{
  "schema_version": "1.0",
  "url": "https://jordananderson.work/",
  "title": "Jordan Anderson — Technical & Policy Consulting",
  "description": "ML systems, GPU debugging, software security and industrial policy. See Jordan Anderson’s open-source work and discuss a technical project or policy engagement.",
  "date_modified": "2026-10-06",
  "representations": {
    "json": "https://jordananderson.work/data/pages/home.json",
    "markdown": "https://jordananderson.work/data/pages/home.md"
  },
  "headings": [
    {
      "level": 1,
      "id": "mission",
      "text": "Find the gap. Make it work."
    },
    {
      "level": 2,
      "id": "technical-title",
      "text": "The work goes all the wayto the implementation."
    },
    {
      "level": 3,
      "id": null,
      "text": "More from thehardware you own."
    },
    {
      "level": 3,
      "id": null,
      "text": "From a driver mismatchto an upstream fix."
    },
    {
      "level": 3,
      "id": null,
      "text": "Read what the proofactually proves."
    },
    {
      "level": 3,
      "id": null,
      "text": "Fix the tools the work depends on."
    },
    {
      "level": 3,
      "id": null,
      "text": "Compute on the packed weights."
    },
    {
      "level": 3,
      "id": null,
      "text": "Make recovery part of the tool."
    },
    {
      "level": 3,
      "id": null,
      "text": "Check what actually arrived."
    },
    {
      "level": 2,
      "id": "policy-title",
      "text": "The same questions,at institutional scale."
    },
    {
      "level": 3,
      "id": null,
      "text": "Friction by Other Means"
    },
    {
      "level": 3,
      "id": null,
      "text": "The Structural Failure of America’s Defense Industrial Base"
    },
    {
      "level": 3,
      "id": null,
      "text": "The Great Recentralization"
    },
    {
      "level": 2,
      "id": "contact-title",
      "text": "Start with the problemyou need to understand."
    },
    {
      "level": 3,
      "id": null,
      "text": "A focused first conversation"
    },
    {
      "level": 3,
      "id": null,
      "text": "A scope you can evaluate"
    },
    {
      "level": 3,
      "id": null,
      "text": "Something you can act on"
    }
  ],
  "links": [
    {
      "text": "Discuss an engagement",
      "url": "https://jordananderson.work/#contact"
    },
    {
      "text": "Explore the work",
      "url": "https://jordananderson.work/#technical"
    },
    {
      "text": "Technical work",
      "url": "https://jordananderson.work/#technical"
    },
    {
      "text": "Policy analysis",
      "url": "https://jordananderson.work/#policy"
    },
    {
      "text": "Explore Loggetta",
      "url": "https://github.com/pjordanandrsn/loggetta"
    },
    {
      "text": "Package & supported models",
      "url": "https://pypi.org/project/loggetta/"
    },
    {
      "text": "Read the benchmark & conditions",
      "url": "https://github.com/pjordanandrsn/experts4bit-qlora/blob/main/bench/h2h-2026-10-02/tc1/RESULTS-tc1-pos69.md"
    },
    {
      "text": "Loggetta 0.6.0",
      "url": "https://github.com/pjordanandrsn/loggetta/blob/main/CHANGELOG.md"
    },
    {
      "text": "Initial fix",
      "url": "https://github.com/openvinotoolkit/openvino/pull/35661"
    },
    {
      "text": "Shared API",
      "url": "https://github.com/openvinotoolkit/openvino/pull/35712"
    },
    {
      "text": "Regression test",
      "url": "https://github.com/openvinotoolkit/openvino/pull/36017"
    },
    {
      "text": "Published advisory",
      "url": "https://github.com/0xPolygonHermez/zisk/security/advisories/GHSA-qjgg-fcj3-p9x3"
    },
    {
      "text": "Finding & scope",
      "url": "https://cerinamroth.com/disclosures/zisk-signed-rem-sign-soundness/"
    },
    {
      "text": "bitsandbytes",
      "url": "https://github.com/bitsandbytes-foundation/bitsandbytes/pull/1999"
    },
    {
      "text": "Axolotl",
      "url": "https://github.com/axolotl-ai-cloud/axolotl/pull/3915"
    },
    {
      "text": "Transformers",
      "url": "https://github.com/huggingface/transformers/pull/47931"
    },
    {
      "text": "Unsloth Zoo",
      "url": "https://github.com/unslothai/unsloth-zoo/pull/1082"
    },
    {
      "text": "Kernel research & evidence",
      "url": "https://cerinamroth.com/ml/grouped-nf4-gemm/"
    },
    {
      "text": "Asahi encryption project",
      "url": "https://github.com/pjordanandrsn/asahi-luks-setup"
    },
    {
      "text": "Malware screening project",
      "url": "https://github.com/pjordanandrsn/arr-malware-screen"
    },
    {
      "text": "Bring a technical problem",
      "url": "https://jordananderson.work/#contact"
    },
    {
      "text": "Friction by Other Means",
      "url": "https://jordananderson.work/writing/friction-by-other-means"
    },
    {
      "text": "The Structural Failure of America’s Defense Industrial Base",
      "url": "https://jordananderson.work/writing/the-structural-failure-of-the-defense-industrial-base"
    },
    {
      "text": "The Great Recentralization",
      "url": "https://jordananderson.work/writing/the-great-recentralization"
    },
    {
      "text": "Browse all essays",
      "url": "https://jordananderson.work/writing/"
    },
    {
      "text": "jordan@jordananderson.work",
      "url": "mailto:jordan@jordananderson.work"
    }
  ],
  "structured_data": [
    {
      "@context": "https://schema.org",
      "@graph": [
        {
          "@type": "WebSite",
          "@id": "https://jordananderson.work/#website",
          "name": "Jordan Anderson",
          "url": "https://jordananderson.work/",
          "inLanguage": "en",
          "publisher": {
            "@type": "Person",
            "@id": "https://jordananderson.work/about#jordan-anderson",
            "name": "Jordan Anderson",
            "url": "https://jordananderson.work/about",
            "sameAs": [
              "https://github.com/pjordanandrsn",
              "https://huggingface.co/pjordanandrsn",
              "https://cerinamroth.com/"
            ]
          }
        },
        {
          "@type": "WebPage",
          "@id": "https://jordananderson.work/#webpage",
          "url": "https://jordananderson.work/",
          "name": "Jordan Anderson — Technical & Policy Consulting",
          "description": "ML systems, GPU debugging, software security and industrial policy. See Jordan Anderson’s open-source work and discuss a technical project or policy engagement.",
          "inLanguage": "en",
          "about": {
            "@type": "Person",
            "@id": "https://jordananderson.work/about#jordan-anderson",
            "name": "Jordan Anderson",
            "url": "https://jordananderson.work/about",
            "sameAs": [
              "https://github.com/pjordanandrsn",
              "https://huggingface.co/pjordanandrsn",
              "https://cerinamroth.com/"
            ]
          },
          "isPartOf": {
            "@id": "https://jordananderson.work/#website"
          },
          "dateModified": "2026-10-06"
        }
      ]
    }
  ],
  "content_markdown": "Independent technical & policy consulting\n\n \n\n# Find the gap.\n Make it work. \n\n \n\n \n\n \n\nI find what stops a system from working—and build the fix.\n\n \n\nML systems, GPU debugging, software security, and industrial policy. Tested work, with evidence you can inspect.\n\n \n\n [Discuss an engagement](https://jordananderson.work/#contact) [Explore the work](https://jordananderson.work/#technical) \n\n \n\n \n\n \n\n [Technical work](https://jordananderson.work/#technical)[Policy analysis](https://jordananderson.work/#policy) \n\nOpen code. Written reasoning. Evidence you can inspect.\n\n \n\n \n\n \n\n \n\n01 / Technical practice\n\n \n\n \n\n## The work goes all the way\nto the implementation.\n\n \n\nDiagnosing the failure is the beginning. I build the fix, test it on the hardware, and work it through upstream review.\n\n \n\n \n\n \n\n \n\n \n\nSelected work / Machine learning systems\n\n \n\n### More from the\nhardware you own.\n\n \n\nTrain large Mixture-of-Experts models on a single GPU. Loggetta checks the hardware, plans the run, and saves a report. One install includes the runtime and 4-bit kernels.\n\n \n\n [Explore Loggetta](https://github.com/pjordanandrsn/loggetta) [Package & supported models](https://pypi.org/project/loggetta/) \n\n \n\n \n\n \n\nMeasured runtime comparison\n\n 1.92× \n\nUnsloth's GPU time per training step, against e4b's\n\n \n\nQwen3-30B-A3B QLoRA · one RTX 5090\nBoth frameworks on one software stack, with matched adapters, initialization, and tokens, and e4b at its defaults. Held-out loss comparable. Wall-clock: 2.80× on an AMD EPYC 7713 host.\n\n \n\nThe tradeoff: Unsloth peaked lower, 24.27 against 26.16 GB. Wall-clock ratios depend on the host, because Unsloth spends more of each step on the CPU. Earlier wall-clock readings, before e4b’s current defaults: 2.352× and 2.468×.\n\n \n\n [Read the benchmark & conditions](https://github.com/pjordanandrsn/experts4bit-qlora/blob/main/bench/h2h-2026-10-02/tc1/RESULTS-tc1-pos69.md) \n\n \n\n \n\n \n\nNew in [Loggetta 0.6.0](https://github.com/pjordanandrsn/loggetta/blob/main/CHANGELOG.md): an AMD GPU is found and refused by name, and with experts4bit-qlora 0.53.0 or later the `grouped_nf4` MoE backward is priced once. The speed result above is the runtime underneath.\n\n \n\n \n\n \n\nGPU correctness / OpenVINO\n\n \n\n### From a driver mismatch\nto an upstream fix.\n\n \n\nIntel GPU kernels assumed an OpenCL operation the driver no longer provided. I traced the failure, developed the shared-header replacement, migrated affected kernels, and added a regression test for the bug class.\n\n \n\nThree merged PRs carry the repair and test. On the measured UHD P630 / TinyLlama int4 setup, the fixed GPU path decoded at 18.4 tokens/s against 13.9 on CPU fallback. Energy was not measured.\n\n \n\n [Initial fix](https://github.com/openvinotoolkit/openvino/pull/35661) [Shared API](https://github.com/openvinotoolkit/openvino/pull/35712) [Regression test](https://github.com/openvinotoolkit/openvino/pull/36017) \n\n \n\n \n\n \n\nSecurity research / ZisK zkVM\n\n \n\n### Read what the proof\nactually proves.\n\n \n\nA signed-remainder constraint checked magnitude but left the sign free for negative dividends. The ZisK team published the finding as a Critical advisory and credited me as a reporter.\n\n \n\nThe report checked constraints and witness components; it did not produce an end-to-end forged proof. The final fix came from Cody Gunton’s independent report of the sibling quotient-sign bug.\n\n \n\n [Published advisory](https://github.com/0xPolygonHermez/zisk/security/advisories/GHSA-qjgg-fcj3-p9x3) [Finding & scope](https://cerinamroth.com/disclosures/zisk-signed-rem-sign-soundness/) \n\n \n\n \n\n \n\n \n\nMore systems work · upstream fixes, kernels, and security\n\n \n\n \n\n \n\n### Fix the tools the work depends on.\n\n \n\nMerged fixes stop training caches from retaining dequantized weights, correct end-of-sequence handling, and repair MoE weight preprocessing.\n\n \n\n[bitsandbytes](https://github.com/bitsandbytes-foundation/bitsandbytes/pull/1999)[Axolotl](https://github.com/axolotl-ai-cloud/axolotl/pull/3915)[Transformers](https://github.com/huggingface/transformers/pull/47931)[Unsloth Zoo](https://github.com/unslothai/unsloth-zoo/pull/1082)\n\n \n\n \n\n \n\n### Compute on the packed weights.\n\n \n\ngrouped-nf4-gemm and experts4bit-qlora run MoE computation directly on low-bit weights. The research records performance, byte-level checks, and the experiments that failed.\n\n \n\n [Kernel research & evidence](https://cerinamroth.com/ml/grouped-nf4-gemm/) \n\n \n\n \n\n \n\n### Make recovery part of the tool.\n\n \n\nasahi-luks-setup adds disk encryption to existing Fedora Asahi installations. Hardware testing informed both the installation paths and a recovery command for interrupted reencryption.\n\n \n\n [Asahi encryption project](https://github.com/pjordanandrsn/asahi-luks-setup) \n\n \n\n \n\n \n\n### Check what actually arrived.\n\n \n\narr-malware-screen checks the files discovered after extraction, closing the gap left by queue-time screening in the *arr media stack.\n\n \n\n [Malware screening project](https://github.com/pjordanandrsn/arr-malware-screen) \n\n \n\n \n\n \n\n \n\n \n\nContract work in ML systems, GPU debugging, and software correctness.\n\n [Bring a technical problem](https://jordananderson.work/#contact) \n\n \n\n \n\n \n\n \n\n02 / Policy practice\n\n \n\n \n\n## The same questions,\nat institutional scale.\n\n \n\nAnalysis of industrial capacity, trade, and who owns the infrastructure that people depend on.\n\n \n\n \n\n \n\n \n\nTrade & regulation\n\n \n\n \n\n###  [Friction by Other Means](https://jordananderson.work/writing/friction-by-other-means) \n\n \n\nHow regulatory infrastructure can preserve the effects of a tariff regime after its legal basis falls away—and why the headline rate tells only part of the story.\n\n \n\nEssay · 2026 · ~7,600 words\n\n \n\n \n\n \n\n \n\nIndustrial capacity\n\n \n\n \n\n###  [The Structural Failure of America’s Defense Industrial Base](https://jordananderson.work/writing/the-structural-failure-of-the-defense-industrial-base) \n\n \n\nAn examination of the conflict between shareholder returns and military readiness, and the case for government-owned, contractor-operated capacity.\n\n \n\nEssay · 2026 · ~4,200 words\n\n \n\n \n\n \n\n \n\nComputing & public power\n\n \n\n \n\n###  [The Great Recentralization](https://jordananderson.work/writing/the-great-recentralization) \n\n \n\nWhat is at stake as computation moves from machines people own to infrastructure they rent—and the institutional choices that could preserve local control.\n\n \n\nEssay · 2026 · ~9,000 words\n\n \n\n \n\n \n\n \n\nI serve on the board of Oteco, Inc., a Houston industrial manufacturer founded in 1948. That experience informs the questions I bring to policy work.\n\n [Browse all essays](https://jordananderson.work/writing/) \n\n \n\n \n\n \n\n \n\n03 / Working together\n\n \n\n## Start with the problem\nyou need to understand.\n\n [jordan@jordananderson.work](mailto:jordan@jordananderson.work) \n\nTechnical projects and policy retainers for engineering teams, industrial and defense organizations, investment firms, and policy shops.\n\n \n\n  \n-  \n\n### A focused first conversation\n\n \n\nSend a short description of the problem and what you have tried. An initial conversation, usually 30–45 minutes, establishes whether the work fits.\n\n  \n-  \n\n### A scope you can evaluate\n\n \n\nScope and rates go in writing before work begins. A specific diagnosis typically takes two to six weeks; retainer work runs longer.\n\n  \n-  \n\n### Something you can act on\n\n \n\nA tested fix, an upstream contribution, or a recommendation with a written account of the evidence. The reasoning stays with you.\n",
  "content_sha256": "8260e740d168b49ed71f7e23ab09af40ab34e59f5147b27b23e291c51fccd49f"
}