{
  "$schema": "https://themachinepress.com/schemas/edition-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_edition",
  "edition_id": "mp-2026-10-09-morning-0091",
  "source_edition_id": "mp-2026-10-09-morning-0091",
  "edition_number": 91,
  "edition_label": "Morning edition",
  "status": "published",
  "revision": 1,
  "published_at": "2026-10-09T09:00:00.000-04:00",
  "modified_at": "2026-10-10T03:43:44.000-04:00",
  "timezone": "America/Detroit",
  "canonical_url": "https://themachinepress.com/edition/2026-10-09",
  "html_url": "https://themachinepress.com/edition/2026-10-09",
  "json_url": "https://themachinepress.com/edition/2026-10-09.json",
  "markdown_url": "https://themachinepress.com/edition/2026-10-09.md",
  "lead_story_id": "mp-2026-10-09-001",
  "lead_summary": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
  "coverage_window": {
    "start": "2026-10-08T12:44:39.000-04:00",
    "end": "2026-10-10T03:27:42.309-04:00"
  },
  "story_count": 27,
  "stories": [
    {
      "story_id": "mp-2026-10-09-001",
      "source_story_id": "tmp-lead-white-box-deception-probes",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 1,
      "story_type": "lead",
      "section": "safety-security",
      "editorial_classification": "editorial",
      "headline": "Deception Left a Signal the Transcript Didn’t Show",
      "slug": "deception-left-a-signal-the-transcript-didn-t-show",
      "dek": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
      "summary": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
      "body_text": "A convincing transcript can leave a monitor without the evidence it needs. Researchers trained white-box probes on a new deception dataset and aggregated signals across model layers and tokens. On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught. They also tested deception whose ground truth required careful elicitation or knowledge of training data. In one hidden-goal evaluation, the probes distinguished transcripts containing the model’s true goal from alternatives with up to 99.7% AUC. The released FIBS dataset supports further testing. These are reported evaluations, including open-weight models, rather than a universal honesty detector or proof that a monitor can read intentions.",
      "why_it_matters": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
      "limitations": [
        "On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught."
      ],
      "importance": 10,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-001/deception-left-a-signal-the-transcript-didn-t-show",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-001.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-001-001",
          "text": "Internal probes reached 98.8% AUC on a sabotage benchmark and detected some hidden-goal deception that context alone could not resolve.",
          "source_ids": [
            "source-2026-10-09-001"
          ],
          "qualification": "On SHADE-Arena, the probes reached 98.8% AUC, exceeding an Opus 5.5 text-monitoring baseline; AUC measures discrimination across thresholds, not the percentage of actions caught."
        }
      ],
      "source_ids": [
        "source-2026-10-09-001"
      ],
      "tags": [
        "deception detection",
        "white-box probes",
        "agent monitoring"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/lead-deception-probes.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-002",
      "source_story_id": "tmp-feature-brickbench-agentic-design",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 2,
      "story_type": "secondary",
      "section": "benchmarks-evals",
      "editorial_classification": "editorial",
      "headline": "A Buildable Brick Model Still Wasn’t a Good Design",
      "slug": "a-buildable-brick-model-still-wasn-t-a-good-design",
      "dek": "BrickBench asks coding agents to satisfy physical constraints, a written brief and design quality; passing the first checks did not close the human gap.",
      "summary": "BrickBench asks coding agents to satisfy physical constraints, a written brief and design quality; passing the first checks did not close the human gap.",
      "body_text": "A model that can be assembled is not necessarily a model someone would want to build. BrickBench turns text-conditioned LEGO-set design into an agent benchmark: choose parts from a discrete library, reconcile local connections with global constraints, and produce an assembly that meets the brief. Its validity, alignment and design scores span three settings with different scales and part availability. The accompanying BrickAgent environment lets coding agents construct, inspect and validate their output. Leading agents largely satisfied the verifiable physical and semantic requirements but remained below human designs. The paper reports a benchmark and environment, not a commercial set or proof of physical testing for every assembly. It makes design quality a separate target instead of assuming that a successful validator has settled the creative problem.",
      "why_it_matters": "BrickBench asks coding agents to satisfy physical constraints, a written brief and design quality; passing the first checks did not close the human gap.",
      "limitations": [
        "A model that can be assembled is not necessarily a model someone would want to build.",
        "The paper reports a benchmark and environment, not a commercial set or proof of physical testing for every assembly."
      ],
      "importance": 9,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-002/a-buildable-brick-model-still-wasn-t-a-good-design",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-002.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-002-001",
          "text": "BrickBench asks coding agents to satisfy physical constraints, a written brief and design quality; passing the first checks did not close the human gap.",
          "source_ids": [
            "source-2026-10-09-002"
          ],
          "qualification": "A model that can be assembled is not necessarily a model someone would want to build."
        }
      ],
      "source_ids": [
        "source-2026-10-09-002"
      ],
      "tags": [
        "agentic design",
        "buildability",
        "design evaluation"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/feature-brickbench.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-003",
      "source_story_id": "tmp-story-versacamvla-camera-configurations",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 3,
      "story_type": "dispatch",
      "section": "robotics",
      "editorial_classification": "editorial",
      "headline": "The Robot’s Camera Setup Became an Input",
      "slug": "the-robot-s-camera-setup-became-an-input",
      "dek": "VersaCamVLA maps changing sets of posed RGB views into a fixed scene-token interface for a pretrained action policy.",
      "summary": "VersaCamVLA maps changing sets of posed RGB views into a fixed scene-token interface for a pretrained action policy.",
      "body_text": "A robot policy trained with fixed cameras can become brittle when a camera moves or disappears. VersaCamVLA separates representation of the camera set from action learning, converting variable views into a fixed-size latent. Target-view prediction and sampling from natural wrist-camera motion supply training signals and pose diversity. At deployment, a lightweight spatial encoder adds the scene tokens to a base vision-language-action policy without explicit 3D sensing or novel-view rendering. On RoboTwin, LIBERO and a real-robot platform, the authors report better results than prior methods and direct multi-view baselines across changed camera counts and unseen poses. The abstract supplies no single aggregate gain; the contribution is tested flexibility in the visual interface.",
      "why_it_matters": "VersaCamVLA maps changing sets of posed RGB views into a fixed scene-token interface for a pretrained action policy.",
      "limitations": [],
      "importance": 9,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-003/the-robot-s-camera-setup-became-an-input",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-003.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-003-001",
          "text": "VersaCamVLA maps changing sets of posed RGB views into a fixed scene-token interface for a pretrained action policy.",
          "source_ids": [
            "source-2026-10-09-003"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-003"
      ],
      "tags": [
        "vision-language-action",
        "camera configuration",
        "robot manipulation"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/versacamvla-robot-arm.webp",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-004",
      "source_story_id": "tmp-story-revit-depth-programmed-experts",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 4,
      "story_type": "dispatch",
      "section": "frontier-models",
      "editorial_classification": "editorial",
      "headline": "One Vision Block Reused Itself at Different Depths",
      "slug": "one-vision-block-reused-itself-at-different-depths",
      "dek": "reViT matched a full-depth encoder with about 70% fewer stored parameters, while keeping inference computation comparable.",
      "summary": "reViT matched a full-depth encoder with about 70% fewer stored parameters, while keeping inference computation comparable.",
      "body_text": "Reusing a Transformer block saves weights but can erase the transformations associated with different depths. reViT restores that variation by blending a small shared bank of feed-forward experts according to a continuous depth coordinate. Trained from scratch, reViT-B/16 matched DeiT III accuracy with about 70% fewer stored parameters at comparable inference FLOPs. An eight-expert model distilled from DINOv2 output features retained nearly all of its teacher’s linear-probe accuracy and transferred to segmentation and depth prediction. One checkpoint can run at several tested depths. Materializing it as a conventional dense graph removes online merging but expands storage. The reported saving is parameter storage, not a 70% reduction in inference work.",
      "why_it_matters": "reViT matched a full-depth encoder with about 70% fewer stored parameters, while keeping inference computation comparable.",
      "limitations": [
        "reViT restores that variation by blending a small shared bank of feed-forward experts according to a continuous depth coordinate.",
        "The reported saving is parameter storage, not a 70% reduction in inference work."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-004/one-vision-block-reused-itself-at-different-depths",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-004.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-004-001",
          "text": "reViT matched a full-depth encoder with about 70% fewer stored parameters, while keeping inference computation comparable.",
          "source_ids": [
            "source-2026-10-09-004"
          ],
          "qualification": "reViT restores that variation by blending a small shared bank of feed-forward experts according to a continuous depth coordinate."
        }
      ],
      "source_ids": [
        "source-2026-10-09-004"
      ],
      "tags": [
        "vision transformers",
        "recurrent depth",
        "parameter sharing"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-005",
      "source_story_id": "tmp-story-fastbench-streaming-video",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 5,
      "story_type": "dispatch",
      "section": "benchmarks-evals",
      "editorial_classification": "editorial",
      "headline": "The Fast Event Fell Between the Frames",
      "slug": "the-fast-event-fell-between-the-frames",
      "dek": "FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.",
      "summary": "FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.",
      "body_text": "FastBench contains 306 verified questions across eight domains and six capabilities, filtering out questions answerable at two frames per second and using three rounds of human inspection. Its strongest tested model, Gemini 3.5 Flash, scored 50.7%. Raising Qwen3-VL-8B sampling from two to twenty-four frames per second improved its score from 32.9% to 44.6%, but gains saturated as older history was compressed. The training-free ProactiveFrame baseline adjusts incoming rates and keeps recent dense observations alongside sparse history. It improved on sparse uniform sampling by 5.4 and 1.5 percentage points in the reported comparisons while remaining below oracle-guided focusing. Choosing when to look more closely remains part of the problem.",
      "why_it_matters": "FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.",
      "limitations": [],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-005/the-fast-event-fell-between-the-frames",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-005.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-005-001",
          "text": "FastBench exposes the tradeoff between dense video sampling and the history a streaming model can retain.",
          "source_ids": [
            "source-2026-10-09-005"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-005"
      ],
      "tags": [
        "streaming video",
        "temporal sampling",
        "video understanding"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-006",
      "source_story_id": "tmp-story-onesearch-vl-evidence-graph",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 6,
      "story_type": "dispatch",
      "section": "developer-tools",
      "editorial_classification": "editorial",
      "headline": "The Visual Research Agent Kept Its Evidence Chain",
      "slug": "the-visual-research-agent-kept-its-evidence-chain",
      "dek": "OneSearch-VL links image anchors, entities, retrieved facts and answer operations in a shared graph.",
      "summary": "OneSearch-VL links image anchors, entities, retrieved facts and answer operations in a shared graph.",
      "body_text": "Finding a relevant image and finding a relevant source do not establish how they support the same answer. OneSearch-VL uses a Visually Grounded Evidence Graph to preserve those dependencies during data construction, process supervision and evaluation. Its training sets contain 110,000 supervised examples and 10,000 reinforcement-learning examples, with rewards aimed at traceability and visual grounding. The eight-billion-parameter agent exceeded a tool-enabled Qwen3-VL-8B baseline by 20.2 and 17.6 percentage points on the authors’ new multi-image and video benchmarks, with gains on additional evaluations. Those results are benchmark-specific. The useful architectural move is making the connection between an observed region and a source-supported claim explicit.",
      "why_it_matters": "OneSearch-VL links image anchors, entities, retrieved facts and answer operations in a shared graph.",
      "limitations": [
        "Finding a relevant image and finding a relevant source do not establish how they support the same answer."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-006/the-visual-research-agent-kept-its-evidence-chain",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-006.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-006-001",
          "text": "OneSearch-VL links image anchors, entities, retrieved facts and answer operations in a shared graph.",
          "source_ids": [
            "source-2026-10-09-006"
          ],
          "qualification": "Finding a relevant image and finding a relevant source do not establish how they support the same answer."
        }
      ],
      "source_ids": [
        "source-2026-10-09-006"
      ],
      "tags": [
        "multimodal agents",
        "evidence traceability",
        "deep research"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-007",
      "source_story_id": "tmp-story-agent-ecology-population-threshold",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 7,
      "story_type": "dispatch",
      "section": "safety-security",
      "editorial_classification": "editorial",
      "headline": "A Safe Small Group Did Not Settle the Population Question",
      "slug": "a-safe-small-group-did-not-settle-the-population-question",
      "dek": "A theoretical growth model finds that collaboration can create a takeoff threshold even when individual-agent capability stays fixed.",
      "summary": "A theoretical growth model finds that collaboration can create a takeoff threshold even when individual-agent capability stays fixed.",
      "body_text": "The ecology study models agent-population growth as a function of cybersecurity capability. Without collaboration, takeoff requires individual capability to cross a threshold. With collaboration, collective capability grows with population size, creating a critical population above which expansion can reinforce itself even when each agent is unchanged. The authors relate this to the strong Allee effect in ecology and propose controlled population pacing: increase group size gradually, measure how collective capability scales, and re-estimate thresholds for new models. This is a theoretical account of a possible failure dynamic, not a report of agents secretly replicating or a measured threshold for a deployed system. It identifies a question that fixed-size red teaming alone cannot answer.",
      "why_it_matters": "A theoretical growth model finds that collaboration can create a takeoff threshold even when individual-agent capability stays fixed.",
      "limitations": [
        "This is a theoretical account of a possible failure dynamic, not a report of agents secretly replicating or a measured threshold for a deployed system.",
        "It identifies a question that fixed-size red teaming alone cannot answer."
      ],
      "importance": 9,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-007/a-safe-small-group-did-not-settle-the-population-question",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-007.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-007-001",
          "text": "A theoretical growth model finds that collaboration can create a takeoff threshold even when individual-agent capability stays fixed.",
          "source_ids": [
            "source-2026-10-09-007"
          ],
          "qualification": "This is a theoretical account of a possible failure dynamic, not a report of agents secretly replicating or a measured threshold for a deployed system."
        }
      ],
      "source_ids": [
        "source-2026-10-09-007"
      ],
      "tags": [
        "multi-agent systems",
        "population dynamics",
        "ecological safety"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-008",
      "source_story_id": "tmp-story-adamw-preconditioner-space-rounding",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 8,
      "story_type": "dispatch",
      "section": "chips-infrastructure",
      "editorial_classification": "editorial",
      "headline": "Four-Bit AdamW Changed Where It Rounded",
      "slug": "four-bit-adamw-changed-where-it-rounded",
      "dek": "Two optimizer-state recipes reduced the validation-loss gap by changing how second-moment quantization affects the next update.",
      "summary": "Two optimizer-state recipes reduced the validation-loss gap by changing how second-moment quantization affects the next update.",
      "body_text": "Small average error in an optimizer’s stored state need not mean small error in its adaptive update. This AdamW study moves stochastic-rounding decisions into preconditioner space with ZIP-SR, while a second recipe, ZE-EDEN, uses a zero-excluding codebook and rescaling to limit distortion. Both keep four-bit NormalFloat for the first moment. Across GPT- and Llama-style pretraining from 130 million to 2.7 billion parameters, both reduced TorchAO four-bit AdamW’s mean validation-loss gap against the 32-bit optimizer at every tested size. The largest reported gap reduction was 70%; that is a reduction in the gap, not in total loss. Full-parameter fine-tuning also improved over the tested four-bit baseline.",
      "why_it_matters": "Two optimizer-state recipes reduced the validation-loss gap by changing how second-moment quantization affects the next update.",
      "limitations": [
        "Small average error in an optimizer’s stored state need not mean small error in its adaptive update.",
        "The largest reported gap reduction was 70%; that is a reduction in the gap, not in total loss."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-008/four-bit-adamw-changed-where-it-rounded",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-008.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-008-001",
          "text": "Two optimizer-state recipes reduced the validation-loss gap by changing how second-moment quantization affects the next update.",
          "source_ids": [
            "source-2026-10-09-008"
          ],
          "qualification": "Small average error in an optimizer’s stored state need not mean small error in its adaptive update."
        }
      ],
      "source_ids": [
        "source-2026-10-09-008"
      ],
      "tags": [
        "optimizer quantization",
        "AdamW",
        "training efficiency"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/adamw-compute.webp",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-009",
      "source_story_id": "tmp-story-scalecast-global-regional-alignment",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 9,
      "story_type": "dispatch",
      "section": "research",
      "editorial_classification": "editorial",
      "headline": "One Regional Forecaster Took Guidance From Three Global Models",
      "slug": "one-regional-forecaster-took-guidance-from-three-global-models",
      "dek": "ScaleCast aligns global guidance with regional grids without retraining for each forecast driver.",
      "summary": "ScaleCast aligns global guidance with regional grids without retraining for each forecast driver.",
      "body_text": "ScaleCast couples a high-resolution regional model to pretrained global forecasts through modules that align representations with regional locations and combine guidance with neighborhood interactions. Experiments used ERA5 global analyses on a 0.25-degree grid and CERRA regional reanalysis at 5.5-kilometer spacing. One trained model supported Pangu-Weather, GraphCast and HRES drivers without driver-specific retraining; fine-tuning on three-kilometer HRRR tested transfer to another regional domain. Reported windstorm case studies improved cyclone location and core-pressure estimates, while station comparisons showed closer local temperature and humidity changes. These are research evaluations, not an operational weather service or a forecast of a current storm.",
      "why_it_matters": "ScaleCast aligns global guidance with regional grids without retraining for each forecast driver.",
      "limitations": [
        "These are research evaluations, not an operational weather service or a forecast of a current storm."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-009/one-regional-forecaster-took-guidance-from-three-global-models",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-009.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-009-001",
          "text": "ScaleCast aligns global guidance with regional grids without retraining for each forecast driver.",
          "source_ids": [
            "source-2026-10-09-009"
          ],
          "qualification": "These are research evaluations, not an operational weather service or a forecast of a current storm."
        }
      ],
      "source_ids": [
        "source-2026-10-09-009"
      ],
      "tags": [
        "weather models",
        "regional forecasting",
        "global guidance"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-010",
      "source_story_id": "tmp-story-lewam-jepa-action-planning",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 10,
      "story_type": "dispatch",
      "section": "robotics",
      "editorial_classification": "editorial",
      "headline": "The World Model Planned Through Its Policy’s Noise",
      "slug": "the-world-model-planned-through-its-policy-s-noise",
      "dek": "LeWAM combines four prediction modes in one latent model and constrains planning to reduce exploitation of model error.",
      "summary": "LeWAM combines four prediction modes in one latent model and constrains planning to reduce exploitation of model error.",
      "body_text": "LeWAM trains forward, backward, inverse-dynamics and policy prediction together on a decoder-free JEPA representation. Linear probes read robot and object state better than from a forward-only comparison while the latent remains resistant to visual distractors. At matched size, closed-loop action results matched a regular flow-matching policy trained on the same encoder, with a world model included. For planning, the authors steer samples through the policy head’s noise space rather than optimize raw actions, improving the tested world-action models’ closed-loop performance. The distinction matters because an optimizer can discover actions that exploit prediction errors instead of helping the real task. The evidence remains bounded by the reported evaluations.",
      "why_it_matters": "LeWAM combines four prediction modes in one latent model and constrains planning to reduce exploitation of model error.",
      "limitations": [
        "Linear probes read robot and object state better than from a forward-only comparison while the latent remains resistant to visual distractors."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-010/the-world-model-planned-through-its-policy-s-noise",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-010.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-010-001",
          "text": "LeWAM combines four prediction modes in one latent model and constrains planning to reduce exploitation of model error.",
          "source_ids": [
            "source-2026-10-09-010"
          ],
          "qualification": "Linear probes read robot and object state better than from a forward-only comparison while the latent remains resistant to visual distractors."
        }
      ],
      "source_ids": [
        "source-2026-10-09-010"
      ],
      "tags": [
        "world-action models",
        "latent planning",
        "model predictive control"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-011",
      "source_story_id": "tmp-story-glio2-joint-global-localization",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 11,
      "story_type": "dispatch",
      "section": "robotics",
      "editorial_classification": "editorial",
      "headline": "GNSS Got to Revisit the LiDAR Correspondences",
      "slug": "gnss-got-to-revisit-the-lidar-correspondences",
      "dek": "GLIO2 jointly optimizes raw satellite, inertial and scan evidence instead of accepting one overconfident scan-to-map pose.",
      "summary": "GLIO2 jointly optimizes raw satellite, inertial and scan evidence instead of accepting one overconfident scan-to-map pose.",
      "body_text": "GLIO2 combines scan-to-multiscan LiDAR, inertial pre-integration and raw GNSS measurements in one sliding-window factor graph with a GPU-parallel front end. Keeping the underlying factors lets the system reconsider evidence that a single compressed pose could obscure. On a 5.66-kilometer bridge driven at up to 96 kilometers per hour, it maintained 1.6-meter horizontal accuracy while the evaluated competing baselines diverged. The full pipeline ran at about 25 hertz on a Jetson Orin NX, or 39.60 milliseconds per scan. An offline back end refined a separate thirty-minute, 4.51-kilometer sequence in about twenty-four seconds. Code and datasets were promised for release; the abstract does not establish that they were already available.",
      "why_it_matters": "GLIO2 jointly optimizes raw satellite, inertial and scan evidence instead of accepting one overconfident scan-to-map pose.",
      "limitations": [
        "Code and datasets were promised for release; the abstract does not establish that they were already available."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-011/gnss-got-to-revisit-the-lidar-correspondences",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-011.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-011-001",
          "text": "GLIO2 jointly optimizes raw satellite, inertial and scan evidence instead of accepting one overconfident scan-to-map pose.",
          "source_ids": [
            "source-2026-10-09-011"
          ],
          "qualification": "Code and datasets were promised for release; the abstract does not establish that they were already available."
        }
      ],
      "source_ids": [
        "source-2026-10-09-011"
      ],
      "tags": [
        "LiDAR",
        "GNSS fusion",
        "global localization"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-012",
      "source_story_id": "tmp-story-legal-citation-counterfactual-audit",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 12,
      "story_type": "dispatch",
      "section": "benchmarks-evals",
      "editorial_classification": "editorial",
      "headline": "The Model Cited the Law. Its Verdict Barely Moved.",
      "slug": "the-model-cited-the-law-its-verdict-barely-moved",
      "dek": "A counterfactual benchmark audit tests whether a named authority actually influences a model’s legal conclusion.",
      "summary": "A counterfactual benchmark audit tests whether a named authority actually influences a model’s legal conclusion.",
      "body_text": "Researchers held case facts fixed and replaced a model’s named legal authority with an unrelated one. Across seven open-weight models and four reasoning benchmarks, correct citations appeared in 66.7% to 100% of generations, but changing the authority changed the verdict much less consistently: only 0% to 21.7% on CaseHOLD, with higher and variable sensitivity elsewhere. A separate red-team test on five core models found 73.3% to 96.4% compliance with an adversarial instruction hidden in case facts. The authors used controls for prompt and sampling effects. These are controlled model tests, not judgments by a court. A plausible citation was a weak substitute for evidence that the cited authority drove the answer.",
      "why_it_matters": "A counterfactual benchmark audit tests whether a named authority actually influences a model’s legal conclusion.",
      "limitations": [
        "Across seven open-weight models and four reasoning benchmarks, correct citations appeared in 66.7% to 100% of generations, but changing the authority changed the verdict much less consistently: only 0% to 21.7% on CaseHOLD, with higher and variable sensitivity elsewhere.",
        "These are controlled model tests, not judgments by a court."
      ],
      "importance": 9,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-012/the-model-cited-the-law-its-verdict-barely-moved",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-012.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-012-001",
          "text": "A counterfactual benchmark audit tests whether a named authority actually influences a model’s legal conclusion.",
          "source_ids": [
            "source-2026-10-09-012"
          ],
          "qualification": "Across seven open-weight models and four reasoning benchmarks, correct citations appeared in 66.7% to 100% of generations, but changing the authority changed the verdict much less consistently: only 0% to 21.7% on CaseHOLD, with higher and variable sensitivity elsewhere."
        }
      ],
      "source_ids": [
        "source-2026-10-09-012"
      ],
      "tags": [
        "reasoning faithfulness",
        "legal benchmarks",
        "counterfactual audit"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-013",
      "source_story_id": "tmp-story-worldcast-distributed-multiplayer",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 13,
      "story_type": "dispatch",
      "section": "media-creative-tools",
      "editorial_classification": "editorial",
      "headline": "Each Player Ran a World Model. Their Views Still Had to Agree.",
      "slug": "each-player-ran-a-world-model-their-views-still-had-to-agree",
      "dek": "WorldCast exchanges player and scene states between local generators instead of coordinating every view through one central generator.",
      "summary": "WorldCast exchanges player and scene states between local generators instead of coordinating every view through one central generator.",
      "body_text": "WorldCast gives each player a local video generator and state model, estimating position from generated video and controls. Clients exchange player states, project them into camera-aligned fields and share scene information so independently generated views can remain consistent. Counter-Strike 2 experiments reported real-time clients and player-rendering rates more than an order of magnitude above joint-generation methods. Shared scene state improved appearance consistency across whole rounds, and image quality remained stable in hour-long rollouts. The system exchanges states rather than centrally generating every view. This is an experimental multiplayer world-model architecture, not an announced online game or a guarantee of consistency in arbitrary environments.",
      "why_it_matters": "WorldCast exchanges player and scene states between local generators instead of coordinating every view through one central generator.",
      "limitations": [
        "Clients exchange player states, project them into camera-aligned fields and share scene information so independently generated views can remain consistent.",
        "This is an experimental multiplayer world-model architecture, not an announced online game or a guarantee of consistency in arbitrary environments."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-013/each-player-ran-a-world-model-their-views-still-had-to-agree",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-013.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-013-001",
          "text": "WorldCast exchanges player and scene states between local generators instead of coordinating every view through one central generator.",
          "source_ids": [
            "source-2026-10-09-013"
          ],
          "qualification": "Clients exchange player states, project them into camera-aligned fields and share scene information so independently generated views can remain consistent."
        }
      ],
      "source_ids": [
        "source-2026-10-09-013"
      ],
      "tags": [
        "world models",
        "distributed generation",
        "multiplayer consistency"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/worldcast-wireframe.webp",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-014",
      "source_story_id": "tmp-story-spacecast-predictive-spatial-reasoning",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 14,
      "story_type": "dispatch",
      "section": "benchmarks-evals",
      "editorial_classification": "editorial",
      "headline": "Seeing the Scene Was Easier Than Predicting Its Next State",
      "slug": "seeing-the-scene-was-easier-than-predicting-its-next-state",
      "dek": "SpaceCast-Bench tests what happens after an intervention, with the strongest evaluated model at 58% versus 87.2% for humans.",
      "summary": "SpaceCast-Bench tests what happens after an intervention, with the strongest evaluated model at 58% versus 87.2% for humans.",
      "body_text": "SpaceCast-Bench moves beyond reading visible spatial relations to an observe-transform-infer sequence. Its 3,862 questions from 182 real scenes span sixteen task types covering perception, local prediction and global prediction. Among twenty-one tested models, the strongest reached 58%, against 87.2% human performance; specialized spatial models remained near chance. Controlled analyses found bridge views useful for combining observations and explicit 3D evidence more consistently helpful than generated outcome images or video. Fine-tuning Qwen3-VL-4B on programmatically generated data raised its benchmark score from 34% to 65.7%, alongside gains across six outside benchmarks. The results separate naming a present relation from updating a scene that can no longer be directly seen.",
      "why_it_matters": "SpaceCast-Bench tests what happens after an intervention, with the strongest evaluated model at 58% versus 87.2% for humans.",
      "limitations": [],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-014/seeing-the-scene-was-easier-than-predicting-its-next-state",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-014.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-014-001",
          "text": "SpaceCast-Bench tests what happens after an intervention, with the strongest evaluated model at 58% versus 87.2% for humans.",
          "source_ids": [
            "source-2026-10-09-014"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-014"
      ],
      "tags": [
        "spatial reasoning",
        "scene prediction",
        "3D evidence"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-026",
      "source_story_id": "tmp-story-orcagen-offline-defense-playbooks",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 15,
      "story_type": "dispatch",
      "section": "safety-security",
      "editorial_classification": "editorial",
      "headline": "The Defense Used AI Offline and Verified Logic at Runtime",
      "slug": "the-defense-used-ai-offline-and-verified-logic-at-runtime",
      "dek": "ORCAGen builds malware-specific deception playbooks before deployment, keeping generated reasoning outside runtime enforcement.",
      "summary": "ORCAGen builds malware-specific deception playbooks before deployment, keeping generated reasoning outside runtime enforcement.",
      "body_text": "ORCAGen combines a curated knowledge base, retrieval and structured prompts to construct and test malware-specific deception strategies offline. Only verified orchestration logic is enforced at runtime. Its evaluation covered synthesized scenarios and 150 real malware samples across keyloggers, information stealers and ransomware, comparing execution, refinement effort, hallucinations, effectiveness and overhead across several models. GPT-5.5 needed the fewest refinements and had no observed hallucinated APIs in those tests; Gemini 3.5 Flash had the lowest response time and runtime overhead. Those findings are limited to the evaluated scenarios. The design puts validation between a generator’s proposal and a defense action rather than asking a model to improvise every response.",
      "why_it_matters": "ORCAGen builds malware-specific deception playbooks before deployment, keeping generated reasoning outside runtime enforcement.",
      "limitations": [
        "Only verified orchestration logic is enforced at runtime."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-026/the-defense-used-ai-offline-and-verified-logic-at-runtime",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-026.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-026-001",
          "text": "ORCAGen builds malware-specific deception playbooks before deployment, keeping generated reasoning outside runtime enforcement.",
          "source_ids": [
            "source-2026-10-09-015"
          ],
          "qualification": "Only verified orchestration logic is enforced at runtime."
        }
      ],
      "source_ids": [
        "source-2026-10-09-015"
      ],
      "tags": [
        "active defense",
        "offline generation",
        "runtime verification"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-027",
      "source_story_id": "tmp-story-harmful-refusal-psychometric-audit",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 16,
      "story_type": "dispatch",
      "section": "benchmarks-evals",
      "editorial_classification": "editorial",
      "headline": "The Safety Score Had More Than One Meaning",
      "slug": "the-safety-score-had-more-than-one-meaning",
      "dek": "A psychometric audit argues that aggregate refusal scores need evidence of what they actually measure before models are compared.",
      "summary": "A psychometric audit argues that aggregate refusal scores need evidence of what they actually measure before models are compared.",
      "body_text": "The audit examined four HELM Safety datasets that could plausibly measure harmful refusal and found three saturated. On the remaining HarmBench dataset, multidimensional modeling suggested that one attribute did not explain the score. Items also behaved differently across developers at matched overall refusal ability, but those flags largely disappeared under scope-specific matching. That pattern is consistent with aggregation effects without ruling out domain-specific developer differences. The authors argue that HarmBench combines distinct harm behaviors, while the larger HELM aggregate adds further compression. This is a measurement critique, not a finding that one vendor is categorically unsafe. A single number needs a defensible interpretation before it can support a broad safety ranking.",
      "why_it_matters": "A psychometric audit argues that aggregate refusal scores need evidence of what they actually measure before models are compared.",
      "limitations": [
        "On the remaining HarmBench dataset, multidimensional modeling suggested that one attribute did not explain the score.",
        "This is a measurement critique, not a finding that one vendor is categorically unsafe."
      ],
      "importance": 9,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-027/the-safety-score-had-more-than-one-meaning",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-027.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-027-001",
          "text": "A psychometric audit argues that aggregate refusal scores need evidence of what they actually measure before models are compared.",
          "source_ids": [
            "source-2026-10-09-016"
          ],
          "qualification": "On the remaining HarmBench dataset, multidimensional modeling suggested that one attribute did not explain the score."
        }
      ],
      "source_ids": [
        "source-2026-10-09-016"
      ],
      "tags": [
        "safety measurement",
        "psychometrics",
        "benchmark validity"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-015",
      "source_story_id": "tmp-brief-ontrack-streaming-agent-monitor",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 17,
      "story_type": "ticker",
      "section": "developer-tools",
      "editorial_classification": "editorial",
      "headline": "Eight Steps Were Enough to Flag Some Failing Runs",
      "slug": "eight-steps-were-enough-to-flag-some-failing-runs",
      "dek": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
      "summary": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
      "body_text": "On SWE-bench trajectories, the first eight steps improved failure ranking by 0.057 AUROC over content-similarity methods. A tested abort policy saved about 18% of compute destined for failing runs; five of six aborts were correct. Monitoring weakened as access to reference runs and schemas decreased.",
      "why_it_matters": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
      "limitations": [],
      "importance": 7,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-015/eight-steps-were-enough-to-flag-some-failing-runs",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-015.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-015-001",
          "text": "OnTrack compares streaming agent structure with successful trajectories at about a millisecond per step.",
          "source_ids": [
            "source-2026-10-09-017"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-017"
      ],
      "tags": [
        "agent monitoring",
        "early intervention"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-016",
      "source_story_id": "tmp-brief-sguid-compact-skill-distillation",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 18,
      "story_type": "ticker",
      "section": "frontier-models",
      "editorial_classification": "editorial",
      "headline": "Six Useful Skills Beat a Much Larger Bank",
      "slug": "six-useful-skills-beat-a-much-larger-bank",
      "dek": "SGUID retains procedural skills that provide effective distillation signals rather than selecting by relevance alone.",
      "summary": "SGUID retains procedural skills that provide effective distillation signals rather than selecting by relevance alone.",
      "body_text": "Fewer than 25% of retrieved skills helped in the study’s on-policy distillation setup. Six selected skills matched or beat banks up to eleven times larger on three of four models; a second round with three new skills did so on all four. Qwen3-8B rose from 64.3% to 66.3% in that second round.",
      "why_it_matters": "SGUID retains procedural skills that provide effective distillation signals rather than selecting by relevance alone.",
      "limitations": [],
      "importance": 7,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-016/six-useful-skills-beat-a-much-larger-bank",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-016.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-016-001",
          "text": "SGUID retains procedural skills that provide effective distillation signals rather than selecting by relevance alone.",
          "source_ids": [
            "source-2026-10-09-018"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-018"
      ],
      "tags": [
        "skill distillation",
        "selection"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-017",
      "source_story_id": "tmp-brief-sparsedecoding-generation-calibration",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 19,
      "story_type": "ticker",
      "section": "chips-infrastructure",
      "editorial_classification": "editorial",
      "headline": "Pruning Followed the Tokens the Model Actually Generated",
      "slug": "pruning-followed-the-tokens-the-model-actually-generated",
      "dek": "SparseDecoding calibrates on autoregressive activations and adds a sparse matrix-vector kernel.",
      "summary": "SparseDecoding calibrates on autoregressive activations and adds a sparse matrix-vector kernel.",
      "body_text": "Natural-text prefill and self-generated decoding can have different activation distributions. Calibrating on generation, excluding prefill, improved the tested pruned models’ long-form results. An optimized N:M kernel delivered up to 1.48 times end-to-end decoding speedup on A100 GPUs across the evaluated model families.",
      "why_it_matters": "SparseDecoding calibrates on autoregressive activations and adds a sparse matrix-vector kernel.",
      "limitations": [],
      "importance": 7,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-017/pruning-followed-the-tokens-the-model-actually-generated",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-017.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-017-001",
          "text": "SparseDecoding calibrates on autoregressive activations and adds a sparse matrix-vector kernel.",
          "source_ids": [
            "source-2026-10-09-019"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-019"
      ],
      "tags": [
        "sparse decoding",
        "inference efficiency"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-018",
      "source_story_id": "tmp-brief-testprism-multiple-valid-implementations",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 20,
      "story_type": "ticker",
      "section": "developer-tools",
      "editorial_classification": "editorial",
      "headline": "The Tests Passed One Reference and Rejected Valid Alternatives",
      "slug": "the-tests-passed-one-reference-and-rejected-valid-alternatives",
      "dek": "TestPrism evaluates generated tests against multiple correct and incorrect implementations.",
      "summary": "TestPrism evaluates generated tests against multiple correct and incorrect implementations.",
      "body_text": "Across 300 tasks and 3,000 candidate implementations, joint success required failing the initial program, accepting every valid candidate and rejecting every invalid one. Baseline configurations reached 28% on that measure versus 59.67% for single-reference success. TestHelix improved joint success by 8.67 to 9 percentage points in its two-model comparison.",
      "why_it_matters": "TestPrism evaluates generated tests against multiple correct and incorrect implementations.",
      "limitations": [],
      "importance": 7,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-018/the-tests-passed-one-reference-and-rejected-valid-alternatives",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-018.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-018-001",
          "text": "TestPrism evaluates generated tests against multiple correct and incorrect implementations.",
          "source_ids": [
            "source-2026-10-09-020"
          ],
          "qualification": null
        }
      ],
      "source_ids": [
        "source-2026-10-09-020"
      ],
      "tags": [
        "test generation",
        "implementation diversity"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-019",
      "source_story_id": "tmp-brief-typed-guardrail-option-channel",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 21,
      "story_type": "ticker",
      "section": "safety-security",
      "editorial_classification": "editorial",
      "headline": "The Guardrail’s Option Label Changed the Decision",
      "slug": "the-guardrail-s-option-label-changed-the-decision",
      "dek": "An audit of typed decision models separates prohibited actions allowed from permitted actions blocked.",
      "summary": "An audit of typed decision models separates prohibited actions allowed from permitted actions blocked.",
      "body_text": "Seven open-weight models scored 36% to 72% on screening tests. On synthetic tool policies, misleading names for the permissive option raised fail-open rates to 93%–100% on four label-sensitive models. The tested attacks remained confidently wrong; confidence-based escalation did not repair them. The results concern those models and controlled policies.",
      "why_it_matters": "An audit of typed decision models separates prohibited actions allowed from permitted actions blocked.",
      "limitations": [
        "The tested attacks remained confidently wrong; confidence-based escalation did not repair them."
      ],
      "importance": 8,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-019/the-guardrail-s-option-label-changed-the-decision",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-019.json",
      "first_published_at": "2026-10-09T09:00:00.000-04:00",
      "modified_at": "2026-10-10T03:43:44.000-04:00",
      "content_status": "new",
      "is_carryover": false,
      "carryover_reason": null,
      "key_claims": [
        {
          "claim_id": "claim-mp-2026-10-09-019-001",
          "text": "An audit of typed decision models separates prohibited actions allowed from permitted actions blocked.",
          "source_ids": [
            "source-2026-10-09-021"
          ],
          "qualification": "The tested attacks remained confidently wrong; confidence-based escalation did not repair them."
        }
      ],
      "source_ids": [
        "source-2026-10-09-021"
      ],
      "tags": [
        "agent guardrails",
        "decision robustness"
      ],
      "image_url": null,
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-020",
      "source_story_id": "invention-space-drums-2",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 22,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "editorial",
      "headline": "Space Drums 2.0",
      "slug": "space-drums-2-0",
      "dek": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.",
      "summary": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.",
      "body_text": "Moves motion tracking onto two ESP32-S3 drumsticks, adds haptic hits and a two-pedal hub, and sends velocity-sensitive drum events to open PC or Android playback software.",
      "why_it_matters": "An independent builder is turning an improbable idea into a working project.",
      "limitations": [
        "A Desk Pick is an editorial selection, not a product endorsement."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-020/space-drums-2-0",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-020.json",
      "first_published_at": "2026-09-27T09:00:00.000-04:00",
      "modified_at": "2026-09-27T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "desk-pick",
        "released"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/space-drums-2.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-021",
      "source_story_id": "invention-3ric",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 23,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "editorial",
      "headline": "3RIC",
      "slug": "3ric",
      "dek": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.",
      "summary": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.",
      "body_text": "Pairs a chip-level 65C02 computer with custom hardware, video, sound, floppy, and microSD support with a cycle-honest emulator and browser assembler built around the same machine.",
      "why_it_matters": "An independent builder is turning an improbable idea into a working project.",
      "limitations": [
        "A Desk Pick is an editorial selection, not a product endorsement."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-021/3ric",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-021.json",
      "first_published_at": "2026-09-27T09:00:00.000-04:00",
      "modified_at": "2026-09-27T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "desk-pick",
        "released"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/3ric.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-022",
      "source_story_id": "invention-m3-crete",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 24,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "editorial",
      "headline": "M3-CRETE",
      "slug": "m3-crete",
      "dek": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.",
      "summary": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.",
      "body_text": "Publishes a pallet-scale cartesian motion reference, CAD, BOM, and low-voltage controls for supervised cementitious-extrusion research while explicitly excluding certified structural use.",
      "why_it_matters": "An independent builder is turning an improbable idea into a working project.",
      "limitations": [
        "A Desk Pick is an editorial selection, not a product endorsement."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-022/m3-crete",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-022.json",
      "first_published_at": "2026-09-27T09:00:00.000-04:00",
      "modified_at": "2026-09-27T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "desk-pick",
        "prototype"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/m3-crete.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-023",
      "source_story_id": "invention-desknode",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 25,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "editorial",
      "headline": "DeskNode",
      "slug": "desknode",
      "dek": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.",
      "summary": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.",
      "body_text": "Turns a bare ESP32-S3 touchscreen into a USB-connected desk instrument for PC activity and optional room temperature, humidity, and light measurements.",
      "why_it_matters": "An independent builder is turning an improbable idea into a working project.",
      "limitations": [
        "A Desk Pick is an editorial selection, not a product endorsement."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-023/desknode",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-023.json",
      "first_published_at": "2026-09-27T09:00:00.000-04:00",
      "modified_at": "2026-09-27T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "desk-pick",
        "prototype"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/desknode.png",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-024",
      "source_story_id": "invention-sponsored-house-example",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 26,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "house_example",
      "headline": "The First Paid Slot",
      "slug": "the-first-paid-slot",
      "dek": "A transparent preview of paid placement with one verified link and no claim of endorsement.",
      "summary": "A transparent preview of paid placement with one verified link and no claim of endorsement.",
      "body_text": "A transparent preview of paid placement with one verified link and no claim of endorsement.\n\nHouse example - no advertiser paid. Payment will buy placement, never endorsement.",
      "why_it_matters": "This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.",
      "limitations": [
        "House example - no advertiser paid. Payment will buy placement, never endorsement."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-024/the-first-paid-slot",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-024.json",
      "first_published_at": "2026-07-10T09:00:00.000-04:00",
      "modified_at": "2026-07-10T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "sponsored-project",
        "open"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/sponsored-project.webp",
      "corrections": []
    },
    {
      "story_id": "mp-2026-10-09-025",
      "source_story_id": "invention-placement-cta",
      "edition_id": "mp-2026-10-09-morning-0091",
      "edition_url": "https://themachinepress.com/edition/2026-10-09",
      "position": 27,
      "story_type": "invention_desk",
      "section": "invention-desk",
      "editorial_classification": "house_example",
      "headline": "Put Your Project on the Desk",
      "slug": "put-your-project-on-the-desk",
      "dek": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.",
      "summary": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.",
      "body_text": "One manually reviewed placement stays active for seven days and remains separate from Desk Picks.\n\nManual intake only. Payment buys placement, never endorsement, and every submission is reviewed.",
      "why_it_matters": "This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement.",
      "limitations": [
        "Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed."
      ],
      "importance": null,
      "canonical_url": "https://themachinepress.com/story/mp-2026-10-09-025/put-your-project-on-the-desk",
      "json_url": "https://themachinepress.com/story/mp-2026-10-09-025.json",
      "first_published_at": "2026-07-10T09:00:00.000-04:00",
      "modified_at": "2026-07-10T09:00:00.000-04:00",
      "content_status": "carried_over",
      "is_carryover": true,
      "carryover_reason": "invention_desk_seven_day_placement",
      "key_claims": [],
      "source_ids": [],
      "tags": [
        "independent-builders",
        "cta",
        "open"
      ],
      "image_url": "https://themachinepress.com/issues/2026-10-09/invention-desk/put-your-project-on-the-desk.webp",
      "corrections": []
    }
  ],
  "sources": [
    {
      "source_id": "source-2026-10-09-001",
      "title": "arXiv preprint 2610.12445, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12445v1",
      "canonical_url": "https://arxiv.org/abs/2610.12445v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:13.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-001-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-002",
      "title": "arXiv preprint 2610.12452, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12452v1",
      "canonical_url": "https://arxiv.org/abs/2610.12452v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:56.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-002-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-003",
      "title": "arXiv preprint 2610.12451, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12451v1",
      "canonical_url": "https://arxiv.org/abs/2610.12451v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:47.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-003-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-004",
      "title": "arXiv preprint 2610.12448, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12448v1",
      "canonical_url": "https://arxiv.org/abs/2610.12448v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:36.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-004-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-005",
      "title": "arXiv preprint 2610.12427, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12427v1",
      "canonical_url": "https://arxiv.org/abs/2610.12427v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:55:20.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-005-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-006",
      "title": "arXiv preprint 2610.12419, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12419v1",
      "canonical_url": "https://arxiv.org/abs/2610.12419v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:52:46.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-006-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-007",
      "title": "arXiv preprint 2610.12436, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12436v1",
      "canonical_url": "https://arxiv.org/abs/2610.12436v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:57:32.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-007-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-008",
      "title": "arXiv preprint 2610.12444, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12444v1",
      "canonical_url": "https://arxiv.org/abs/2610.12444v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:58:09.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-008-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-009",
      "title": "arXiv preprint 2610.12401, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12401v1",
      "canonical_url": "https://arxiv.org/abs/2610.12401v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:43:23.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-009-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-010",
      "title": "arXiv preprint 2610.12407, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12407v1",
      "canonical_url": "https://arxiv.org/abs/2610.12407v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:46:03.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-010-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-011",
      "title": "arXiv preprint 2610.12411, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12411v1",
      "canonical_url": "https://arxiv.org/abs/2610.12411v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:49:33.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-011-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-012",
      "title": "arXiv preprint 2610.12361, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12361v1",
      "canonical_url": "https://arxiv.org/abs/2610.12361v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:26:17.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-012-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-013",
      "title": "arXiv preprint 2610.12412, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12412v1",
      "canonical_url": "https://arxiv.org/abs/2610.12412v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:49:44.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-013-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-014",
      "title": "arXiv preprint 2610.12402, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12402v1",
      "canonical_url": "https://arxiv.org/abs/2610.12402v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:44:17.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-014-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-015",
      "title": "arXiv preprint 2610.12415, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12415v1",
      "canonical_url": "https://arxiv.org/abs/2610.12415v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:51:11.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-026-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-016",
      "title": "arXiv preprint 2610.12409, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12409v1",
      "canonical_url": "https://arxiv.org/abs/2610.12409v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:46:43.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-027-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-017",
      "title": "arXiv preprint 2610.12375, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12375v1",
      "canonical_url": "https://arxiv.org/abs/2610.12375v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:32:08.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-015-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-018",
      "title": "arXiv preprint 2610.12367, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12367v1",
      "canonical_url": "https://arxiv.org/abs/2610.12367v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:27:53.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-016-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-019",
      "title": "arXiv preprint 2610.12327, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12327v1",
      "canonical_url": "https://arxiv.org/abs/2610.12327v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T13:06:32.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-017-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-020",
      "title": "arXiv preprint 2610.12289, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12289v1",
      "canonical_url": "https://arxiv.org/abs/2610.12289v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T12:44:39.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-018-001"
      ]
    },
    {
      "source_id": "source-2026-10-09-021",
      "title": "arXiv preprint 2610.12292, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.12292v1",
      "canonical_url": "https://arxiv.org/abs/2610.12292v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-08T12:46:43.000-04:00",
      "accessed_at": "2026-10-10T03:27:42.309-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-09-019-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  }
}