have invented transformers from memory",
      "url": "https://alexwang.ai/posts/associative-memory/",
      "domain": "alexwang.ai",
      "site": "Alex Wang",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Analysis",
      "depth": "three_judges",
      "published_at": "2026-11-14T04:49:47.000Z",
      "jev": 0.5974,
      "judge_a": 0.6651,
      "judge_b": 0.7908,
      "score": 0.5844,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0.1,
      "score_rank": 44,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "This post argues that transformers' feedforward layers and self-attention are both associative memory stores, and that raising similarities to higher powers, or exponentiating them as softmax does, sharply increases how many memories can be recalled without interference.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/49/",
      "card": "https://rubric.hrih.si/og/2026-11-15/49.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.61,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.6467,
          "spread": 0.23
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.68,
          "judge_a": 0.82,
          "judge_b": 0.95,
          "mean": 0.8167,
          "spread": 0.27
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.663,
          "judge_a": 0.67,
          "judge_b": 0.72,
          "mean": 0.6844,
          "spread": 0.057
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.56,
          "judge_a": 0.6,
          "judge_b": 0.66,
          "mean": 0.6067,
          "spread": 0.1
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.733,
          "judge_a": 0.8,
          "judge_b": 0.88,
          "mean": 0.8044,
          "spread": 0.147
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.993,
          "judge_a": 0.92,
          "judge_b": 0.97,
          "mean": 0.9608,
          "spread": 0.073
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.85,
          "judge_b": 0.94,
          "mean": 0.9,
          "spread": 0.09
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.04,
          "judge_a": 0.03,
          "judge_b": 0.01,
          "mean": 0.0267,
          "spread": 0.03
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.54,
          "judge_a": 0.2,
          "judge_b": 0.18,
          "mean": 0.3067,
          "spread": 0.36
        }
      }
    },
    {
      "rank": 50,
      "id": "st_20261115_25364",
      "title": "TriusLD – A Small Tokenizer for LLMs with Robust Multilingual Support",
      "url": "https://trius.canton.graphics/blog/000003.html",
      "domain": "trius.canton.graphics",
      "site": "trius.canton.graphics",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T15:23:03.000Z",
      "jev": 0.6116,
      "judge_a": 0.3895,
      "judge_b": 0.747,
      "score": 0.5826,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 47,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "TriusLD is a 65,535-slot tokenizer for LLMs that allocates vocabulary across 19 writing systems using per-script quotas plus a shared byte fallback. Its author reports 0.52 tokens per character averaged over 58 languages, versus 0.84 for GPT-4's vocabulary, while English gets 3.85 characters per token, about 18% behind a 200,000-token…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/50/",
      "card": "https://rubric.hrih.si/og/2026-11-15/50.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.528,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.4825,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.63,
          "judge_a": 0.45,
          "judge_b": 0.9,
          "mean": 0.66,
          "spread": 0.45
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.587,
          "judge_a": 0.4,
          "judge_b": 0.72,
          "mean": 0.5689,
          "spread": 0.32
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.645,
          "judge_a": 0.5,
          "judge_b": 0.78,
          "mean": 0.6417,
          "spread": 0.28
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.977,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7856,
          "spread": 0.377
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.998,
          "judge_a": 0.85,
          "judge_b": 0.93,
          "mean": 0.9258,
          "spread": 0.148
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.85,
          "judge_b": 0.97,
          "mean": 0.9133,
          "spread": 0.12
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.1167,
          "spread": 0.13
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.26,
          "judge_a": 0.35,
          "judge_b": 0.18,
          "mean": 0.2633,
          "spread": 0.17
        }
      }
    },
    {
      "rank": 51,
      "id": "st_20261115_25086",
      "title": "Writing a DSL for high performance parallel programming in OxCaml",
      "url": "https://blog.janestreet.com/writing-a-dsl-for-high-performance-parallel-programming-in-oxcaml/",
      "domain": "blog.janestreet.com",
      "site": "Jane Street Blog",
      "categories": [
        "programming"
      ],
      "category_names": [
        "Programming & dev tools"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T02:57:49.000Z",
      "jev": 0.519,
      "judge_a": 0.415,
      "judge_b": 0.7415,
      "score": 0.5585,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 54,
      "topic_penalty": 0.06,
      "repeat_penalty": 0,
      "summary": "Jane Street's blog describes an SPMD domain-specific language built by intern Sai Konkimalla in OxCaml, aimed at the firm's SIMD-heavy options pricing and risk code. Users write a per-element kernel once, and a runtime JIT compiler, using OxCaml's runtime metaprogramming, generates architecture-specific vectorized code, specializing branches on runtime values.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/51/",
      "card": "https://rubric.hrih.si/og/2026-11-15/51.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.49,
          "judge_a": 0.3,
          "judge_b": 0.7,
          "mean": 0.4967,
          "spread": 0.4
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.56,
          "judge_a": 0.55,
          "judge_b": 0.88,
          "mean": 0.6633,
          "spread": 0.33
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.363,
          "judge_a": 0.25,
          "judge_b": 0.45,
          "mean": 0.3544,
          "spread": 0.2
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.563,
          "judge_a": 0.4,
          "judge_b": 0.78,
          "mean": 0.5808,
          "spread": 0.38
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.747,
          "judge_a": 0.6,
          "judge_b": 0.92,
          "mean": 0.7556,
          "spread": 0.32
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.95,
          "judge_a": 0.8,
          "judge_b": 0.96,
          "mean": 0.9033,
          "spread": 0.16
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.88,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.92,
          "spread": 0.1
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.1,
          "judge_b": 0.08,
          "mean": 0.0833,
          "spread": 0.03
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.28,
          "judge_a": 0.15,
          "judge_b": 0.12,
          "mean": 0.1833,
          "spread": 0.16
        }
      }
    },
    {
      "rank": 52,
      "id": "st_20261115_25407",
      "title": "Before I Trust an Agent, I Try to Break Its Box",
      "url": "https://kunalkushwaha.github.io/2026/11/06/before-i-trust-an-agent/",
      "domain": "kunalkushwaha.github.io",
      "site": "Kunal Kushwaha",
      "categories": [
        "programming",
        "ai",
        "security",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Security",
        "Infrastructure & DevOps"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T16:14:06.000Z",
      "jev": 0.6896,
      "judge_a": 0.536,
      "judge_b": 0.8064,
      "score": 0.5773,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0.1,
      "score_rank": 50,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Kunal Kushwaha describes Koala, a tool that runs each AI agent in its own Linux VM on macOS or Linux, with network policy and API keys kept on the host. He attacks it from inside a root-compromised guest, judging results by what host-side witnesses record.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/52/",
      "card": "https://rubric.hrih.si/og/2026-11-15/52.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.563,
          "judge_a": 0.4,
          "judge_b": 0.72,
          "mean": 0.5608,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.79,
          "judge_a": 0.8,
          "judge_b": 0.93,
          "mean": 0.84,
          "spread": 0.14
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.557,
          "judge_a": 0.55,
          "judge_b": 0.62,
          "mean": 0.5756,
          "spread": 0.07
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.815,
          "judge_a": 0.6,
          "judge_b": 0.88,
          "mean": 0.765,
          "spread": 0.28
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.94,
          "judge_a": 0.72,
          "judge_b": 0.91,
          "mean": 0.8567,
          "spread": 0.22
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.968,
          "judge_a": 0.8,
          "judge_b": 0.91,
          "mean": 0.8925,
          "spread": 0.167
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.79,
          "judge_a": 0.85,
          "judge_b": 0.86,
          "mean": 0.8333,
          "spread": 0.07
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.3,
          "judge_b": 0.04,
          "mean": 0.1367,
          "spread": 0.26
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.22,
          "judge_a": 0.3,
          "judge_b": 0.08,
          "mean": 0.2,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 53,
      "id": "st_20261115_25035",
      "title": "DeepSeek V4 Pro 0813",
      "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
      "domain": "huggingface.co",
      "site": "huggingface.co",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T00:50:33.000Z",
      "jev": 0.6532,
      "judge_a": 0.6246,
      "judge_b": 0.7448,
      "score": 0.5742,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0.1,
      "score_rank": 51,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "DeepSeek has released DeepSeek-V4-Pro-0813, the official version of its V4-Pro model, which supersedes the earlier preview and adds a DSpark speculative decoding module. According to its published benchmarks, the model improves on the preview on agentic and coding tasks such as Terminal Bench 2.1 and DeepSWE, and is broadly competitive…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/53/",
      "card": "https://rubric.hrih.si/og/2026-11-15/53.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.66,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.68,
          "spread": 0.18
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.62,
          "judge_a": 0.8,
          "judge_b": 0.9,
          "mean": 0.7733,
          "spread": 0.28
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.563,
          "judge_a": 0.67,
          "judge_b": 0.72,
          "mean": 0.6511,
          "spread": 0.157
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.853,
          "judge_a": 0.7,
          "judge_b": 0.8,
          "mean": 0.7842,
          "spread": 0.153
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.937,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7722,
          "spread": 0.337
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.863,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7275,
          "spread": 0.263
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.85,
          "judge_b": 1,
          "mean": 0.92,
          "spread": 0.15
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.11,
          "judge_a": 0.2,
          "judge_b": 0.2,
          "mean": 0.17,
          "spread": 0.09
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.38,
          "judge_a": 0.3,
          "judge_b": 0.3,
          "mean": 0.3267,
          "spread": 0.08
        }
      }
    },
    {
      "rank": 54,
      "id": "st_20261115_25166",
      "title": "microsoft/onnxruntime plugin-ep-cuda/v0.4.0",
      "url": "https://github.com/microsoft/onnxruntime/releases/tag/plugin-ep-cuda%2Fv0.4.0",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai",
        "hardware"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Hardware & devices"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T06:34:19.000Z",
      "jev": 0.6482,
      "judge_a": 0.438,
      "judge_b": 0.7608,
      "score": 0.6156,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 27,
      "topic_penalty": 0.08,
      "repeat_penalty": 0.05,
      "summary": "The microsoft/onnxruntime plugin-ep-cuda v0.4.0 release, published by Microsoft's ONNX Runtime project, adds CPU/GPU hybrid expert offloading for Mixture-of-Experts models that exceed GPU memory, plus new quantized MoE kernels (INT2/INT3/INT4, offline NVFP4) and long-context and speculative-decoding speedups.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/54/",
      "card": "https://rubric.hrih.si/og/2026-11-15/54.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.668,
          "judge_a": 0.4,
          "judge_b": 0.72,
          "mean": 0.5958,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.59,
          "judge_a": 0.4,
          "judge_b": 0.78,
          "mean": 0.59,
          "spread": 0.38
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.41,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.43,
          "spread": 0.28
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.813,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.7442,
          "spread": 0.22
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.82,
          "judge_a": 0.6,
          "judge_b": 0.91,
          "mean": 0.7767,
          "spread": 0.31
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.968,
          "judge_a": 0.85,
          "judge_b": 0.98,
          "mean": 0.9325,
          "spread": 0.13
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.9,
          "judge_b": 0.96,
          "mean": 0.9267,
          "spread": 0.06
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.15,
          "judge_b": 0.04,
          "mean": 0.0867,
          "spread": 0.11
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.18,
          "judge_a": 0.2,
          "judge_b": 0.05,
          "mean": 0.1433,
          "spread": 0.15
        }
      }
    },
    {
      "rank": 55,
      "id": "st_20261115_25542",
      "title": "Loft – offline Q&A over your local files, with citations (llama.cpp)",
      "url": "https://github.com/nwaoga/loft",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T19:42:17.000Z",
      "jev": 0.5919,
      "judge_a": 0.452,
      "judge_b": 0.6454,
      "score": 0.563,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 52,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Loft is an open-source desktop app, posted as a Show HN project by its author on GitHub, that answers questions about user-chosen local folders fully offline using llama.cpp, with cited excerpts shown under each answer.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/55/",
      "card": "https://rubric.hrih.si/og/2026-11-15/55.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.433,
          "judge_a": 0.3,
          "judge_b": 0.42,
          "mean": 0.3842,
          "spread": 0.133
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.62,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.65,
          "spread": 0.23
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.357,
          "judge_a": 0.35,
          "judge_b": 0.48,
          "mean": 0.3956,
          "spread": 0.13
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.79,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.7033,
          "spread": 0.19
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.84,
          "judge_a": 0.6,
          "judge_b": 0.86,
          "mean": 0.7667,
          "spread": 0.26
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.945,
          "judge_a": 0.7,
          "judge_b": 0.88,
          "mean": 0.8417,
          "spread": 0.245
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.9,
          "judge_b": 0.97,
          "mean": 0.93,
          "spread": 0.07
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.07,
          "judge_a": 0.15,
          "judge_b": 0.03,
          "mean": 0.0833,
          "spread": 0.12
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.14,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.14,
          "spread": 0.12
        }
      }
    },
    {
      "rank": 56,
      "id": "st_20261115_25147",
      "title": "Not Everything Needs an LLM",
      "url": "https://posthog.com/newsletter/decision-models",
      "domain": "posthog.com",
      "site": "posthog.com",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T06:25:03.000Z",
      "jev": 0.6436,
      "judge_a": 0.3885,
      "judge_b": 0.6517,
      "score": 0.5612,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 53,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "PostHog's engineering team describes how they use \"decision models\" (Jev, a typed-answer model returning calibrated probabilities) instead of LLMs for narrow yes/no and choice tasks.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/56/",
      "card": "https://rubric.hrih.si/og/2026-11-15/56.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.518,
          "judge_a": 0.35,
          "judge_b": 0.62,
          "mean": 0.4958,
          "spread": 0.27
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.69,
          "judge_a": 0.6,
          "judge_b": 0.82,
          "mean": 0.7033,
          "spread": 0.22
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.713,
          "judge_a": 0.5,
          "judge_b": 0.68,
          "mean": 0.6311,
          "spread": 0.213
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.91,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.81,
          "spread": 0.21
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.5,
          "judge_a": 0.55,
          "judge_b": 0.78,
          "mean": 0.61,
          "spread": 0.28
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.8,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6933,
          "spread": 0.2
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.87,
          "judge_a": 0.6,
          "judge_b": 0.93,
          "mean": 0.8,
          "spread": 0.33
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.15,
          "judge_a": 0.6,
          "judge_b": 0.38,
          "mean": 0.3767,
          "spread": 0.45
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.3,
          "judge_a": 0.4,
          "judge_b": 0.24,
          "mean": 0.3133,
          "spread": 0.16
        }
      }
    },
    {
      "rank": 57,
      "id": "st_20261115_25267",
      "title": "Build Your Own AI Coding Agent Harness",
      "url": "https://vercel.com/academy/build-ai-agent-harness",
      "domain": "vercel.com",
      "site": "vercel.com",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Tutorial",
      "depth": "three_judges",
      "published_at": "2026-11-14T12:09:46.000Z",
      "jev": 0.6536,
      "judge_a": 0.465,
      "judge_b": 0.5551,
      "score": 0.5578,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 55,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Vercel Academy has published a course, \"Build Your Own AI Coding Agent Harness,\" that walks through building TeensyCode, a TypeScript coding agent with a tool loop, approval gates, a swappable sandbox layer (local and in-memory), context pruning, subagent delegation, and extensibility hooks.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/57/",
      "card": "https://rubric.hrih.si/og/2026-11-15/57.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.578,
          "judge_a": 0.35,
          "judge_b": 0.48,
          "mean": 0.4692,
          "spread": 0.228
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.74,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6733,
          "spread": 0.14
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.547,
          "judge_a": 0.6,
          "judge_b": 0.55,
          "mean": 0.5656,
          "spread": 0.053
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.928,
          "judge_a": 0.8,
          "judge_b": 0.8,
          "mean": 0.8425,
          "spread": 0.127
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.353,
          "judge_a": 0.45,
          "judge_b": 0.62,
          "mean": 0.4744,
          "spread": 0.267
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.823,
          "judge_a": 0.7,
          "judge_b": 0.72,
          "mean": 0.7475,
          "spread": 0.123
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.86,
          "judge_a": 0.8,
          "judge_b": 0.68,
          "mean": 0.78,
          "spread": 0.18
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.16,
          "judge_a": 0.55,
          "judge_b": 0.38,
          "mean": 0.3633,
          "spread": 0.39
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.22,
          "judge_a": 0.2,
          "judge_b": 0.12,
          "mean": 0.18,
          "spread": 0.1
        }
      }
    },
    {
      "rank": 58,
      "id": "st_20261115_25494",
      "title": "Chopper, a tiny edge model that keeps only useful text in 50 ms",
      "url": "https://github.com/wiiiktor/chopper_demo",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T18:05:32.000Z",
      "jev": 0.6331,
      "judge_a": 0.313,
      "judge_b": 0.7246,
      "score": 0.5569,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 56,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Chopper is a 0.31B-parameter encoder that reads a question and 20 retrieved passages in one pass and marks each word keep or drop, producing exact-quote fragments instead of generated summaries.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/58/",
      "card": "https://rubric.hrih.si/og/2026-11-15/58.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.55,
          "judge_a": 0.3,
          "judge_b": 0.68,
          "mean": 0.51,
          "spread": 0.38
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.64,
          "judge_a": 0.5,
          "judge_b": 0.9,
          "mean": 0.68,
          "spread": 0.4
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.577,
          "judge_a": 0.4,
          "judge_b": 0.72,
          "mean": 0.5656,
          "spread": 0.32
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.915,
          "judge_a": 0.6,
          "judge_b": 0.86,
          "mean": 0.7917,
          "spread": 0.315
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.937,
          "judge_a": 0.45,
          "judge_b": 0.82,
          "mean": 0.7356,
          "spread": 0.487
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.6,
          "judge_b": 0.84,
          "mean": 0.78,
          "spread": 0.3
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.9,
          "judge_b": 0.98,
          "mean": 0.9333,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.1,
          "judge_a": 0.55,
          "judge_b": 0.24,
          "mean": 0.2967,
          "spread": 0.45
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.55,
          "judge_a": 0.6,
          "judge_b": 0.38,
          "mean": 0.51,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 59,
      "id": "st_20261115_25463",
      "title": "Use LLM for direct robot control",
      "url": "https://arxiv.org/abs/2610.28259",
      "domain": "arxiv.org",
      "site": "arXiv",
      "categories": [
        "ai"
      ],
      "category_names": [
        "AI & machine learning"
      ],
      "format": "Paper",
      "depth": "three_judges",
      "published_at": "2026-11-14T17:55:04.000Z",
      "jev": 0.5573,
      "judge_a": 0.471,
      "judge_b": 0.6422,
      "score": 0.5568,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 57,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "A systematic study of GPT-6 Astra used directly as a closed-loop robot controller, with no specialized action policy, through a lightweight agent framework called GPT-GEAR. The authors test depth perception, in-context learning, and tactile feedback, benchmark it on five manipulation suites, and introduce LIBERO-GeoMotion, where it reaches only 11.1% zero-shot…",
      "page": "https://rubric.hrih.si/editions/2026-11-15/59/",
      "card": "https://rubric.hrih.si/og/2026-11-15/59.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.643,
          "judge_a": 0.5,
          "judge_b": 0.67,
          "mean": 0.6042,
          "spread": 0.17
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.53,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.6367,
          "spread": 0.25
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.55,
          "judge_a": 0.6,
          "judge_b": 0.42,
          "mean": 0.5233,
          "spread": 0.18
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.448,
          "judge_a": 0.45,
          "judge_b": 0.55,
          "mean": 0.4825,
          "spread": 0.103
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.923,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7678,
          "spread": 0.323
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.935,
          "judge_a": 0.7,
          "judge_b": 0.86,
          "mean": 0.8317,
          "spread": 0.235
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.92,
          "judge_a": 0.7,
          "judge_b": 0.93,
          "mean": 0.85,
          "spread": 0.23
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.08,
          "judge_a": 0.2,
          "judge_b": 0.03,
          "mean": 0.1033,
          "spread": 0.17
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.26,
          "judge_a": 0.4,
          "judge_b": 0.18,
          "mean": 0.28,
          "spread": 0.22
        }
      }
    },
    {
      "rank": 60,
      "id": "st_20261115_25190",
      "title": "Skill Recorder: Desktop app that records your on-screen work session",
      "url": "https://github.com/microsoft/skill-recorder",
      "domain": "github.com",
      "site": "GitHub",
      "categories": [
        "programming",
        "ai"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning"
      ],
      "format": "Release",
      "depth": "three_judges",
      "published_at": "2026-11-14T08:40:42.000Z",
      "jev": 0.5917,
      "judge_a": 0.4035,
      "judge_b": 0.6715,
      "score": 0.5555,
      "contested": true,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 59,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "Skill Recorder is an open-source desktop app from Microsoft that records a user's screen, clicks, window switches, optional terminal output, and narration during a task, then uses the GitHub Copilot CLI to turn that single run into a reusable SKILL.md procedure or scheduled automation for AI agents.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/60/",
      "card": "https://rubric.hrih.si/og/2026-11-15/60.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.49,
          "judge_a": 0.3,
          "judge_b": 0.58,
          "mean": 0.4567,
          "spread": 0.28
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.55,
          "judge_a": 0.55,
          "judge_b": 0.82,
          "mean": 0.64,
          "spread": 0.27
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.437,
          "judge_a": 0.45,
          "judge_b": 0.58,
          "mean": 0.4889,
          "spread": 0.143
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.785,
          "judge_a": 0.6,
          "judge_b": 0.78,
          "mean": 0.7217,
          "spread": 0.185
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.833,
          "judge_a": 0.45,
          "judge_b": 0.84,
          "mean": 0.7078,
          "spread": 0.39
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.96,
          "judge_a": 0.55,
          "judge_b": 0.83,
          "mean": 0.78,
          "spread": 0.41
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.91,
          "judge_a": 0.95,
          "judge_b": 0.94,
          "mean": 0.9333,
          "spread": 0.04
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.35,
          "judge_b": 0.18,
          "mean": 0.2067,
          "spread": 0.26
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.15,
          "judge_a": 0.3,
          "judge_b": 0.22,
          "mean": 0.2233,
          "spread": 0.15
        }
      }
    },
    {
      "rank": 61,
      "id": "st_20261115_25547",
      "title": "WhatWeRun, a database of real production AI agent stacks",
      "url": "https://whatwerun.dev/",
      "domain": "whatwerun.dev",
      "site": "WhatWeRun",
      "categories": [
        "programming",
        "ai",
        "devops"
      ],
      "category_names": [
        "Programming & dev tools",
        "AI & machine learning",
        "Infrastructure & DevOps"
      ],
      "format": "Other",
      "depth": "three_judges",
      "published_at": "2026-11-14T19:42:04.000Z",
      "jev": 0.5755,
      "judge_a": 0.415,
      "judge_b": 0.6754,
      "score": 0.5553,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 60,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "WhatWeRun is a database cataloguing the real production AI agent stacks used by companies such as Bkper, Natera, Pinterest, Corezoid, AWS, LangChain, Workday, Temporal, Dropbox, and Duolingo. Each entry names the agent's purpose, tooling, and models, drawn from public posts and documentation.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/61/",
      "card": "https://rubric.hrih.si/og/2026-11-15/61.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.502,
          "judge_a": 0.3,
          "judge_b": 0.62,
          "mean": 0.4742,
          "spread": 0.32
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.71,
          "judge_a": 0.55,
          "judge_b": 0.84,
          "mean": 0.7,
          "spread": 0.29
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.717,
          "judge_a": 0.5,
          "judge_b": 0.76,
          "mean": 0.6589,
          "spread": 0.26
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.655,
          "judge_a": 0.6,
          "judge_b": 0.72,
          "mean": 0.6583,
          "spread": 0.12
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.47,
          "judge_a": 0.35,
          "judge_b": 0.58,
          "mean": 0.4667,
          "spread": 0.23
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.555,
          "judge_a": 0.4,
          "judge_b": 0.52,
          "mean": 0.4917,
          "spread": 0.155
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.89,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.79,
          "spread": 0.19
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.15,
          "judge_a": 0.25,
          "judge_b": 0.18,
          "mean": 0.1933,
          "spread": 0.1
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.31,
          "judge_a": 0.2,
          "judge_b": 0.08,
          "mean": 0.1967,
          "spread": 0.23
        }
      }
    },
    {
      "rank": 62,
      "id": "st_20261115_25059",
      "title": "SSH Keys in the Secure Enclave, Without an Agent",
      "url": "https://lograker.app/blog/secure-enclave-ssh-keys-without-an-agent",
      "domain": "lograker.app",
      "site": "LogRaker",
      "categories": [
        "programming",
        "security"
      ],
      "category_names": [
        "Programming & dev tools",
        "Security"
      ],
      "format": "Write-up",
      "depth": "three_judges",
      "published_at": "2026-11-14T01:43:43.000Z",
      "jev": 0.574,
      "judge_a": 0.485,
      "judge_b": 0.5974,
      "score": 0.5521,
      "contested": false,
      "flag_penalty": 0,
      "freshness_penalty": 0,
      "score_rank": 61,
      "topic_penalty": 0.08,
      "repeat_penalty": 0,
      "summary": "The author describes how their Mac app LogRaker stores SSH keys in the Secure Enclave and signs logins directly through swift-nio-ssh, avoiding the need for an ssh-agent like Secretive.",
      "page": "https://rubric.hrih.si/editions/2026-11-15/62/",
      "card": "https://rubric.hrih.si/og/2026-11-15/62.png",
      "breakdown": {
        "significance": {
          "name": "Significance",
          "weight": 0.24,
          "jev": 0.453,
          "judge_a": 0.3,
          "judge_b": 0.48,
          "mean": 0.4108,
          "spread": 0.18
        },
        "reader_interest": {
          "name": "Reader interest",
          "weight": 0.26,
          "jev": 0.62,
          "judge_a": 0.7,
          "judge_b": 0.78,
          "mean": 0.7,
          "spread": 0.16
        },
        "audience_breadth": {
          "name": "Audience breadth",
          "weight": 0.13,
          "jev": 0.393,
          "judge_a": 0.35,
          "judge_b": 0.35,
          "mean": 0.3644,
          "spread": 0.043
        },
        "practical_relevance": {
          "name": "Practical relevance",
          "weight": 0.22,
          "jev": 0.778,
          "judge_a": 0.6,
          "judge_b": 0.68,
          "mean": 0.6858,
          "spread": 0.178
        },
        "substantiated": {
          "name": "Substantiated",
          "weight": 0.05,
          "jev": 0.563,
          "judge_a": 0.7,
          "judge_b": 0.82,
          "mean": 0.6944,
          "spread": 0.257
        },
        "technical_depth": {
          "name": "Technical depth",
          "weight": 0.05,
          "jev": 0.955,
          "judge_a": 0.8,
          "judge_b": 0.72,
          "mean": 0.825,
          "spread": 0.235
        },
        "new_information": {
          "name": "New information",
          "weight": 0.05,
          "jev": 0.9,
          "judge_a": 0.85,
          "judge_b": 0.93,
          "mean": 0.8933,
          "spread": 0.08
        },
        "promotional": {
          "name": "Promotional",
          "weight": -0.19,
          "jev": 0.09,
          "judge_a": 0.25,
          "judge_b": 0.16,
          "mean": 0.1667,
          "spread": 0.16
        },
        "overclaim": {
          "name": "Overclaim",
          "weight": -0.11,
          "jev": 0.2,
          "judge_a": 0.15,
          "judge_b": 0.08,
          "mean": 0.1433,
          "spread": 0.12
        }
      }
    }
  ]
}
