--- schema_version: "1.0.0" edition_id: "mp-2026-10-17-morning-0099" published_at: "2026-10-17T09:00:00.000-04:00" modified_at: "2026-10-17T09:00:00.000-04:00" canonical_url: "https://themachinepress.com/edition/2026-10-17" story_count: 27 lead_story_id: "mp-2026-10-17-001" --- # The Machine Press — Morning edition Edition ID: `mp-2026-10-17-morning-0099` Published: 2026-10-17T09:00:00.000-04:00 Canonical edition: https://themachinepress.com/edition/2026-10-17 3DVerifierBench finds a 21.6-point gap between people and the strongest tested model—and shows why one plausible projection is not enough. ## 1. The Object Looks Right Until the Other Views Arrive {#mp-2026-10-17-001} - Story ID: `mp-2026-10-17-001` - Type: `lead` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-001/the-object-looks-right-until-the-other-views-arrive **Dek:** 3DVerifierBench finds a 21.6-point gap between people and the strongest tested model—and shows why one plausible projection is not enough. A generated 3D object can look convincing from one camera while violating a requirement elsewhere. 3DVerifierBench tests that problem across ten tasks, scoring acceptance decisions, rejection explanations and responses to controlled changes in both specifications and candidate objects. In the authors' human study, the strongest evaluated model, GPT-6 Astra, trailed people by 21.6 percentage points. Errors concentrated around exact values, binding a requirement to the right target and deciding whether an apparent mismatch was merely projection-dependent. The team also built 14,126 training examples with contradiction explanations and linked counterfactuals. In human-labeled pools of sixteen candidates, an oracle choice raised complete factual success from 26.7 to 90 percent. Those are benchmark results, not proof that every 3D generator fails. They do show that visual plausibility and requirement-level correctness are separate jobs—and that reliable verification must see the whole object. ### Why it matters {#why-it-matters-mp-2026-10-17-001} 3DVerifierBench finds a 21.6-point gap between people and the strongest tested model—and shows why one plausible projection is not enough. ### Limits and context {#limitations-mp-2026-10-17-001} - Those are benchmark results, not proof that every 3D generator fails. ### Claims and sources {#claims-mp-2026-10-17-001} - 3DVerifierBench finds a 21.6-point gap between people and the strongest tested model—and shows why one plausible projection is not enough. [source-2026-10-17-001] — Qualification: Those are benchmark results, not proof that every 3D generator fails. ## 2. The Safety Guard Notices the Wrapper {#mp-2026-10-17-002} - Story ID: `mp-2026-10-17-002` - Type: `secondary` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-002/the-safety-guard-notices-the-wrapper **Dek:** A three-level audit finds that command length, tool names and formatting markers can change both guard decisions and reported benchmark scores. A safety benchmark can accidentally reward a guard for recognizing the way a risky action is written instead of the risk itself. The Shortcut Sensitivity study separates three questions: whether labels can be predicted from superficial cues, whether guard decisions correlate with those cues and whether controlled rewrites actually change a decision. In LinuxArena, shorter average shell commands strongly predicted trajectory labels. Under domain holdout, an exact-tool lookup baseline on AgentDojo-Traj fell from an F1 of 0.901 to 0.495. On a fixed safe-row cohort, rewriting an information marker lowered blocking by 8.33 percentage points, while replacing tool names with opaque aliases raised it by 11.46 points; all unsafe calls in that screen remained blocked. The paired effects appeared for the tested Haiku configuration but not Luna or deterministic Q8 TS-Guard. The result is not a universal ranking of guards. It is evidence that representation shortcuts can alter both behavior and the score meant to measure safety. ### Why it matters {#why-it-matters-mp-2026-10-17-002} A three-level audit finds that command length, tool names and formatting markers can change both guard decisions and reported benchmark scores. ### Limits and context {#limitations-mp-2026-10-17-002} - The paired effects appeared for the tested Haiku configuration but not Luna or deterministic Q8 TS-Guard. - The result is not a universal ranking of guards. ### Claims and sources {#claims-mp-2026-10-17-002} - A three-level audit finds that command length, tool names and formatting markers can change both guard decisions and reported benchmark scores. [source-2026-10-17-002] — Qualification: The paired effects appeared for the tested Haiku configuration but not Luna or deterministic Q8 TS-Guard. ## 3. The Robot Remembers How Close It Was {#mp-2026-10-17-003} - Story ID: `mp-2026-10-17-003` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-003/the-robot-remembers-how-close-it-was **Dek:** Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice. Partial observability can leave a robot unsure whether a long task is advancing or repeating. Value as Memory caches recent value predictions as a compact progress signal and combines offline demonstrations, human interventions and autonomous rollouts for online reinforcement learning. On the authors' deformable-object and contact-rich tasks, value caching produced a 29.4-point mean absolute success-rate gain, while online learning added 32.8 points over matched offline training. Concurrent collection and training also completed eight times as many policy iterations under the same gradient-step budget. These results cover the reported tasks and hardware, not general household autonomy. ### Why it matters {#why-it-matters-mp-2026-10-17-003} Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice. ### Limits and context {#limitations-mp-2026-10-17-003} - These results cover the reported tasks and hardware, not general household autonomy. ### Claims and sources {#claims-mp-2026-10-17-003} - Value as Memory feeds recent value predictions back into a general policy, then updates it during real-world practice. [source-2026-10-17-003] — Qualification: These results cover the reported tasks and hardware, not general household autonomy. ## 4. One Trial Teaches the World Model What the Camera Cannot See {#mp-2026-10-17-004} - Story ID: `mp-2026-10-17-004` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-004/one-trial-teaches-the-world-model-what-the-camera-cannot-see **Dek:** T2-Z adapts a compact environment latent at test time to represent friction, mass and other hidden conditions. A world model can observe an object without seeing the friction, mass or local rule that governs what happens next. T2-Z keeps the model frozen at deployment and optimizes a newly initialized environment latent from one or more interactions. Across seven simulated settings and four real robot tasks, the authors report object-prediction error falling from 15.20 to 7.00 and action success rising from 33.73 to 72.96 percent. Some useful latent structure emerged from a single trajectory. The evaluation does not establish that the learned coordinates correspond cleanly to every physical property, but it offers a compact route to test-time adaptation. ### Why it matters {#why-it-matters-mp-2026-10-17-004} T2-Z adapts a compact environment latent at test time to represent friction, mass and other hidden conditions. ### Limits and context {#limitations-mp-2026-10-17-004} - The evaluation does not establish that the learned coordinates correspond cleanly to every physical property, but it offers a compact route to test-time adaptation. ### Claims and sources {#claims-mp-2026-10-17-004} - T2-Z adapts a compact environment latent at test time to represent friction, mass and other hidden conditions. [source-2026-10-17-004] — Qualification: The evaluation does not establish that the learned coordinates correspond cleanly to every physical property, but it offers a compact route to test-time adaptation. ## 5. The Simulator Learns Which Failures to Rebuild {#mp-2026-10-17-005} - Story ID: `mp-2026-10-17-005` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-005/the-simulator-learns-which-failures-to-rebuild **Dek:** CoDA shares a memory between the data engine and robot policy so new environments target diagnosed weaknesses. Physically valid simulated scenes are not necessarily useful training examples. CoDA stores environment programs, policy failures and learning progress in one heterogeneous memory. A vision-language critic turns validated diagnoses into new generation programs, while staged reward potentials supervise policy updates. Across four manipulation families, the authors report out-of-distribution success rising from 5.3 percent after supervised fine-tuning to 27.6 percent, 9.5 points above the strongest closed-loop baseline at the same generation budget. On LIBERO-Pro, fifteen rounds produced larger gains under position and instruction perturbations. The measurements remain simulator- and benchmark-specific. ### Why it matters {#why-it-matters-mp-2026-10-17-005} CoDA shares a memory between the data engine and robot policy so new environments target diagnosed weaknesses. ### Limits and context {#limitations-mp-2026-10-17-005} - Physically valid simulated scenes are not necessarily useful training examples. - The measurements remain simulator- and benchmark-specific. ### Claims and sources {#claims-mp-2026-10-17-005} - CoDA shares a memory between the data engine and robot policy so new environments target diagnosed weaknesses. [source-2026-10-17-005] — Qualification: Physically valid simulated scenes are not necessarily useful training examples. ## 6. The Forecast Keeps the Futures Together {#mp-2026-10-17-006} - Story ID: `mp-2026-10-17-006` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-006/the-forecast-keeps-the-futures-together **Dek:** PathFM generates complete multivariate paths instead of predicting each future quantile in isolation. Marginal forecasts can score well while missing whether peaks, totals and event times cohere along one plausible trajectory. PathFM uses flow matching to generate complete future paths, distills sampling to one or two steps and shares the history-attention cache across draws. The authors report competitive marginal accuracy on fev-bench and TIME, plus the best aggregate path score among their external baselines on a new benchmark that adds energy, variogram and downstream metrics. In chaotic-system tests, the flow model preserved long-run variability where comparison systems collapsed. The new benchmark and model are author-defined; the result does not settle forecasting across every operational domain. ### Why it matters {#why-it-matters-mp-2026-10-17-006} PathFM generates complete multivariate paths instead of predicting each future quantile in isolation. ### Limits and context {#limitations-mp-2026-10-17-006} - The new benchmark and model are author-defined; the result does not settle forecasting across every operational domain. ### Claims and sources {#claims-mp-2026-10-17-006} - PathFM generates complete multivariate paths instead of predicting each future quantile in isolation. [source-2026-10-17-006] — Qualification: The new benchmark and model are author-defined; the result does not settle forecasting across every operational domain. ## 7. The Theorem Prover Follows the Library's Dependency Trails {#mp-2026-10-17-007} - Story ID: `mp-2026-10-17-007` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-007/the-theorem-prover-follows-the-library-s-dependency-trails **Dek:** LeanGraphSearch propagates relevance through Mathlib's proof graph before reranking lemmas against the target. A useful lemma may be several proof dependencies away from anything that looks lexically similar to the theorem at hand. LeanGraphSearch builds a graph from Mathlib's forward dependencies, spreads relevance from semantically matched declarations and reranks the expanded candidates. Across two retrieval benchmarks, the authors report recall gains of up to 12 percent over LeanSearch v2 and coverage of additional complete proof routes. In iterative Lean-guided proving, pass rates improved by as much as 40 percent over the vanilla prover and at most 10 percent over LeanSearch v2 at matched reflection-turn budgets. The gains are benchmark-bound, but they connect better retrieval to machine-checked downstream proofs. ### Why it matters {#why-it-matters-mp-2026-10-17-007} LeanGraphSearch propagates relevance through Mathlib's proof graph before reranking lemmas against the target. ### Limits and context {#limitations-mp-2026-10-17-007} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-17-007} - LeanGraphSearch propagates relevance through Mathlib's proof graph before reranking lemmas against the target. [source-2026-10-17-007] ## 8. The Next Idea Is Rewarded for Not Looking Like the Last One {#mp-2026-10-17-008} - Story ID: `mp-2026-10-17-008` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-008/the-next-idea-is-rewarded-for-not-looking-like-the-last-one **Dek:** A set-level objective trains one model to generate grounded scientific proposals that remain distinct across the batch. Research-idea systems usually optimize each suggestion alone, which can produce a polished set of near-duplicates. This work scores a generated set jointly, weighting each idea by both quality and distance from ideas already proposed. On a contamination-free benchmark, the authors' 8-billion-parameter model achieved higher creative utility than the tested Claude Opus 4.8 and GPT-6 Astra configurations. Four experts judged its sets about as promising as Claude's and marginally more unique; coding-agent proof-of-concept runs kept comparable success-at-k. The pilot is small and does not establish scientific novelty or validity, but it makes diversity an explicit training target rather than a prompt request. ### Why it matters {#why-it-matters-mp-2026-10-17-008} A set-level objective trains one model to generate grounded scientific proposals that remain distinct across the batch. ### Limits and context {#limitations-mp-2026-10-17-008} - The pilot is small and does not establish scientific novelty or validity, but it makes diversity an explicit training target rather than a prompt request. ### Claims and sources {#claims-mp-2026-10-17-008} - A set-level objective trains one model to generate grounded scientific proposals that remain distinct across the batch. [source-2026-10-17-008] — Qualification: The pilot is small and does not establish scientific novelty or validity, but it makes diversity an explicit training target rather than a prompt request. ## 9. The Swarm Learns Who Should Speak {#mp-2026-10-17-009} - Story ID: `mp-2026-10-17-009` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-009/the-swarm-learns-who-should-speak **Dek:** Market-based post-training lets decentralized agents bid for participation and learn coordination with their individual policies. Fixed orchestration can separate what each agent learns from how the group organizes itself. Market-based post-training lets agents bid to participate, uses those bids to determine who acts and converts the collective outcome into local training credit. The authors report that learned self-orchestration supports larger populations at inference time and loses less performance when agents drop out. The paper evaluates controlled multi-agent tasks rather than an open-world organization, so bidding is a training mechanism, not evidence that markets are universally optimal. Its useful move is to make coordination structure itself learnable. ### Why it matters {#why-it-matters-mp-2026-10-17-009} Market-based post-training lets decentralized agents bid for participation and learn coordination with their individual policies. ### Limits and context {#limitations-mp-2026-10-17-009} - The paper evaluates controlled multi-agent tasks rather than an open-world organization, so bidding is a training mechanism, not evidence that markets are universally optimal. ### Claims and sources {#claims-mp-2026-10-17-009} - Market-based post-training lets decentralized agents bid for participation and learn coordination with their individual policies. [source-2026-10-17-009] — Qualification: The paper evaluates controlled multi-agent tasks rather than an open-world organization, so bidding is a training mechanism, not evidence that markets are universally optimal. ## 10. The Second Opinion Improves When the Roles Reconcile {#mp-2026-10-17-010} - Story ID: `mp-2026-10-17-010` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-010/the-second-opinion-improves-when-the-roles-reconcile **Dek:** AI Consilium evaluates 3,040 reports across grounding, coherence, management and multidisciplinary synthesis. A plausible differential diagnosis is not enough for a useful clinical second opinion. AI Consilium uses 300 resolved NEJM–MedQA cases, with 110 for physician reference ratings and judge calibration and 190 held out. Eight pipelines produced independent role-specific opinions and interactive multidisciplinary reports; a physician-calibrated model judge scored 3,040 final reports and their structured primary-role outputs. Structured synthesis improved pooled scores across all five shared criteria, especially prioritization, grounding and coherence, while model identity explained more variation than format. This is a retrospective benchmark using an LLM judge, not clinical validation or medical advice. ### Why it matters {#why-it-matters-mp-2026-10-17-010} AI Consilium evaluates 3,040 reports across grounding, coherence, management and multidisciplinary synthesis. ### Limits and context {#limitations-mp-2026-10-17-010} - A plausible differential diagnosis is not enough for a useful clinical second opinion. - This is a retrospective benchmark using an LLM judge, not clinical validation or medical advice. ### Claims and sources {#claims-mp-2026-10-17-010} - AI Consilium evaluates 3,040 reports across grounding, coherence, management and multidisciplinary synthesis. [source-2026-10-17-010] — Qualification: A plausible differential diagnosis is not enough for a useful clinical second opinion. ## 11. One Instruction Changes How the Agent Returns {#mp-2026-10-17-011} - Story ID: `mp-2026-10-17-011` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-011/one-instruction-changes-how-the-agent-returns **Dek:** An environment-agnostic coding agent improves when it reports difficulty and yields control for automatic continuation. Access to tools and instructions does not guarantee that an agent will use them effectively. Across ARC-AGI-3, AGI Maze and Terminal-Bench, one prompt change—reporting difficulty and returning control for continuation—improved mean scores on the first two environments while producing small, mixed changes on the third. Moving reset decisions into a simple scripted controller recovered part of a specialized agent's advantage, while extra tutorial experience could hurt. The experiments cannot separate better procedure use from additional or differently allocated inference, and gains cost more tokens. They nevertheless show that control scaffolding remains consequential for the tested models. ### Why it matters {#why-it-matters-mp-2026-10-17-011} An environment-agnostic coding agent improves when it reports difficulty and yields control for automatic continuation. ### Limits and context {#limitations-mp-2026-10-17-011} - Access to tools and instructions does not guarantee that an agent will use them effectively. - The experiments cannot separate better procedure use from additional or differently allocated inference, and gains cost more tokens. ### Claims and sources {#claims-mp-2026-10-17-011} - An environment-agnostic coding agent improves when it reports difficulty and yields control for automatic continuation. [source-2026-10-17-011] — Qualification: Access to tools and instructions does not guarantee that an agent will use them effectively. ## 12. The Navigation Plan Keeps the Law Partially Ordered {#mp-2026-10-17-012} - Story ID: `mp-2026-10-17-012` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-012/the-navigation-plan-keeps-the-law-partially-ordered **Dek:** Posetal games translate legal guidelines into auditable multi-agent preferences without collapsing every duty into one score. Real movement rules often prioritize some duties without ranking every possible conflict. This framework derives scenario-specific agent preferences from partially ordered legal guidelines, then lets agents add private objectives without creating new illegal equilibria. Belief play coordinates without revealing those private preferences. In seven maritime encounters with up to four vessels, every violation in the committed equilibria was certified as forced by the hierarchy or unsatisfiable, and tested defectors were identified from public information. The cases are limited and do not constitute legal certification, but the method makes the path from written guidance to a motion plan inspectable. ### Why it matters {#why-it-matters-mp-2026-10-17-012} Posetal games translate legal guidelines into auditable multi-agent preferences without collapsing every duty into one score. ### Limits and context {#limitations-mp-2026-10-17-012} - The cases are limited and do not constitute legal certification, but the method makes the path from written guidance to a motion plan inspectable. ### Claims and sources {#claims-mp-2026-10-17-012} - Posetal games translate legal guidelines into auditable multi-agent preferences without collapsing every duty into one score. [source-2026-10-17-012] — Qualification: The cases are limited and do not constitute legal certification, but the method makes the path from written guidance to a motion plan inspectable. ## 13. The Desktop Agent Stops Treating Code as a Bolt-On {#mp-2026-10-17-013} - Story ID: `mp-2026-10-17-013` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-013/the-desktop-agent-stops-treating-code-as-a-bolt-on **Dek:** OS-Helix trains GUI actions and code execution in one action space after naive tool addition made the visual agent worse. Desktop work crosses visible interfaces, files, scripts and hidden system state. OS-Helix trains GUI actions and code execution together, using application-artifact composition to build executable Ubuntu tasks with solution-independent final-state checks. Step-balanced reinforcement learning aggregates importance by action step so long code blocks do not dominate optimization. On OSWorld-Verified, the authors report 33.7 to 35 percent relative gains over matched hybrid baselines; the 27-billion-parameter RL version reached 66.44 percent success and transferred to WindowsAgentArena. These are benchmark results in constructed environments, not proof of safe unattended desktop operation. ### Why it matters {#why-it-matters-mp-2026-10-17-013} OS-Helix trains GUI actions and code execution in one action space after naive tool addition made the visual agent worse. ### Limits and context {#limitations-mp-2026-10-17-013} - Step-balanced reinforcement learning aggregates importance by action step so long code blocks do not dominate optimization. - These are benchmark results in constructed environments, not proof of safe unattended desktop operation. ### Claims and sources {#claims-mp-2026-10-17-013} - OS-Helix trains GUI actions and code execution in one action space after naive tool addition made the visual agent worse. [source-2026-10-17-013] — Qualification: Step-balanced reinforcement learning aggregates importance by action step so long code blocks do not dominate optimization. ## 14. The Humanoid Uses Its Torso as Part of the Grasp {#mp-2026-10-17-014} - Story ID: `mp-2026-10-17-014` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-014/the-humanoid-uses-its-torso-as-part-of-the-grasp **Dek:** Whole-Body Dexterity trains a policy to brace and slide bulky objects against arms and body surfaces. Dexterous manipulation is often reduced to the hands. Whole-Body Dexterity first generates feasible body-contact grasps, trains reinforcement-learning experts to move between them and distills those experts into one policy using proprioception and object geometry. Transferred without hardware-specific fine-tuning to a Unitree G1, the controller repositioned a basketball and box around the torso and completed a teleoperated sequence that stowed, carried and placed a box. Across the reported real-world tasks, it completed 90 percent of trials, versus 18 percent for direct retargeting and 4 percent for a generalist whole-body controller. The task set is narrow and still uses coarse human references. ### Why it matters {#why-it-matters-mp-2026-10-17-014} Whole-Body Dexterity trains a policy to brace and slide bulky objects against arms and body surfaces. ### Limits and context {#limitations-mp-2026-10-17-014} - No additional limitation was separately recorded. ### Claims and sources {#claims-mp-2026-10-17-014} - Whole-Body Dexterity trains a policy to brace and slide bulky objects against arms and body surfaces. [source-2026-10-17-014] ## 15. The Diffusion Model Keeps the Details That Are Still Moving {#mp-2026-10-17-026} - Story ID: `mp-2026-10-17-026` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-026/the-diffusion-model-keeps-the-details-that-are-still-moving **Dek:** DaTMe decides which image tokens to merge from denoising dynamics and local structure, not prompt relevance alone. Training-free token merging can speed image diffusion by collapsing redundant computation, but preserving only prompt-aligned foreground regions can erase background structure and unmentioned detail. Dynamics-aware Token Merging builds an importance map from the predicted clean image: temporal change marks unresolved regions, while local high-frequency content identifies boundaries and fine structure. Important tokens remain anchors; stable, homogeneous regions become merge candidates. Across PixArt-a and FLUX, the authors report better fidelity across the full image than comparison methods at matched compression settings. The evidence is benchmark- and model-specific, but it replaces a semantic shortcut with signals from the generation process itself. ### Why it matters {#why-it-matters-mp-2026-10-17-026} DaTMe decides which image tokens to merge from denoising dynamics and local structure, not prompt relevance alone. ### Limits and context {#limitations-mp-2026-10-17-026} - Training-free token merging can speed image diffusion by collapsing redundant computation, but preserving only prompt-aligned foreground regions can erase background structure and unmentioned detail. - Important tokens remain anchors; stable, homogeneous regions become merge candidates. ### Claims and sources {#claims-mp-2026-10-17-026} - DaTMe decides which image tokens to merge from denoising dynamics and local structure, not prompt relevance alone. [source-2026-10-17-015] — Qualification: Training-free token merging can speed image diffusion by collapsing redundant computation, but preserving only prompt-aligned foreground regions can erase background structure and unmentioned detail. ## 16. A Correct Answer Can Hide a Broken Memory {#mp-2026-10-17-027} - Story ID: `mp-2026-10-17-027` - Type: `dispatch` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-027/a-correct-answer-can-hide-a-broken-memory **Dek:** FEB-30 inspects the store after each fact change instead of using downstream question answering as the only proxy. Question-answering accuracy can blame a good memory for a retrieval failure or reward a broken memory when generation guesses correctly. FEB-30 inverts the usual benchmark: known fact-evolution trajectories generate the documents, emails and conversations, producing deterministic ground truth for direct store inspection. On one representative system, intrinsic and downstream evaluation differed by 16 points on average, with 44 percent of trajectories crossing the authors' meaningful-divergence threshold. The system classified create, update, reinforce and supersede operations correctly only 20 percent of the time. Thirty trajectories are not a comprehensive memory benchmark, but the result exposes failures that answer-only scoring cannot see. ### Why it matters {#why-it-matters-mp-2026-10-17-027} FEB-30 inspects the store after each fact change instead of using downstream question answering as the only proxy. ### Limits and context {#limitations-mp-2026-10-17-027} - The system classified create, update, reinforce and supersede operations correctly only 20 percent of the time. - Thirty trajectories are not a comprehensive memory benchmark, but the result exposes failures that answer-only scoring cannot see. ### Claims and sources {#claims-mp-2026-10-17-027} - FEB-30 inspects the store after each fact change instead of using downstream question answering as the only proxy. [source-2026-10-17-016] — Qualification: The system classified create, update, reinforce and supersede operations correctly only 20 percent of the time. ## 17. Quantum-Code Correlations Survive Two Noise Regimes {#mp-2026-10-17-015} - Story ID: `mp-2026-10-17-015` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-015/quantum-code-correlations-survive-two-noise-regimes **Dek:** A proof finds exponential decay of conditional mutual information under weak noise and under strong noise when the buffer carries no logical operator. For normalized projector states of Z-d qLDPC stabilizer codes, the authors derive system-size-independent bounds and, on finite-dimensional lattices, a quasi-local recovery channel in the weak-noise regime. This is a mathematical stability result, not a hardware demonstration. ### Why it matters {#why-it-matters-mp-2026-10-17-015} A proof finds exponential decay of conditional mutual information under weak noise and under strong noise when the buffer carries no logical operator. ### Limits and context {#limitations-mp-2026-10-17-015} - This is a mathematical stability result, not a hardware demonstration. ### Claims and sources {#claims-mp-2026-10-17-015} - A proof finds exponential decay of conditional mutual information under weak noise and under strong noise when the buffer carries no logical operator. [source-2026-10-17-017] — Qualification: This is a mathematical stability result, not a hardware demonstration. ## 18. Three Hot Jupiters Rise to the Top of the Deformation List {#mp-2026-10-17-016} - Story ID: `mp-2026-10-17-016` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-016/three-hot-jupiters-rise-to-the-top-of-the-deformation-list **Dek:** A new metric combines expected tidal shape change with photometric signal-to-noise for JWST and Ariel phase curves. The authors identify WASP-33 b, WASP-76 b and HAT-P-32 b as strong unmeasured targets. Simulations suggest one continuous JWST observation of WASP-76 b could constrain its fluid Love number to about 0.09 precision; that is a forecast, not a completed observation. ### Why it matters {#why-it-matters-mp-2026-10-17-016} A new metric combines expected tidal shape change with photometric signal-to-noise for JWST and Ariel phase curves. ### Limits and context {#limitations-mp-2026-10-17-016} - Simulations suggest one continuous JWST observation of WASP-76 b could constrain its fluid Love number to about 0.09 precision; that is a forecast, not a completed observation. ### Claims and sources {#claims-mp-2026-10-17-016} - A new metric combines expected tidal shape change with photometric signal-to-noise for JWST and Ariel phase curves. [source-2026-10-17-018] — Qualification: Simulations suggest one continuous JWST observation of WASP-76 b could constrain its fluid Love number to about 0.09 precision; that is a forecast, not a completed observation. ## 19. Private Network Counts Reach the Noninteractive Limit {#mp-2026-10-17-017} - Story ID: `mp-2026-10-17-017` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-017/private-network-counts-reach-the-noninteractive-limit **Dek:** A local node-differential-privacy algorithm counts any constant-size template subgraph with near-optimal additive error. The matching lower bound establishes the first polynomial-in-n separation between central and local node privacy for accuracy on all input graphs. The result is theoretical and expands beyond prior degree-distribution statistics; it is not a deployment benchmark. ### Why it matters {#why-it-matters-mp-2026-10-17-017} A local node-differential-privacy algorithm counts any constant-size template subgraph with near-optimal additive error. ### Limits and context {#limitations-mp-2026-10-17-017} - The result is theoretical and expands beyond prior degree-distribution statistics; it is not a deployment benchmark. ### Claims and sources {#claims-mp-2026-10-17-017} - A local node-differential-privacy algorithm counts any constant-size template subgraph with near-optimal additive error. [source-2026-10-17-019] — Qualification: The result is theoretical and expands beyond prior degree-distribution statistics; it is not a deployment benchmark. ## 20. Meta Reworks the Model Around the Accelerator {#mp-2026-10-17-018} - Story ID: `mp-2026-10-17-018` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-018/meta-reworks-the-model-around-the-accelerator **Dek:** A production case study reports up to 1.5-times higher training throughput than a direct AMD MI350X port. The paper combines hardware-model co-design, zero-warmup kernel selection and agent-assisted porting for recommendation workloads, emphasizing MI350X memory capacity rather than CUDA-shaped assumptions. The comparison is Meta's own case study, not an independent fleet-wide benchmark. ### Why it matters {#why-it-matters-mp-2026-10-17-018} A production case study reports up to 1.5-times higher training throughput than a direct AMD MI350X port. ### Limits and context {#limitations-mp-2026-10-17-018} - The comparison is Meta's own case study, not an independent fleet-wide benchmark. ### Claims and sources {#claims-mp-2026-10-17-018} - A production case study reports up to 1.5-times higher training throughput than a direct AMD MI350X port. [source-2026-10-17-020] — Qualification: The comparison is Meta's own case study, not an independent fleet-wide benchmark. ## 21. Change the Rating Scale, Change One in Five Labels {#mp-2026-10-17-019} - Story ID: `mp-2026-10-17-019` - Type: `ticker` - Classification: `editorial` - Content status: `new` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-019/change-the-rating-scale-change-one-in-five-labels **Dek:** Across five models and three social-science tasks, one scale-design change moved normalized scores by about 0.10 on average. Roughly 20 percent of texts changed category across alternative formats, ranges or signs, with the largest sensitivity on toxicity. Rubric prompting did not remove the effect. The finding argues for robustness checks; it does not invalidate every model-based score. ### Why it matters {#why-it-matters-mp-2026-10-17-019} Across five models and three social-science tasks, one scale-design change moved normalized scores by about 0.10 on average. ### Limits and context {#limitations-mp-2026-10-17-019} - Rubric prompting did not remove the effect. - The finding argues for robustness checks; it does not invalidate every model-based score. ### Claims and sources {#claims-mp-2026-10-17-019} - Across five models and three social-science tasks, one scale-design change moved normalized scores by about 0.10 on average. [source-2026-10-17-021] — Qualification: Rubric prompting did not remove the effect. ## 22. Space Drums 2.0 {#mp-2026-10-17-020} - Story ID: `mp-2026-10-17-020` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-020/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. 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 {#why-it-matters-mp-2026-10-17-020} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-17-020} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-17-020} - This Invention Desk entry makes no independently sourced news claim. ## 23. 3RIC {#mp-2026-10-17-021} - Story ID: `mp-2026-10-17-021` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-021/3ric **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. 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 {#why-it-matters-mp-2026-10-17-021} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-17-021} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-17-021} - This Invention Desk entry makes no independently sourced news claim. ## 24. M3-CRETE {#mp-2026-10-17-022} - Story ID: `mp-2026-10-17-022` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-022/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. 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 {#why-it-matters-mp-2026-10-17-022} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-17-022} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-17-022} - This Invention Desk entry makes no independently sourced news claim. ## 25. DeskNode {#mp-2026-10-17-023} - Story ID: `mp-2026-10-17-023` - Type: `invention_desk` - Classification: `editorial` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-023/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. 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 {#why-it-matters-mp-2026-10-17-023} An independent builder is turning an improbable idea into a working project. ### Limits and context {#limitations-mp-2026-10-17-023} - A Desk Pick is an editorial selection, not a product endorsement. ### Claims and sources {#claims-mp-2026-10-17-023} - This Invention Desk entry makes no independently sourced news claim. ## 26. The First Paid Slot {#mp-2026-10-17-024} - Story ID: `mp-2026-10-17-024` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-024/the-first-paid-slot **Dek:** A transparent preview of paid placement with one verified link and no claim of endorsement. A transparent preview of paid placement with one verified link and no claim of endorsement. House example - no advertiser paid. Payment will buy placement, never endorsement. ### Why it matters {#why-it-matters-mp-2026-10-17-024} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-10-17-024} - House example - no advertiser paid. Payment will buy placement, never endorsement. ### Claims and sources {#claims-mp-2026-10-17-024} - This Invention Desk entry makes no independently sourced news claim. ## 27. Put Your Project on the Desk {#mp-2026-10-17-025} - Story ID: `mp-2026-10-17-025` - Type: `invention_desk` - Classification: `house_example` - Content status: `carried_over` - Permanent URL: https://themachinepress.com/story/mp-2026-10-17-025/put-your-project-on-the-desk **Dek:** One manually reviewed placement stays active for seven days and remains separate from Desk Picks. One manually reviewed placement stays active for seven days and remains separate from Desk Picks. Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed. ### Why it matters {#why-it-matters-mp-2026-10-17-025} This placement explains how builders can appear in The Invention Desk without purchasing editorial endorsement. ### Limits and context {#limitations-mp-2026-10-17-025} - Manual intake only. Payment buys placement, never endorsement, and every submission is reviewed. ### Claims and sources {#claims-mp-2026-10-17-025} - This Invention Desk entry makes no independently sourced news claim. ## Normalized sources - **source-2026-10-17-001:** [arXiv preprint 2610.22360, original v1](https://arxiv.org/abs/2610.22360v1) — arXiv; primary_research - **source-2026-10-17-002:** [arXiv preprint 2610.22243, original v1](https://arxiv.org/abs/2610.22243v1) — arXiv; primary_research - **source-2026-10-17-003:** [arXiv preprint 2610.22368, original v1](https://arxiv.org/abs/2610.22368v1) — arXiv; primary_research - **source-2026-10-17-004:** [arXiv preprint 2610.22364, original v1](https://arxiv.org/abs/2610.22364v1) — arXiv; primary_research - **source-2026-10-17-005:** [arXiv preprint 2610.22362, original v1](https://arxiv.org/abs/2610.22362v1) — arXiv; primary_research - **source-2026-10-17-006:** [arXiv preprint 2610.22358, original v1](https://arxiv.org/abs/2610.22358v1) — arXiv; primary_research - **source-2026-10-17-007:** [arXiv preprint 2610.22339, original v1](https://arxiv.org/abs/2610.22339v1) — arXiv; primary_research - **source-2026-10-17-008:** [arXiv preprint 2610.22324, original v1](https://arxiv.org/abs/2610.22324v1) — arXiv; primary_research - **source-2026-10-17-009:** [arXiv preprint 2610.22322, original v1](https://arxiv.org/abs/2610.22322v1) — arXiv; primary_research - **source-2026-10-17-010:** [arXiv preprint 2610.22298, original v1](https://arxiv.org/abs/2610.22298v1) — arXiv; primary_research - **source-2026-10-17-011:** [arXiv preprint 2610.22289, original v1](https://arxiv.org/abs/2610.22289v1) — arXiv; primary_research - **source-2026-10-17-012:** [arXiv preprint 2610.22277, original v1](https://arxiv.org/abs/2610.22277v1) — arXiv; primary_research - **source-2026-10-17-013:** [arXiv preprint 2610.22234, original v1](https://arxiv.org/abs/2610.22234v1) — arXiv; primary_research - **source-2026-10-17-014:** [arXiv preprint 2610.22184, original v1](https://arxiv.org/abs/2610.22184v1) — arXiv; primary_research - **source-2026-10-17-015:** [arXiv preprint 2610.22370, original v1](https://arxiv.org/abs/2610.22370v1) — arXiv; primary_research - **source-2026-10-17-016:** [arXiv preprint 2610.22209, original v1](https://arxiv.org/abs/2610.22209v1) — arXiv; primary_research - **source-2026-10-17-017:** [arXiv preprint 2610.22357, original v1](https://arxiv.org/abs/2610.22357v1) — arXiv; primary_research - **source-2026-10-17-018:** [arXiv preprint 2610.22338, original v1](https://arxiv.org/abs/2610.22338v1) — arXiv; primary_research - **source-2026-10-17-019:** [arXiv preprint 2610.22278, original v1](https://arxiv.org/abs/2610.22278v1) — arXiv; primary_research - **source-2026-10-17-020:** [arXiv preprint 2610.22230, original v1](https://arxiv.org/abs/2610.22230v1) — arXiv; primary_research - **source-2026-10-17-021:** [arXiv preprint 2610.22187, original v1](https://arxiv.org/abs/2610.22187v1) — arXiv; primary_research